doi
stringlengths 9
10
| chunk-id
stringlengths 1
4
| chunk
stringlengths 1
1.57k
| id
stringlengths 9
10
| title
stringlengths 10
127
| summary
stringlengths 581
1.92k
| source
stringlengths 30
31
| authors
sequence | categories
sequence | comment
stringclasses 28
values | journal_ref
stringclasses 1
value | primary_category
stringclasses 6
values | published
stringlengths 8
8
| updated
stringlengths 8
8
| references
list |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
1806.01261 | 868 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 869 | sha1_base64="Zv5ybfKo/H8QvXQaOWqQVyOFtg4=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ73JkJnZZWZWCCGf4MWDIl79Im/+jZNkDxotaCiquunuilLBjfX9L6+wtr6xuVXcLu3s7u0flA+PWibJNMMmS0SiOxE1KLjCpuVWYCfVSGUksB2Nb+Z++xG14Yl6sJMUQ0mHisecUeuke3kd9MsVv+YvQP6SICcVyNHolz97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 870 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 871 | sha1_base64="hP+6LrUf2d3tZaldqaQQvEKMXyw=">AAAB2XicbZDNSgMxFIXv1L86Vq1rN8EiuCozbnQpuHFZwbZCO5RM5k4bmskMyR2hDH0BF25EfC93vo3pz0JbDwQ+zknIvSculLQUBN9ebWd3b/+gfugfNfzjk9Nmo2fz0gjsilzl5jnmFpXU2CVJCp8LgzyLFfbj6f0i77+gsTLXTzQrMMr4WMtUCk7O6oyaraAdLMW2IVxDC9YaNb+GSS7KDDUJxa0dhEFBUcUNSaFw7g9LiwUXUz7GgUPNM7RRt | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 872 | DUJxa0dhEFBUcUNSaFw7g9LiwUXUz7GgUPNM7RRtRxzzi6dk7A0N+5oYkv394uKZ9bOstjdzDhN7Ga2MP/LBiWlt1EldVESarH6KC0Vo5wtdmaJNChIzRxwYaSblYkJN1yQa8Z3HYSbG29D77odBu3wMYA6nMMFXEEIN3AHD9CBLghI4BXevYn35n2suqp569LO4I+8zx84xIo4</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 873 | sha1_base64="41wuc+jmacSzA8ipzMzk3JQ30Cs=">AAAB33icbZDNSgMxFIXv1L9aq1a3boJFcFUybnQjCG5cVrS10A4lk95pQ5PMkGSEMvQR3LhQxLdy59uY/iy09UDg45yE3HviTArrKP0OShubW9s75d3KXnX/4LB2VG3bNDccWzyVqenEzKIUGltOOImdzCBTscSneHw7y5+e0ViR6kc3yTBSbKhFIjhz3npQ12G/VqcNOhdZh3AJdViq2a999QYpzxVqxyWzthvSzEUFM05widNKL7eYMT5mQ+x61 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 874 | QYpzxVqxyWzthvSzEUFM05widNKL7eYMT5mQ+x61EyhjYr5qFNy5p0BSVLjj3Zk7v5+UTBl7UTF/qZibmRXs5n5X9bNXXIVFUJnuUPNFx8luSQuJbO9yUAY5E5OPDBuhJ+V8BEzjDvfTsWXEK6uvA7ti0ZIG+E9hTKcwCmcQwiXcAN30IQWcBjCC7zBeyCD1+BjUVcpWPZ2DH8UfP4Arc6MHg==</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 875 | sha1_base64="41wuc+jmacSzA8ipzMzk3JQ30Cs=">AAAB33icbZDNSgMxFIXv1L9aq1a3boJFcFUybnQjCG5cVrS10A4lk95pQ5PMkGSEMvQR3LhQxLdy59uY/iy09UDg45yE3HviTArrKP0OShubW9s75d3KXnX/4LB2VG3bNDccWzyVqenEzKIUGltOOImdzCBTscSneHw7y5+e0ViR6kc3yTBSbKhFIjhz3npQ12G/VqcNOhdZh3AJdViq2a999QYpzxVqxyWzthvSzEUFM05widNKL7eYMT5mQ+x61 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 876 | QYpzxVqxyWzthvSzEUFM05widNKL7eYMT5mQ+x61EyhjYr5qFNy5p0BSVLjj3Zk7v5+UTBl7UTF/qZibmRXs5n5X9bNXXIVFUJnuUPNFx8luSQuJbO9yUAY5E5OPDBuhJ+V8BEzjDvfTsWXEK6uvA7ti0ZIG+E9hTKcwCmcQwiXcAN30IQWcBjCC7zBeyCD1+BjUVcpWPZ2DH8UfP4Arc6MHg==</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 877 | sha1_base64="c5O5JuZLnHr2YYnYcAr/a2hwAn0=">AAAB6nicbVA9SwNBEJ2LXzF+RS1tFoOQKtzZaCMEbCwjmg9IjrC3mSRLdveO3T0hHPkJNhaK2PqL7Pw3bpIrNPHBwOO9GWbmRYngxvr+t1fY2Nza3inulvb2Dw6PyscnLROnmmGTxSLWnYgaFFxh03IrsJNopDIS2I4mt3O//YTa8Fg92mmCoaQjxYecUeukB3kT9MsVv+YvQNZJkJMK5Gj0y1+9QcxSicoyQY3pBn5iw4xqy5nAWamXGkwom9ARd | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 878 | 1+9QcxSicoyQY3pBn5iw4xqy5nAWamXGkwom9ARdh1VVKIJs8WpM3LhlAEZxtqVsmSh/p7IqDRmKiPXKakdm1VvLv7ndVM7vA4zrpLUomLLRcNUEBuT+d9kwDUyK6aOUKa5u5WwMdWUWZdOyYUQrL68TlqXtcCvBfd+pV7N4yjCGZxDFQK4gjrcQQOawGAEz/AKb57wXrx372PZWvDymVP4A+/zB74jjVU=</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 879 | sha1_base64="Zv5ybfKo/H8QvXQaOWqQVyOFtg4=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ73JkJnZZWZWCCGf4MWDIl79Im/+jZNkDxotaCiquunuilLBjfX9L6+wtr6xuVXcLu3s7u0flA+PWibJNMMmS0SiOxE1KLjCpuVWYCfVSGUksB2Nb+Z++xG14Yl6sJMUQ0mHisecUeuke3kd9MsVv+YvQP6SICcVyNHolz97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 