Datasets:

jamescalam
/

llama-2-arxiv-papers-chunked

Dataset card Data Studio Files Files and versions Community

Split (1)

train · 4.84k rows

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	268	sha1_base64="74MJShuZzxGyM2nLY3EY87InuhI=">AAAB7XicbVC7SgNBFJ2NrxhfUcHGZjAIVmHXRguLgI1lBLMJJGuYndxNxszODjOzyrLkH2wsFNHS/7HzA/wPJ49CEw9cOJxzL/feE0rOtHHdL6ewtLyyulZcL21sbm3vlHf3fJ2kikKDJjxRrZBo4ExAwzDDoSUVkDjk0AyHl2O/eQ9Ks0TcmExCEJO+YBGjxFjJ78gBu4VuueJW3QnwIvFmpFI78B9k9v1e75Y/O72EpjEIQznRuu250gQ5UYZRD	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	269	FI78B9k9v1e75Y/O72EpjEIQznRuu250gQ5UYZRDqNSJ9UgCR2SPrQtFSQGHeSTa0f42Co9HCXKljB4ov6eyEmsdRaHtjMmZqDnvbH4n9dOTXQe5EzI1ICg00VRyrFJ8Ph13GMKqOGZJYQqZm/FdEAUocYGVLIhePMvLxL/tOq5Ve/apnGBpiiiQ3SETpCHzlANXaE6aiCK7tAjekYvTuI8Oa/O27S14Mxm9tEfOB8/TZOSnA==</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	270	sha1_base64="74MJShuZzxGyM2nLY3EY87InuhI=">AAAB7XicbVC7SgNBFJ2NrxhfUcHGZjAIVmHXRguLgI1lBLMJJGuYndxNxszODjOzyrLkH2wsFNHS/7HzA/wPJ49CEw9cOJxzL/feE0rOtHHdL6ewtLyyulZcL21sbm3vlHf3fJ2kikKDJjxRrZBo4ExAwzDDoSUVkDjk0AyHl2O/eQ9Ks0TcmExCEJO+YBGjxFjJ78gBu4VuueJW3QnwIvFmpFI78B9k9v1e75Y/O72EpjEIQznRuu250gQ5UYZRD	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	271	FI78B9k9v1e75Y/O72EpjEIQznRuu250gQ5UYZRDqNSJ9UgCR2SPrQtFSQGHeSTa0f42Co9HCXKljB4ov6eyEmsdRaHtjMmZqDnvbH4n9dOTXQe5EzI1ICg00VRyrFJ8Ph13GMKqOGZJYQqZm/FdEAUocYGVLIhePMvLxL/tOq5Ve/apnGBpiiiQ3SETpCHzlANXaE6aiCK7tAjekYvTuI8Oa/O27S14Mxm9tEfOB8/TZOSnA==</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	272	sha1_base64="pLq6KB/1S9uyUeWp/G4byg43mK0=">AAAB7XicbVA9SwNBEJ2LXzF+RS1tFoNgFe5stLAI2FhGMB+QnGFvM5es2ds9dveEEPIfbCwUsfX/2Plv3CRXaOKDgcd7M8zMi1LBjfX9b6+wtr6xuVXcLu3s7u0flA+PmkZlmmGDKaF0O6IGBZfYsNwKbKcaaRIJbEWjm5nfekJtuJL3dpximNCB5DFn1Dqp2U2H/AF75Ypf9ecgqyTISQVy1Hvlr25fsSxBaZmgxnQCP7XhhGrLmcBpqZsZTCkb0	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	273	Hvlr25fsSxBaZmgxnQCP7XhhGrLmcBpqZsZTCkb0QF2HJU0QRNO5tdOyZlT+iRW2pW0ZK7+npjQxJhxErnOhNqhWfZm4n9eJ7PxVTjhMs0sSrZYFGeCWEVmr5M+18isGDtCmebuVsKGVFNmXUAlF0Kw/PIqaV5UA78a3PmV2nUeRxFO4BTOIYBLqMEt1KEBDB7hGV7hzVPei/fufSxaC14+cwx/4H3+AIWDjww=</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	274	(b) Independent recurrent block	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	275	Edge blockNode blockGlobal blockV0	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	276	<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	277	sha1_base64="gAQ7qdt3IKvK5oBqK3uN1PHYi1k=">AAAB6XicbVA9SwNBEJ2LXzF+RS1tFoNoFe4koIVFwMYyivmA5Ah7m7lkyd7esbsnhCP/wMZCEVv/kZ3/xk1yhSY+GHi8N8PMvCARXBvX/XYKa+sbm1vF7dLO7t7+QfnwqKXjVDFssljEqhNQjYJLbBpuBHYShTQKBLaD8e3Mbz+h0jyWj2aSoB/RoeQhZ9RY6aF13i9X3Ko7B1klXk4qkKPRL3/1BjFLI5SGCap113MT42dUGc4ETku9VGNC2ZgOs	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	278	3/1BjFLI5SGCap113MT42dUGc4ETku9VGNC2ZgOsWuppBFqP5tfOiVnVhmQMFa2pCFz9fdERiOtJ1FgOyNqRnrZm4n/ed3UhNd+xmWSGpRssShMBTExmb1NBlwhM2JiCWWK21sJG1FFmbHhlGwI3vLLq6R1WfXcqndfq9Rv8jiKcAKncAEeXEEd7qABTWAQwjO8wpszdl6cd+dj0Vpw8plj+APn8wcRSI0F</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	279	sha1_base64="gAQ7qdt3IKvK5oBqK3uN1PHYi1k=">AAAB6XicbVA9SwNBEJ2LXzF+RS1tFoNoFe4koIVFwMYyivmA5Ah7m7lkyd7esbsnhCP/wMZCEVv/kZ3/xk1yhSY+GHi8N8PMvCARXBvX/XYKa+sbm1vF7dLO7t7+QfnwqKXjVDFssljEqhNQjYJLbBpuBHYShTQKBLaD8e3Mbz+h0jyWj2aSoB/RoeQhZ9RY6aF13i9X3Ko7B1klXk4qkKPRL3/1BjFLI5SGCap113MT42dUGc4ETku9VGNC2ZgOs	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	280	3/1BjFLI5SGCap113MT42dUGc4ETku9VGNC2ZgOsWuppBFqP5tfOiVnVhmQMFa2pCFz9fdERiOtJ1FgOyNqRnrZm4n/ed3UhNd+xmWSGpRssShMBTExmb1NBlwhM2JiCWWK21sJG1FFmbHhlGwI3vLLq6R1WfXcqndfq9Rv8jiKcAKncAEeXEEd7qABTWAQwjO8wpszdl6cd+dj0Vpw8plj+APn8wcRSI0F</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	281	sha1_base64="gAQ7qdt3IKvK5oBqK3uN1PHYi1k=">AAAB6XicbVA9SwNBEJ2LXzF+RS1tFoNoFe4koIVFwMYyivmA5Ah7m7lkyd7esbsnhCP/wMZCEVv/kZ3/xk1yhSY+GHi8N8PMvCARXBvX/XYKa+sbm1vF7dLO7t7+QfnwqKXjVDFssljEqhNQjYJLbBpuBHYShTQKBLaD8e3Mbz+h0jyWj2aSoB/RoeQhZ9RY6aF13i9X3Ko7B1klXk4qkKPRL3/1BjFLI5SGCap113MT42dUGc4ETku9VGNC2ZgOs	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	282	3/1BjFLI5SGCap113MT42dUGc4ETku9VGNC2ZgOsWuppBFqP5tfOiVnVhmQMFa2pCFz9fdERiOtJ1FgOyNqRnrZm4n/ed3UhNd+xmWSGpRssShMBTExmb1NBlwhM2JiCWWK21sJG1FFmbHhlGwI3vLLq6R1WfXcqndfq9Rv8jiKcAKncAEeXEEd7qABTWAQwjO8wpszdl6cd+dj0Vpw8plj+APn8wcRSI0F</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	283	sha1_base64="gAQ7qdt3IKvK5oBqK3uN1PHYi1k=">AAAB6XicbVA9SwNBEJ2LXzF+RS1tFoNoFe4koIVFwMYyivmA5Ah7m7lkyd7esbsnhCP/wMZCEVv/kZ3/xk1yhSY+GHi8N8PMvCARXBvX/XYKa+sbm1vF7dLO7t7+QfnwqKXjVDFssljEqhNQjYJLbBpuBHYShTQKBLaD8e3Mbz+h0jyWj2aSoB/RoeQhZ9RY6aF13i9X3Ko7B1klXk4qkKPRL3/1BjFLI5SGCap113MT42dUGc4ETku9VGNC2ZgOs	