880 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 881 | sha1_base64="Zv5ybfKo/H8QvXQaOWqQVyOFtg4=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ73JkJnZZWZWCCGf4MWDIl79Im/+jZNkDxotaCiquunuilLBjfX9L6+wtr6xuVXcLu3s7u0flA+PWibJNMMmS0SiOxE1KLjCpuVWYCfVSGUksB2Nb+Z++xG14Yl6sJMUQ0mHisecUeuke3kd9MsVv+YvQP6SICcVyNHolz97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 882 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 883 | sha1_base64="Zv5ybfKo/H8QvXQaOWqQVyOFtg4=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ73JkJnZZWZWCCGf4MWDIl79Im/+jZNkDxotaCiquunuilLBjfX9L6+wtr6xuVXcLu3s7u0flA+PWibJNMMmS0SiOxE1KLjCpuVWYCfVSGUksB2Nb+Z++xG14Yl6sJMUQ0mHisecUeuke3kd9MsVv+YvQP6SICcVyNHolz97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 884 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 885 | sha1_base64="Zv5ybfKo/H8QvXQaOWqQVyOFtg4=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ73JkJnZZWZWCCGf4MWDIl79Im/+jZNkDxotaCiquunuilLBjfX9L6+wtr6xuVXcLu3s7u0flA+PWibJNMMmS0SiOxE1KLjCpuVWYCfVSGUksB2Nb+Z++xG14Yl6sJMUQ0mHisecUeuke3kd9MsVv+YvQP6SICcVyNHolz97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 886 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 887 | sha1_base64="Zv5ybfKo/H8QvXQaOWqQVyOFtg4=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ73JkJnZZWZWCCGf4MWDIl79Im/+jZNkDxotaCiquunuilLBjfX9L6+wtr6xuVXcLu3s7u0flA+PWibJNMMmS0SiOxE1KLjCpuVWYCfVSGUksB2Nb+Z++xG14Yl6sJMUQ0mHisecUeuke3kd9MsVv+YvQP6SICcVyNHolz97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 888 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 889 | sha1_base64="Zv5ybfKo/H8QvXQaOWqQVyOFtg4=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ73JkJnZZWZWCCGf4MWDIl79Im/+jZNkDxotaCiquunuilLBjfX9L6+wtr6xuVXcLu3s7u0flA+PWibJNMMmS0SiOxE1KLjCpuVWYCfVSGUksB2Nb+Z++xG14Yl6sJMUQ0mHisecUeuke3kd9MsVv+YvQP6SICcVyNHolz97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 890 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit>m=2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 891 | <latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 892 | sha1_base64="oVNcNejQAmVb9Nn6VIeeV7f50ls=">AAAB6nicbVBNSwMxEJ3Ur1q/qh69BIvQU9ktgl6EghePFe0HtEvJptk2NMkuSVYoS3+CFw+KePUXefPfmLZ70NYHA4/3ZpiZFyaCG+t536iwsbm1vVPcLe3tHxwelY9P2iZONWUtGotYd0NimOCKtSy3gnUTzYgMBeuEk9u533li2vBYPdppwgJJRopHnBLrpAd5Ux+UK17NWwCvEz8nFcjRHJS/+sOYppIpSwUxpud7iQ0yoi2ngs1K/dSwhNAJG | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 893 | JS/+sOYppIpSwUxpud7iQ0yoi2ngs1K/dSwhNAJGbGeo4pIZoJsceoMXzhliKNYu1IWL9TfExmRxkxl6DolsWOz6s3F/7xeaqPrIOMqSS1TdLkoSgW2MZ7/jYdcM2rF1BFCNXe3YjommlDr0im5EPzVl9dJu17zvZp/f1lpVPM4inAG51AFH66gAXfQhBZQGMEzvMIbEugFvaOPZWsB5TOn8Afo8wfA541a</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 894 | sha1_base64="oVNcNejQAmVb9Nn6VIeeV7f50ls=">AAAB6nicbVBNSwMxEJ3Ur1q/qh69BIvQU9ktgl6EghePFe0HtEvJptk2NMkuSVYoS3+CFw+KePUXefPfmLZ70NYHA4/3ZpiZFyaCG+t536iwsbm1vVPcLe3tHxwelY9P2iZONWUtGotYd0NimOCKtSy3gnUTzYgMBeuEk9u533li2vBYPdppwgJJRopHnBLrpAd5Ux+UK17NWwCvEz8nFcjRHJS/+sOYppIpSwUxpud7iQ0yoi2ngs1K/dSwhNAJG | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 895 | JS/+sOYppIpSwUxpud7iQ0yoi2ngs1K/dSwhNAJGbGeo4pIZoJsceoMXzhliKNYu1IWL9TfExmRxkxl6DolsWOz6s3F/7xeaqPrIOMqSS1TdLkoSgW2MZ7/jYdcM2rF1BFCNXe3YjommlDr0im5EPzVl9dJu17zvZp/f1lpVPM4inAG51AFH66gAXfQhBZQGMEzvMIbEugFvaOPZWsB5TOn8Afo8wfA541a</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 896 | sha1_base64="oVNcNejQAmVb9Nn6VIeeV7f50ls=">AAAB6nicbVBNSwMxEJ3Ur1q/qh69BIvQU9ktgl6EghePFe0HtEvJptk2NMkuSVYoS3+CFw+KePUXefPfmLZ70NYHA4/3ZpiZFyaCG+t536iwsbm1vVPcLe3tHxwelY9P2iZONWUtGotYd0NimOCKtSy3gnUTzYgMBeuEk9u533li2vBYPdppwgJJRopHnBLrpAd5Ux+UK17NWwCvEz8nFcjRHJS/+sOYppIpSwUxpud7iQ0yoi2ngs1K/dSwhNAJG | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 897 | JS/+sOYppIpSwUxpud7iQ0yoi2ngs1K/dSwhNAJGbGeo4pIZoJsceoMXzhliKNYu1IWL9TfExmRxkxl6DolsWOz6s3F/7xeaqPrIOMqSS1TdLkoSgW2MZ7/jYdcM2rF1BFCNXe3YjommlDr0im5EPzVl9dJu17zvZp/f1lpVPM4inAG51AFH66gAXfQhBZQGMEzvMIbEugFvaOPZWsB5TOn8Afo8wfA541a</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 898 | sha1_base64="oVNcNejQAmVb9Nn6VIeeV7f50ls=">AAAB6nicbVBNSwMxEJ3Ur1q/qh69BIvQU9ktgl6EghePFe0HtEvJptk2NMkuSVYoS3+CFw+KePUXefPfmLZ70NYHA4/3ZpiZFyaCG+t536iwsbm1vVPcLe3tHxwelY9P2iZONWUtGotYd0NimOCKtSy3gnUTzYgMBeuEk9u533li2vBYPdppwgJJRopHnBLrpAd5Ux+UK17NWwCvEz8nFcjRHJS/+sOYppIpSwUxpud7iQ0yoi2ngs1K/dSwhNAJG | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 899 | JS/+sOYppIpSwUxpud7iQ0yoi2ngs1K/dSwhNAJGbGeo4pIZoJsceoMXzhliKNYu1IWL9TfExmRxkxl6DolsWOz6s3F/7xeaqPrIOMqSS1TdLkoSgW2MZ7/jYdcM2rF1BFCNXe3YjommlDr0im5EPzVl9dJu17zvZp/f1lpVPM4inAG51AFH66gAXfQhBZQGMEzvMIbEugFvaOPZWsB5TOn8Afo8wfA541a</latexit>m=3 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 900 | <latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 901 | sha1_base64="+jOsYdJ9SNJymWDVPaXJffNRSPM=">AAAB6nicbVBNSwMxEJ3Ur1q/qh69BIvQU9lVQS9CwYvHivYD2qVk02wbmmSXJCuUpT/BiwdFvPqLvPlvTNs9aOuDgcd7M8zMCxPBjfW8b1RYW9/Y3Cpul3Z29/YPyodHLROnmrImjUWsOyExTHDFmpZbwTqJZkSGgrXD8e3Mbz8xbXisHu0kYYEkQ8UjTol10oO8ueiXK17NmwOvEj8nFcjR6Je/eoOYppIpSwUxput7iQ0yoi2ngk1LvdSwhNAxG | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 902 | Je/eoOYppIpSwUxput7iQ0yoi2ngk1LvdSwhNAxGbKuo4pIZoJsfuoUnzllgKNYu1IWz9XfExmRxkxk6DolsSOz7M3E/7xuaqPrIOMqSS1TdLEoSgW2MZ79jQdcM2rFxBFCNXe3YjoimlDr0im5EPzll1dJ67zmezX//rJSr+ZxFOEETqEKPlxBHe6gAU2gMIRneIU3JNALekcfi9YCymeO4Q/Q5w/Ca41b</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 903 | sha1_base64="+jOsYdJ9SNJymWDVPaXJffNRSPM=">AAAB6nicbVBNSwMxEJ3Ur1q/qh69BIvQU9lVQS9CwYvHivYD2qVk02wbmmSXJCuUpT/BiwdFvPqLvPlvTNs9aOuDgcd7M8zMCxPBjfW8b1RYW9/Y3Cpul3Z29/YPyodHLROnmrImjUWsOyExTHDFmpZbwTqJZkSGgrXD8e3Mbz8xbXisHu0kYYEkQ8UjTol10oO8ueiXK17NmwOvEj8nFcjR6Je/eoOYppIpSwUxput7iQ0yoi2ngk1LvdSwhNAxG | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 904 | Je/eoOYppIpSwUxput7iQ0yoi2ngk1LvdSwhNAxGbKuo4pIZoJsfuoUnzllgKNYu1IWz9XfExmRxkxk6DolsSOz7M3E/7xuaqPrIOMqSS1TdLEoSgW2MZ79jQdcM2rFxBFCNXe3YjoimlDr0im5EPzll1dJ67zmezX//rJSr+ZxFOEETqEKPlxBHe6gAU2gMIRneIU3JNALekcfi9YCymeO4Q/Q5w/Ca41b</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 905 | sha1_base64="+jOsYdJ9SNJymWDVPaXJffNRSPM=">AAAB6nicbVBNSwMxEJ3Ur1q/qh69BIvQU9lVQS9CwYvHivYD2qVk02wbmmSXJCuUpT/BiwdFvPqLvPlvTNs9aOuDgcd7M8zMCxPBjfW8b1RYW9/Y3Cpul3Z29/YPyodHLROnmrImjUWsOyExTHDFmpZbwTqJZkSGgrXD8e3Mbz8xbXisHu0kYYEkQ8UjTol10oO8ueiXK17NmwOvEj8nFcjR6Je/eoOYppIpSwUxput7iQ0yoi2ngk1LvdSwhNAxG | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 906 | Je/eoOYppIpSwUxput7iQ0yoi2ngk1LvdSwhNAxGbKuo4pIZoJsfuoUnzllgKNYu1IWz9XfExmRxkxk6DolsSOz7M3E/7xuaqPrIOMqSS1TdLEoSgW2MZ79jQdcM2rFxBFCNXe3YjoimlDr0im5EPzll1dJ67zmezX//rJSr+ZxFOEETqEKPlxBHe6gAU2gMIRneIU3JNALekcfi9YCymeO4Q/Q5w/Ca41b</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 907 | sha1_base64="+jOsYdJ9SNJymWDVPaXJffNRSPM=">AAAB6nicbVBNSwMxEJ3Ur1q/qh69BIvQU9lVQS9CwYvHivYD2qVk02wbmmSXJCuUpT/BiwdFvPqLvPlvTNs9aOuDgcd7M8zMCxPBjfW8b1RYW9/Y3Cpul3Z29/YPyodHLROnmrImjUWsOyExTHDFmpZbwTqJZkSGgrXD8e3Mbz8xbXisHu0kYYEkQ8UjTol10oO8ueiXK17NmwOvEj8nFcjR6Je/eoOYppIpSwUxput7iQ0yoi2ngk1LvdSwhNAxG | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 908 | Je/eoOYppIpSwUxput7iQ0yoi2ngk1LvdSwhNAxGbKuo4pIZoJsfuoUnzllgKNYu1IWz9XfExmRxkxk6DolsSOz7M3E/7xuaqPrIOMqSS1TdLEoSgW2MZ79jQdcM2rFxBFCNXe3YjoimlDr0im5EPzll1dJ67zmezX//rJSr+ZxFOEETqEKPlxBHe6gAU2gMIRneIU3JNALekcfi9YCymeO4Q/Q5w/Ca41b</latexit> | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 909 | m=0 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 910 | <latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 911 | sha1_base64="ic6WezPV7TWar890N1QnpVOPjNA=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ7PJkJnZZaZXCCGf4MWDIl79Im/+jZNkDxotaCiquunuilIpLPr+l1dYW9/Y3Cpul3Z29/YPyodHLZtkhvEmS2RiOhG1XArNmyhQ8k5qOFWR5O1ofDP324/cWJHoB5ykPFR0qEUsGEUn3atrv1+u+DV/AfKXBDmpQI5Gv/zZGyQsU1wjk9TabuCnGE6pQcEkn5V6meUpZWM65 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 912 | /zZGyQsU1wjk9TabuCnGE6pQcEkn5V6meUpZWM65F1HNVXchtPFqTNy5pQBiRPjSiNZqD8nplRZO1GR61QUR3bVm4v/ed0M46twKnSaIddsuSjOJMGEzP8mA2E4QzlxhDIj3K2EjaihDF06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NWT3rP35r0vWwtePnMMv+B9fAO9341Y</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 913 | sha1_base64="ic6WezPV7TWar890N1QnpVOPjNA=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ7PJkJnZZaZXCCGf4MWDIl79Im/+jZNkDxotaCiquunuilIpLPr+l1dYW9/Y3Cpul3Z29/YPyodHLZtkhvEmS2RiOhG1XArNmyhQ8k5qOFWR5O1ofDP324/cWJHoB5ykPFR0qEUsGEUn3atrv1+u+DV/AfKXBDmpQI5Gv/zZGyQsU1wjk9TabuCnGE6pQcEkn5V6meUpZWM65 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 914 | /zZGyQsU1wjk9TabuCnGE6pQcEkn5V6meUpZWM65F1HNVXchtPFqTNy5pQBiRPjSiNZqD8nplRZO1GR61QUR3bVm4v/ed0M46twKnSaIddsuSjOJMGEzP8mA2E4QzlxhDIj3K2EjaihDF06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NWT3rP35r0vWwtePnMMv+B9fAO9341Y</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 915 | sha1_base64="ic6WezPV7TWar890N1QnpVOPjNA=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ7PJkJnZZaZXCCGf4MWDIl79Im/+jZNkDxotaCiquunuilIpLPr+l1dYW9/Y3Cpul3Z29/YPyodHLZtkhvEmS2RiOhG1XArNmyhQ8k5qOFWR5O1ofDP324/cWJHoB5ykPFR0qEUsGEUn3atrv1+u+DV/AfKXBDmpQI5Gv/zZGyQsU1wjk9TabuCnGE6pQcEkn5V6meUpZWM65 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 916 | /zZGyQsU1wjk9TabuCnGE6pQcEkn5V6meUpZWM65F1HNVXchtPFqTNy5pQBiRPjSiNZqD8nplRZO1GR61QUR3bVm4v/ed0M46twKnSaIddsuSjOJMGEzP8mA2E4QzlxhDIj3K2EjaihDF06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NWT3rP35r0vWwtePnMMv+B9fAO9341Y</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 917 | sha1_base64="ic6WezPV7TWar890N1QnpVOPjNA=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ7PJkJnZZaZXCCGf4MWDIl79Im/+jZNkDxotaCiquunuilIpLPr+l1dYW9/Y3Cpul3Z29/YPyodHLZtkhvEmS2RiOhG1XArNmyhQ8k5qOFWR5O1ofDP324/cWJHoB5ykPFR0qEUsGEUn3atrv1+u+DV/AfKXBDmpQI5Gv/zZGyQsU1wjk9TabuCnGE6pQcEkn5V6meUpZWM65 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 918 | /zZGyQsU1wjk9TabuCnGE6pQcEkn5V6meUpZWM65F1HNVXchtPFqTNy5pQBiRPjSiNZqD8nplRZO1GR61QUR3bVm4v/ed0M46twKnSaIddsuSjOJMGEzP8mA2E4QzlxhDIj3K2EjaihDF06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NWT3rP35r0vWwtePnMMv+B9fAO9341Y</latexit>m=1 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 919 | <latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 920 | sha1_base64="Zv5ybfKo/H8QvXQaOWqQVyOFtg4=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ73JkJnZZWZWCCGf4MWDIl79Im/+jZNkDxotaCiquunuilLBjfX9L6+wtr6xuVXcLu3s7u0flA+PWibJNMMmS0SiOxE1KLjCpuVWYCfVSGUksB2Nb+Z++xG14Yl6sJMUQ0mHisecUeuke3kd9MsVv+YvQP6SICcVyNHolz97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 921 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 922 | sha1_base64="Zv5ybfKo/H8QvXQaOWqQVyOFtg4=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ73JkJnZZWZWCCGf4MWDIl79Im/+jZNkDxotaCiquunuilLBjfX9L6+wtr6xuVXcLu3s7u0flA+PWibJNMMmS0SiOxE1KLjCpuVWYCfVSGUksB2Nb+Z++xG14Yl6sJMUQ0mHisecUeuke3kd9MsVv+YvQP6SICcVyNHolz97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 923 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 924 | sha1_base64="Zv5ybfKo/H8QvXQaOWqQVyOFtg4=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ73JkJnZZWZWCCGf4MWDIl79Im/+jZNkDxotaCiquunuilLBjfX9L6+wtr6xuVXcLu3s7u0flA+PWibJNMMmS0SiOxE1KLjCpuVWYCfVSGUksB2Nb+Z++xG14Yl6sJMUQ0mHisecUeuke3kd9MsVv+YvQP6SICcVyNHolz97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 