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	284	3/1BjFLI5SGCap113MT42dUGc4ETku9VGNC2ZgOsWuppBFqP5tfOiVnVhmQMFa2pCFz9fdERiOtJ1FgOyNqRnrZm4n/ed3UhNd+xmWSGpRssShMBTExmb1NBlwhM2JiCWWK21sJG1FFmbHhlGwI3vLLq6R1WfXcqndfq9Rv8jiKcAKncAEeXEEd7qABTWAQwjO8wpszdl6cd+dj0Vpw8plj+APn8wcRSI0F</latexit>u0	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	285	<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	286	sha1_base64="Z/n1gIms2/ONBt0R58c8NGdBbqU=">AAAB8nicbVDLSsNAFL2pr1pfVZduBovoqiQi6MJFwY3LCvYBbSiT6aQdOpmEmRuhhH6GGxeKuPVr3Pk3TtostPXAwOGce5lzT5BIYdB1v53S2vrG5lZ5u7Kzu7d/UD08aps41Yy3WCxj3Q2o4VIo3kKBkncTzWkUSN4JJne533ni2ohYPeI04X5ER0qEglG0Uq8fURwHYZbOzgfVmlt35yCrxCtIDQo0B9Wv/jBmacQVMkmN6Xlugn5GNQom+azST	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	287	CtIDQo0B9Wv/jBmacQVMkmN6Xlugn5GNQom+azSTw1PKJvQEe9ZqmjEjZ/NI8/ImVWGJIy1fQrJXP29kdHImGkU2Mk8oln2cvE/r5dieONnQiUpcsUWH4WpJBiT/H4yFJozlFNLKNPCZiVsTDVlaFuq2BK85ZNXSfuy7rl17+Gq1rgt6ijDCZzCBXhwDQ24hya0gEEMz/AKbw46L86787EYLTnFzjH8gfP5A1s4kUQ=</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	288	sha1_base64="Z/n1gIms2/ONBt0R58c8NGdBbqU=">AAAB8nicbVDLSsNAFL2pr1pfVZduBovoqiQi6MJFwY3LCvYBbSiT6aQdOpmEmRuhhH6GGxeKuPVr3Pk3TtostPXAwOGce5lzT5BIYdB1v53S2vrG5lZ5u7Kzu7d/UD08aps41Yy3WCxj3Q2o4VIo3kKBkncTzWkUSN4JJne533ni2ohYPeI04X5ER0qEglG0Uq8fURwHYZbOzgfVmlt35yCrxCtIDQo0B9Wv/jBmacQVMkmN6Xlugn5GNQom+azST	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	289	CtIDQo0B9Wv/jBmacQVMkmN6Xlugn5GNQom+azSTw1PKJvQEe9ZqmjEjZ/NI8/ImVWGJIy1fQrJXP29kdHImGkU2Mk8oln2cvE/r5dieONnQiUpcsUWH4WpJBiT/H4yFJozlFNLKNPCZiVsTDVlaFuq2BK85ZNXSfuy7rl17+Gq1rgt6ijDCZzCBXhwDQ24hya0gEEMz/AKbw46L86787EYLTnFzjH8gfP5A1s4kUQ=</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	290	sha1_base64="Z/n1gIms2/ONBt0R58c8NGdBbqU=">AAAB8nicbVDLSsNAFL2pr1pfVZduBovoqiQi6MJFwY3LCvYBbSiT6aQdOpmEmRuhhH6GGxeKuPVr3Pk3TtostPXAwOGce5lzT5BIYdB1v53S2vrG5lZ5u7Kzu7d/UD08aps41Yy3WCxj3Q2o4VIo3kKBkncTzWkUSN4JJne533ni2ohYPeI04X5ER0qEglG0Uq8fURwHYZbOzgfVmlt35yCrxCtIDQo0B9Wv/jBmacQVMkmN6Xlugn5GNQom+azST	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	291	CtIDQo0B9Wv/jBmacQVMkmN6Xlugn5GNQom+azSTw1PKJvQEe9ZqmjEjZ/NI8/ImVWGJIy1fQrJXP29kdHImGkU2Mk8oln2cvE/r5dieONnQiUpcsUWH4WpJBiT/H4yFJozlFNLKNPCZiVsTDVlaFuq2BK85ZNXSfuy7rl17+Gq1rgt6ijDCZzCBXhwDQ24hya0gEEMz/AKbw46L86787EYLTnFzjH8gfP5A1s4kUQ=</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	292	sha1_base64="Z/n1gIms2/ONBt0R58c8NGdBbqU=">AAAB8nicbVDLSsNAFL2pr1pfVZduBovoqiQi6MJFwY3LCvYBbSiT6aQdOpmEmRuhhH6GGxeKuPVr3Pk3TtostPXAwOGce5lzT5BIYdB1v53S2vrG5lZ5u7Kzu7d/UD08aps41Yy3WCxj3Q2o4VIo3kKBkncTzWkUSN4JJne533ni2ohYPeI04X5ER0qEglG0Uq8fURwHYZbOzgfVmlt35yCrxCtIDQo0B9Wv/jBmacQVMkmN6Xlugn5GNQom+azST	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	293	CtIDQo0B9Wv/jBmacQVMkmN6Xlugn5GNQom+azSTw1PKJvQEe9ZqmjEjZ/NI8/ImVWGJIy1fQrJXP29kdHImGkU2Mk8oln2cvE/r5dieONnQiUpcsUWH4WpJBiT/H4yFJozlFNLKNPCZiVsTDVlaFuq2BK85ZNXSfuy7rl17+Gq1rgt6ijDCZzCBXhwDQ24hya0gEEMz/AKbw46L86787EYLTnFzjH8gfP5A1s4kUQ=</latexit>u	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	294	<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	295	sha1_base64="znt8hwWv6wryqwCugrweUa+jkM8=">AAAB7XicbVA9SwNBEJ3zM8avqGBjsxgEq3Bno4VFwMYygrkEkjPubfaSNXu7y+6eEo78BxsLRWz9P3b+GzcfhSY+GHi8N8PMvFhxZqzvf3tLyyura+uFjeLm1vbObmlvPzQy04TWieRSN2NsKGeC1i2znDaVpjiNOW3Eg6ux33ik2jApbu1Q0SjFPcESRrB1UthWfXaXdUplv+JPgBZJMCPl6mH4dA8AtU7pq92VJEupsIRjY1qBr2yUY20Z4XRUb	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	296	mH4dA8AtU7pq92VJEupsIRjY1qBr2yUY20Z4XRUbGeGKkwGuEdbjgqcUhPlk2tH6MQpXZRI7UpYNFF/T+Q4NWaYxq4zxbZv5r2x+J/XymxyEeVMqMxSQaaLkowjK9H4ddRlmhLLh45gopm7FZE+1phYF1DRhRDMv7xIwrNK4FeCG5fGJUxRgCM4hlMI4ByqcA01qAOBB3iGV3jzpPfivXsf09YlbzZzAH/gff4AqE6Qnw==</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	297	sha1_base64="Nc0DXje6uYB+/0fHlXL99yCq0no=">AAAB7XicbVC7SgNBFL0bXzG+ooKNzWAQrMKujRYWARvLCGYTSNYwO5lNxszODjOzyrLkH2wsFNHS/7HzA/wPJ49CEw9cOJxzL/feE0rOtHHdL6ewtLyyulZcL21sbm3vlHf3fJ2kitAGSXiiWiHWlDNBG4YZTltSURyHnDbD4eXYb95TpVkibkwmaRDjvmARI9hYye/IAbtNu+WKW3UnQIvEm5FK7cB/kNn3e71b/uz0EpLGVBjCsdZtz5UmyLEyj	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	298	5FK7cB/kNn3e71b/uz0EpLGVBjCsdZtz5UmyLEyjHA6KnVSTSUmQ9ynbUsFjqkO8sm1I3RslR6KEmVLGDRRf0/kONY6i0PbGWMz0PPeWPzPa6cmOg9yJmRqqCDTRVHKkUnQ+HXUY4oSwzNLMFHM3orIACtMjA2oZEPw5l9eJP5p1XOr3rVN4wKmKMIhHMEJeHAGNbiCOjSAwB08wjO8OInz5Lw6b9PWgjOb2Yc/cD5+AGXTkqw=</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	299	sha1_base64="Nc0DXje6uYB+/0fHlXL99yCq0no=">AAAB7XicbVC7SgNBFL0bXzG+ooKNzWAQrMKujRYWARvLCGYTSNYwO5lNxszODjOzyrLkH2wsFNHS/7HzA/wPJ49CEw9cOJxzL/feE0rOtHHdL6ewtLyyulZcL21sbm3vlHf3fJ2kitAGSXiiWiHWlDNBG4YZTltSURyHnDbD4eXYb95TpVkibkwmaRDjvmARI9hYye/IAbtNu+WKW3UnQIvEm5FK7cB/kNn3e71b/uz0EpLGVBjCsdZtz5UmyLEyj	