925 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 926 | sha1_base64="hP+6LrUf2d3tZaldqaQQvEKMXyw=">AAAB2XicbZDNSgMxFIXv1L86Vq1rN8EiuCozbnQpuHFZwbZCO5RM5k4bmskMyR2hDH0BF25EfC93vo3pz0JbDwQ+zknIvSculLQUBN9ebWd3b/+gfugfNfzjk9Nmo2fz0gjsilzl5jnmFpXU2CVJCp8LgzyLFfbj6f0i77+gsTLXTzQrMMr4WMtUCk7O6oyaraAdLMW2IVxDC9YaNb+GSS7KDDUJxa0dhEFBUcUNSaFw7g9LiwUXUz7GgUPNM7RRt | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 927 | DUJxa0dhEFBUcUNSaFw7g9LiwUXUz7GgUPNM7RRtRxzzi6dk7A0N+5oYkv394uKZ9bOstjdzDhN7Ga2MP/LBiWlt1EldVESarH6KC0Vo5wtdmaJNChIzRxwYaSblYkJN1yQa8Z3HYSbG29D77odBu3wMYA6nMMFXEEIN3AHD9CBLghI4BXevYn35n2suqp569LO4I+8zx84xIo4</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 928 | sha1_base64="41wuc+jmacSzA8ipzMzk3JQ30Cs=">AAAB33icbZDNSgMxFIXv1L9aq1a3boJFcFUybnQjCG5cVrS10A4lk95pQ5PMkGSEMvQR3LhQxLdy59uY/iy09UDg45yE3HviTArrKP0OShubW9s75d3KXnX/4LB2VG3bNDccWzyVqenEzKIUGltOOImdzCBTscSneHw7y5+e0ViR6kc3yTBSbKhFIjhz3npQ12G/VqcNOhdZh3AJdViq2a999QYpzxVqxyWzthvSzEUFM05widNKL7eYMT5mQ+x61 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 929 | QYpzxVqxyWzthvSzEUFM05widNKL7eYMT5mQ+x61EyhjYr5qFNy5p0BSVLjj3Zk7v5+UTBl7UTF/qZibmRXs5n5X9bNXXIVFUJnuUPNFx8luSQuJbO9yUAY5E5OPDBuhJ+V8BEzjDvfTsWXEK6uvA7ti0ZIG+E9hTKcwCmcQwiXcAN30IQWcBjCC7zBeyCD1+BjUVcpWPZ2DH8UfP4Arc6MHg==</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 930 | sha1_base64="41wuc+jmacSzA8ipzMzk3JQ30Cs=">AAAB33icbZDNSgMxFIXv1L9aq1a3boJFcFUybnQjCG5cVrS10A4lk95pQ5PMkGSEMvQR3LhQxLdy59uY/iy09UDg45yE3HviTArrKP0OShubW9s75d3KXnX/4LB2VG3bNDccWzyVqenEzKIUGltOOImdzCBTscSneHw7y5+e0ViR6kc3yTBSbKhFIjhz3npQ12G/VqcNOhdZh3AJdViq2a999QYpzxVqxyWzthvSzEUFM05widNKL7eYMT5mQ+x61 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 931 | QYpzxVqxyWzthvSzEUFM05widNKL7eYMT5mQ+x61EyhjYr5qFNy5p0BSVLjj3Zk7v5+UTBl7UTF/qZibmRXs5n5X9bNXXIVFUJnuUPNFx8luSQuJbO9yUAY5E5OPDBuhJ+V8BEzjDvfTsWXEK6uvA7ti0ZIG+E9hTKcwCmcQwiXcAN30IQWcBjCC7zBeyCD1+BjUVcpWPZ2DH8UfP4Arc6MHg==</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 932 | sha1_base64="c5O5JuZLnHr2YYnYcAr/a2hwAn0=">AAAB6nicbVA9SwNBEJ2LXzF+RS1tFoOQKtzZaCMEbCwjmg9IjrC3mSRLdveO3T0hHPkJNhaK2PqL7Pw3bpIrNPHBwOO9GWbmRYngxvr+t1fY2Nza3inulvb2Dw6PyscnLROnmmGTxSLWnYgaFFxh03IrsJNopDIS2I4mt3O//YTa8Fg92mmCoaQjxYecUeukB3kT9MsVv+YvQNZJkJMK5Gj0y1+9QcxSicoyQY3pBn5iw4xqy5nAWamXGkwom9ARd | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 933 | 1+9QcxSicoyQY3pBn5iw4xqy5nAWamXGkwom9ARdh1VVKIJs8WpM3LhlAEZxtqVsmSh/p7IqDRmKiPXKakdm1VvLv7ndVM7vA4zrpLUomLLRcNUEBuT+d9kwDUyK6aOUKa5u5WwMdWUWZdOyYUQrL68TlqXtcCvBfd+pV7N4yjCGZxDFQK4gjrcQQOawGAEz/AKb57wXrx372PZWvDymVP4A+/zB74jjVU=</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 934 | sha1_base64="Zv5ybfKo/H8QvXQaOWqQVyOFtg4=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ73JkJnZZWZWCCGf4MWDIl79Im/+jZNkDxotaCiquunuilLBjfX9L6+wtr6xuVXcLu3s7u0flA+PWibJNMMmS0SiOxE1KLjCpuVWYCfVSGUksB2Nb+Z++xG14Yl6sJMUQ0mHisecUeuke3kd9MsVv+YvQP6SICcVyNHolz97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 935 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 936 | sha1_base64="Zv5ybfKo/H8QvXQaOWqQVyOFtg4=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ73JkJnZZWZWCCGf4MWDIl79Im/+jZNkDxotaCiquunuilLBjfX9L6+wtr6xuVXcLu3s7u0flA+PWibJNMMmS0SiOxE1KLjCpuVWYCfVSGUksB2Nb+Z++xG14Yl6sJMUQ0mHisecUeuke3kd9MsVv+YvQP6SICcVyNHolz97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 937 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 938 | sha1_base64="Zv5ybfKo/H8QvXQaOWqQVyOFtg4=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ73JkJnZZWZWCCGf4MWDIl79Im/+jZNkDxotaCiquunuilLBjfX9L6+wtr6xuVXcLu3s7u0flA+PWibJNMMmS0SiOxE1KLjCpuVWYCfVSGUksB2Nb+Z++xG14Yl6sJMUQ0mHisecUeuke3kd9MsVv+YvQP6SICcVyNHolz97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 939 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 940 | sha1_base64="Zv5ybfKo/H8QvXQaOWqQVyOFtg4=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ73JkJnZZWZWCCGf4MWDIl79Im/+jZNkDxotaCiquunuilLBjfX9L6+wtr6xuVXcLu3s7u0flA+PWibJNMMmS0SiOxE1KLjCpuVWYCfVSGUksB2Nb+Z++xG14Yl6sJMUQ0mHisecUeuke3kd9MsVv+YvQP6SICcVyNHolz97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 941 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 942 | sha1_base64="Zv5ybfKo/H8QvXQaOWqQVyOFtg4=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ73JkJnZZWZWCCGf4MWDIl79Im/+jZNkDxotaCiquunuilLBjfX9L6+wtr6xuVXcLu3s7u0flA+PWibJNMMmS0SiOxE1KLjCpuVWYCfVSGUksB2Nb+Z++xG14Yl6sJMUQ0mHisecUeuke3kd9MsVv+YvQP6SICcVyNHolz97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 943 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 944 | sha1_base64="Zv5ybfKo/H8QvXQaOWqQVyOFtg4=">AAAB6nicbVDLSgNBEOyNrxhfUY9eBoOQU9gVQS9CwIvHiOYByRJmJ73JkJnZZWZWCCGf4MWDIl79Im/+jZNkDxotaCiquunuilLBjfX9L6+wtr6xuVXcLu3s7u0flA+PWibJNMMmS0SiOxE1KLjCpuVWYCfVSGUksB2Nb+Z++xG14Yl6sJMUQ0mHisecUeuke3kd9MsVv+YvQP6SICcVyNHolz97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 945 | z97g4RlEpVlghrTDfzUhlOqLWcCZ6VeZjClbEyH2HVUUYkmnC5OnZEzpwxInGhXypKF+nNiSqUxExm5TkntyKx6c/E/r5vZ+CqccpVmFhVbLoozQWxC5n+TAdfIrJg4Qpnm7lbCRlRTZl06JRdCsPryX9I6rwV+Lbi7qNSreRxFOIFTqEIAl1CHW2hAExgM4Qle4NUT3rP35r0vWwtePnMMv+B9fAO/Y41Z</latexit>m=2 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 946 | <latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 947 | sha1_base64="oVNcNejQAmVb9Nn6VIeeV7f50ls=">AAAB6nicbVBNSwMxEJ3Ur1q/qh69BIvQU9ktgl6EghePFe0HtEvJptk2NMkuSVYoS3+CFw+KePUXefPfmLZ70NYHA4/3ZpiZFyaCG+t536iwsbm1vVPcLe3tHxwelY9P2iZONWUtGotYd0NimOCKtSy3gnUTzYgMBeuEk9u533li2vBYPdppwgJJRopHnBLrpAd5Ux+UK17NWwCvEz8nFcjRHJS/+sOYppIpSwUxpud7iQ0yoi2ngs1K/dSwhNAJG | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 948 | JS/+sOYppIpSwUxpud7iQ0yoi2ngs1K/dSwhNAJGbGeo4pIZoJsceoMXzhliKNYu1IWL9TfExmRxkxl6DolsWOz6s3F/7xeaqPrIOMqSS1TdLkoSgW2MZ7/jYdcM2rF1BFCNXe3YjommlDr0im5EPzVl9dJu17zvZp/f1lpVPM4inAG51AFH66gAXfQhBZQGMEzvMIbEugFvaOPZWsB5TOn8Afo8wfA541a</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 949 | sha1_base64="oVNcNejQAmVb9Nn6VIeeV7f50ls=">AAAB6nicbVBNSwMxEJ3Ur1q/qh69BIvQU9ktgl6EghePFe0HtEvJptk2NMkuSVYoS3+CFw+KePUXefPfmLZ70NYHA4/3ZpiZFyaCG+t536iwsbm1vVPcLe3tHxwelY9P2iZONWUtGotYd0NimOCKtSy3gnUTzYgMBeuEk9u533li2vBYPdppwgJJRopHnBLrpAd5Ux+UK17NWwCvEz8nFcjRHJS/+sOYppIpSwUxpud7iQ0yoi2ngs1K/dSwhNAJG | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 950 | JS/+sOYppIpSwUxpud7iQ0yoi2ngs1K/dSwhNAJGbGeo4pIZoJsceoMXzhliKNYu1IWL9TfExmRxkxl6DolsWOz6s3F/7xeaqPrIOMqSS1TdLkoSgW2MZ7/jYdcM2rF1BFCNXe3YjommlDr0im5EPzVl9dJu17zvZp/f1lpVPM4inAG51AFH66gAXfQhBZQGMEzvMIbEugFvaOPZWsB5TOn8Afo8wfA541a</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 951 | sha1_base64="oVNcNejQAmVb9Nn6VIeeV7f50ls=">AAAB6nicbVBNSwMxEJ3Ur1q/qh69BIvQU9ktgl6EghePFe0HtEvJptk2NMkuSVYoS3+CFw+KePUXefPfmLZ70NYHA4/3ZpiZFyaCG+t536iwsbm1vVPcLe3tHxwelY9P2iZONWUtGotYd0NimOCKtSy3gnUTzYgMBeuEk9u533li2vBYPdppwgJJRopHnBLrpAd5Ux+UK17NWwCvEz8nFcjRHJS/+sOYppIpSwUxpud7iQ0yoi2ngs1K/dSwhNAJG | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 952 | JS/+sOYppIpSwUxpud7iQ0yoi2ngs1K/dSwhNAJGbGeo4pIZoJsceoMXzhliKNYu1IWL9TfExmRxkxl6DolsWOz6s3F/7xeaqPrIOMqSS1TdLkoSgW2MZ7/jYdcM2rF1BFCNXe3YjommlDr0im5EPzVl9dJu17zvZp/f1lpVPM4inAG51AFH66gAXfQhBZQGMEzvMIbEugFvaOPZWsB5TOn8Afo8wfA541a</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 953 | sha1_base64="oVNcNejQAmVb9Nn6VIeeV7f50ls=">AAAB6nicbVBNSwMxEJ3Ur1q/qh69BIvQU9ktgl6EghePFe0HtEvJptk2NMkuSVYoS3+CFw+KePUXefPfmLZ70NYHA4/3ZpiZFyaCG+t536iwsbm1vVPcLe3tHxwelY9P2iZONWUtGotYd0NimOCKtSy3gnUTzYgMBeuEk9u533li2vBYPdppwgJJRopHnBLrpAd5Ux+UK17NWwCvEz8nFcjRHJS/+sOYppIpSwUxpud7iQ0yoi2ngs1K/dSwhNAJG | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 954 | JS/+sOYppIpSwUxpud7iQ0yoi2ngs1K/dSwhNAJGbGeo4pIZoJsceoMXzhliKNYu1IWL9TfExmRxkxl6DolsWOz6s3F/7xeaqPrIOMqSS1TdLkoSgW2MZ7/jYdcM2rF1BFCNXe3YjommlDr0im5EPzVl9dJu17zvZp/f1lpVPM4inAG51AFH66gAXfQhBZQGMEzvMIbEugFvaOPZWsB5TOn8Afo8wfA541a</latexit>m=3 | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 955 | <latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 956 | sha1_base64="+jOsYdJ9SNJymWDVPaXJffNRSPM=">AAAB6nicbVBNSwMxEJ3Ur1q/qh69BIvQU9lVQS9CwYvHivYD2qVk02wbmmSXJCuUpT/BiwdFvPqLvPlvTNs9aOuDgcd7M8zMCxPBjfW8b1RYW9/Y3Cpul3Z29/YPyodHLROnmrImjUWsOyExTHDFmpZbwTqJZkSGgrXD8e3Mbz8xbXisHu0kYYEkQ8UjTol10oO8ueiXK17NmwOvEj8nFcjR6Je/eoOYppIpSwUxput7iQ0yoi2ngk1LvdSwhNAxG | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 957 | Je/eoOYppIpSwUxput7iQ0yoi2ngk1LvdSwhNAxGbKuo4pIZoJsfuoUnzllgKNYu1IWz9XfExmRxkxk6DolsSOz7M3E/7xuaqPrIOMqSS1TdLEoSgW2MZ79jQdcM2rFxBFCNXe3YjoimlDr0im5EPzll1dJ67zmezX//rJSr+ZxFOEETqEKPlxBHe6gAU2gMIRneIU3JNALekcfi9YCymeO4Q/Q5w/Ca41b</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 958 | sha1_base64="+jOsYdJ9SNJymWDVPaXJffNRSPM=">AAAB6nicbVBNSwMxEJ3Ur1q/qh69BIvQU9lVQS9CwYvHivYD2qVk02wbmmSXJCuUpT/BiwdFvPqLvPlvTNs9aOuDgcd7M8zMCxPBjfW8b1RYW9/Y3Cpul3Z29/YPyodHLROnmrImjUWsOyExTHDFmpZbwTqJZkSGgrXD8e3Mbz8xbXisHu0kYYEkQ8UjTol10oO8ueiXK17NmwOvEj8nFcjR6Je/eoOYppIpSwUxput7iQ0yoi2ngk1LvdSwhNAxG | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 959 | Je/eoOYppIpSwUxput7iQ0yoi2ngk1LvdSwhNAxGbKuo4pIZoJsfuoUnzllgKNYu1IWz9XfExmRxkxk6DolsSOz7M3E/7xuaqPrIOMqSS1TdLEoSgW2MZ79jQdcM2rFxBFCNXe3YjoimlDr0im5EPzll1dJ67zmezX//rJSr+ZxFOEETqEKPlxBHe6gAU2gMIRneIU3JNALekcfi9YCymeO4Q/Q5w/Ca41b</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 960 | sha1_base64="+jOsYdJ9SNJymWDVPaXJffNRSPM=">AAAB6nicbVBNSwMxEJ3Ur1q/qh69BIvQU9lVQS9CwYvHivYD2qVk02wbmmSXJCuUpT/BiwdFvPqLvPlvTNs9aOuDgcd7M8zMCxPBjfW8b1RYW9/Y3Cpul3Z29/YPyodHLROnmrImjUWsOyExTHDFmpZbwTqJZkSGgrXD8e3Mbz8xbXisHu0kYYEkQ8UjTol10oO8ueiXK17NmwOvEj8nFcjR6Je/eoOYppIpSwUxput7iQ0yoi2ngk1LvdSwhNAxG | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 961 | Je/eoOYppIpSwUxput7iQ0yoi2ngk1LvdSwhNAxGbKuo4pIZoJsfuoUnzllgKNYu1IWz9XfExmRxkxk6DolsSOz7M3E/7xuaqPrIOMqSS1TdLEoSgW2MZ79jQdcM2rFxBFCNXe3YjoimlDr0im5EPzll1dJ67zmezX//rJSr+ZxFOEETqEKPlxBHe6gAU2gMIRneIU3JNALekcfi9YCymeO4Q/Q5w/Ca41b</latexit><latexit | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 962 | sha1_base64="+jOsYdJ9SNJymWDVPaXJffNRSPM=">AAAB6nicbVBNSwMxEJ3Ur1q/qh69BIvQU9lVQS9CwYvHivYD2qVk02wbmmSXJCuUpT/BiwdFvPqLvPlvTNs9aOuDgcd7M8zMCxPBjfW8b1RYW9/Y3Cpul3Z29/YPyodHLROnmrImjUWsOyExTHDFmpZbwTqJZkSGgrXD8e3Mbz8xbXisHu0kYYEkQ8UjTol10oO8ueiXK17NmwOvEj8nFcjR6Je/eoOYppIpSwUxput7iQ0yoi2ngk1LvdSwhNAxG | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 963 | Je/eoOYppIpSwUxput7iQ0yoi2ngk1LvdSwhNAxGbKuo4pIZoJsfuoUnzllgKNYu1IWz9XfExmRxkxk6DolsSOz7M3E/7xuaqPrIOMqSS1TdLEoSgW2MZ79jQdcM2rFxBFCNXe3YjoimlDr0im5EPzll1dJ67zmezX//rJSr+ZxFOEETqEKPlxBHe6gAU2gMIRneIU3JNALekcfi9YCymeO4Q/Q5w/Ca41b</latexit>Figure | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 964 | 7: Example of message passing. Each row highlights the information that di uses through | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 965 | the graph starting from a particular node. In the top row, the node of interest is in the upper
right; in the bottom row, the node of interest is in the bottom right. Shaded nodes indicate how far
information from the original node can travel in msteps of message passing; bolded edges indicate
which edges that information has the potential to travel across. Note that during the full message
passing procedure, this propagation of information happens simultaneously for all nodes and edges
in the graph (not just the two shown here).