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	300	5FK7cB/kNn3e71b/uz0EpLGVBjCsdZtz5UmyLEyjHA6KnVSTSUmQ9ynbUsFjqkO8sm1I3RslR6KEmVLGDRRf0/kONY6i0PbGWMz0PPeWPzPa6cmOg9yJmRqqCDTRVHKkUnQ+HXUY4oSwzNLMFHM3orIACtMjA2oZEPw5l9eJP5p1XOr3rVN4wKmKMIhHMEJeHAGNbiCOjSAwB08wjO8OInz5Lw6b9PWgjOb2Yc/cD5+AGXTkqw=</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	301	sha1_base64="S5XnA5iYIAgqxiI+i0ptSwAiKP4=">AAAB7XicbVA9SwNBEJ2LXzF+RS1tFoNgFe5stLAI2FhGMB+QnGFvM5es2ds9dveEEPIfbCwUsfX/2Plv3CRXaOKDgcd7M8zMi1LBjfX9b6+wtr6xuVXcLu3s7u0flA+PmkZlmmGDKaF0O6IGBZfYsNwKbKcaaRIJbEWjm5nfekJtuJL3dpximNCB5DFn1Dqp2U2H/CHrlSt+1Z+DrJIgJxXIUe+Vv7p9xbIEpWWCGtMJ/NSGE6otZwKnpW5mMKVsR	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	302	e+Vv7p9xbIEpWWCGtMJ/NSGE6otZwKnpW5mMKVsRAfYcVTSBE04mV87JWdO6ZNYaVfSkrn6e2JCE2PGSeQ6E2qHZtmbif95nczGV+GEyzSzKNliUZwJYhWZvU76XCOzYuwIZZq7WwkbUk2ZdQGVXAjB8surpHlRDfxqcOdXatd5HEU4gVM4hwAuoQa3UIcGMHiEZ3iFN095L96797FoLXj5zDH8gff5A53Djxw=</latexit>⇢v!u	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	303	<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	304	sha1_base64="8QVocR3pGD0i/QL+G1OhPZDl9fE=">AAAB/nicbVBNS8NAEN3Ur1q/ouLJy2IRPJVEBD0WvXisYD+giWWz3TRLN9mwO6mUUPCvePGgiFd/hzf/jds2B219MPB4b4aZeUEquAbH+bZKK6tr6xvlzcrW9s7unr1/0NIyU5Q1qRRSdQKimeAJawIHwTqpYiQOBGsHw5up3x4xpblM7mGcMj8mg4SHnBIwUs8+8lQkH/KRp/ggAqKUfMTZpGdXnZozA14mbkGqqECjZ395fUmzmCVABdG66zop+	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	305	GdXnZozA14mbkGqqECjZ395fUmzmCVABdG66zop+DlRwKlgk4qXaZYSOiQD1jU0ITHTfj47f4JPjdLHoVSmEsAz9fdETmKtx3FgOmMCkV70puJ/XjeD8MrPeZJmwBI6XxRmAoPE0yxwnytGQYwNIVRxcyumEVGEgkmsYkJwF19eJq3zmuvU3LuLav26iKOMjtEJOkMuukR1dIsaqIkoytEzekVv1pP1Yr1bH/PWklXMHKI/sD5/AAdVlig=</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	306	sha1_base64="8QVocR3pGD0i/QL+G1OhPZDl9fE=">AAAB/nicbVBNS8NAEN3Ur1q/ouLJy2IRPJVEBD0WvXisYD+giWWz3TRLN9mwO6mUUPCvePGgiFd/hzf/jds2B219MPB4b4aZeUEquAbH+bZKK6tr6xvlzcrW9s7unr1/0NIyU5Q1qRRSdQKimeAJawIHwTqpYiQOBGsHw5up3x4xpblM7mGcMj8mg4SHnBIwUs8+8lQkH/KRp/ggAqKUfMTZpGdXnZozA14mbkGqqECjZ395fUmzmCVABdG66zop+	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	307	GdXnZozA14mbkGqqECjZ395fUmzmCVABdG66zop+DlRwKlgk4qXaZYSOiQD1jU0ITHTfj47f4JPjdLHoVSmEsAz9fdETmKtx3FgOmMCkV70puJ/XjeD8MrPeZJmwBI6XxRmAoPE0yxwnytGQYwNIVRxcyumEVGEgkmsYkJwF19eJq3zmuvU3LuLav26iKOMjtEJOkMuukR1dIsaqIkoytEzekVv1pP1Yr1bH/PWklXMHKI/sD5/AAdVlig=</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	