apply an elementary dynamics update, and the decoder might read out the nal positions from the
updated graph state.
Similar to the encode-process-decode design, recurrent GN-based architectures can be built by
maintaining a hidden graph, Gt
hid, taking as input an observed graph, Gt
inp, and returning an output
graph,Gt
out, on each step (see Figure 6c). This type of architecture can be particularly useful for
predicting sequences of graphs, such as predicting the trajectory of a dynamical system over time
(e.g. Sanchez-Gonzalez et al., 2018). The encoded graph, output by GNenc, must have the same
structure as Gt
hid, and they can be easily combined by concatenating their corresponding ek,vi, | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 966 | structure as Gt
hid, and they can be easily combined by concatenating their corresponding ek,vi,
anduvectors (where the upward arrow merges into the left-hand horizontal arrow in Figure 6c),
before being passed to GNcore. For the output, the Gt
hidis copied (where the right-hand horizontal
arrow splits into the downward arrow in Figure 6c) and decoded by GNdec. This design reuses GN
blocks in several ways: GNenc,GNdec, and GNcoreare shared across each step, t; and within each
step, GN coremay perform multiple shared sub-steps.
Various other techniques for designing GN-based architectures can be useful. Graph skip
connections, for example, would concatenate a GN block's input graph, Gm, with its output graph,
Gm+1, before proceeding to further computations. Merging and smoothing input and hidden graph
information, as in Figure 6c, can use LSTM- or GRU-style gating schemes, instead of simple
concatenation (Li et al., 2016). Or distinct, recurrent GN blocks (e.g. Figure 4b) can be composed
before and/or after other GN blocks, to improve stability in the representations over multiple
propagation steps (Sanchez-Gonzalez et al., 2018). | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
1806.01261 | 967 | before and/or after other GN blocks, to improve stability in the representations over multiple
propagation steps (Sanchez-Gonzalez et al., 2018).
4.4 Implementing graph networks in code
Similar to CNNs (see Figure 1), which are naturally parallelizable (e.g. on GPUs), GNs have a
natural parallel structure: since the eand vfunctions in Equation 1 are shared over the edges
and nodes, respectively, they can be computed in parallel. In practice, this means that with respect
20
Box 4: Graph Nets open-source software library: github.com/deepmind/graph nets
We have released an open-source library for building GNs in Tensor
ow/Sonnet. It includes
demos of how to create, manipulate, and train GNs to reason about graph-structured data, on
a shortest path- nding task, a sorting task, and a physical prediction task. Each demo uses the
same GN architecture, which highlights the
exibility of the approach.
Shortest path demo: tinyurl.com/gn-shortest-path-demo
This demo creates random graphs, and trains a GN to label the nodes and edges on the shortest
path between any two nodes. Over a sequence of message-passing steps (as depicted by each | 1806.01261 | Relational inductive biases, deep learning, and graph networks | Artificial intelligence (AI) has undergone a renaissance recently, making
major progress in key domains such as vision, language, control, and
decision-making. This has been due, in part, to cheap data and cheap compute
resources, which have fit the natural strengths of deep learning. However, many
defining characteristics of human intelligence, which developed under much
different pressures, remain out of reach for current approaches. In particular,
generalizing beyond one's experiences--a hallmark of human intelligence from
infancy--remains a formidable challenge for modern AI.
The following is part position paper, part review, and part unification. We
argue that combinatorial generalization must be a top priority for AI to
achieve human-like abilities, and that structured representations and
computations are key to realizing this objective. Just as biology uses nature
and nurture cooperatively, we reject the false choice between
"hand-engineering" and "end-to-end" learning, and instead advocate for an
approach which benefits from their complementary strengths. We explore how
using relational inductive biases within deep learning architectures can
facilitate learning about entities, relations, and rules for composing them. We
present a new building block for the AI toolkit with a strong relational
inductive bias--the graph network--which generalizes and extends various
approaches for neural networks that operate on graphs, and provides a
straightforward interface for manipulating structured knowledge and producing
structured behaviors. We discuss how graph networks can support relational
reasoning and combinatorial generalization, laying the foundation for more
sophisticated, interpretable, and flexible patterns of reasoning. As a
companion to this paper, we have released an open-source software library for
building graph networks, with demonstrations of how to use them in practice. | http://arxiv.org/pdf/1806.01261 | [
"Peter W. Battaglia",
"Jessica B. Hamrick",
"Victor Bapst",
"Alvaro Sanchez-Gonzalez",
"Vinicius Zambaldi",
"Mateusz Malinowski",
"Andrea Tacchetti",
"David Raposo",
"Adam Santoro",
"Ryan Faulkner",
"Caglar Gulcehre",
"Francis Song",
"Andrew Ballard",
"Justin Gilmer",
"George Dahl",
"Ashish Vaswani",
"Kelsey Allen",
"Charles Nash",
"Victoria Langston",
"Chris Dyer",
"Nicolas Heess",
"Daan Wierstra",
"Pushmeet Kohli",
"Matt Botvinick",
"Oriol Vinyals",
"Yujia Li",
"Razvan Pascanu"
] | [
"cs.LG",
"cs.AI",
"stat.ML"
] | null | null | cs.LG | 20180604 | 20181017 | [] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.