308	sha1_base64="8QVocR3pGD0i/QL+G1OhPZDl9fE=">AAAB/nicbVBNS8NAEN3Ur1q/ouLJy2IRPJVEBD0WvXisYD+giWWz3TRLN9mwO6mUUPCvePGgiFd/hzf/jds2B219MPB4b4aZeUEquAbH+bZKK6tr6xvlzcrW9s7unr1/0NIyU5Q1qRRSdQKimeAJawIHwTqpYiQOBGsHw5up3x4xpblM7mGcMj8mg4SHnBIwUs8+8lQkH/KRp/ggAqKUfMTZpGdXnZozA14mbkGqqECjZ395fUmzmCVABdG66zop+	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	309	GdXnZozA14mbkGqqECjZ395fUmzmCVABdG66zop+DlRwKlgk4qXaZYSOiQD1jU0ITHTfj47f4JPjdLHoVSmEsAz9fdETmKtx3FgOmMCkV70puJ/XjeD8MrPeZJmwBI6XxRmAoPE0yxwnytGQYwNIVRxcyumEVGEgkmsYkJwF19eJq3zmuvU3LuLav26iKOMjtEJOkMuukR1dIsaqIkoytEzekVv1pP1Yr1bH/PWklXMHKI/sD5/AAdVlig=</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	310	sha1_base64="8QVocR3pGD0i/QL+G1OhPZDl9fE=">AAAB/nicbVBNS8NAEN3Ur1q/ouLJy2IRPJVEBD0WvXisYD+giWWz3TRLN9mwO6mUUPCvePGgiFd/hzf/jds2B219MPB4b4aZeUEquAbH+bZKK6tr6xvlzcrW9s7unr1/0NIyU5Q1qRRSdQKimeAJawIHwTqpYiQOBGsHw5up3x4xpblM7mGcMj8mg4SHnBIwUs8+8lQkH/KRp/ggAqKUfMTZpGdXnZozA14mbkGqqECjZ395fUmzmCVABdG66zop+	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	311	GdXnZozA14mbkGqqECjZ395fUmzmCVABdG66zop+DlRwKlgk4qXaZYSOiQD1jU0ITHTfj47f4JPjdLHoVSmEsAz9fdETmKtx3FgOmMCkV70puJ/XjeD8MrPeZJmwBI6XxRmAoPE0yxwnytGQYwNIVRxcyumEVGEgkmsYkJwF19eJq3zmuvU3LuLav26iKOMjtEJOkMuukR1dIsaqIkoytEzekVv1pP1Yr1bH/PWklXMHKI/sD5/AAdVlig=</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	312	<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	313	sha1_base64="m8MJ1M94ujO0d0COo5n2Dsol6rc=">AAAB53icbVC7SgNBFL0bXzG+opY2g0GwCrs2phAM2FhGMA9IFpmdvZsMmZ1dZmaFsKS0sbFQxNZP8Rfs/AZ/wsmj0MQDFw7nnMt9BKng2rjul1NYWV1b3yhulra2d3b3yvsHLZ1kimGTJSJRnYBqFFxi03AjsJMqpHEgsB0MryZ++x6V5om8NaMU/Zj2JY84o8ZKjbtyxa26U5Bl4s1J5fLj+4EAgM1/9sKEZTFKwwTVuuu5qfFzqgxnAselXqYxp	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	314	4EAgM1/9sKEZTFKwwTVuuu5qfFzqgxnAselXqYxpWxI+9i1VNIYtZ9P9xyTE6uEJEqULWnIVP3dkdNY61Ec2GRMzUAvehPxP6+bmajm51ymmUHJZoOiTBCTkMnRJOQKmREjSyhT3O5K2IAqyox9Tck+wVs8eZm0zqqeW/Vu3Er9AmYowhEcwyl4cA51uIYGNIFBCI/wDC8Od56cV+dtFi04855D+APn/Qd/sI8A</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	315	sha1_base64="q1zM5ZsCNMJAtNUZI97B98ATlaA=">AAAB53icbVC7SgNBFL3rM8ZX1FKQwSBYhV0bLQQDNpYJmAckQWZn7yZDZmeXmVkhLCltbCwUsfUvbP0FO79BP8LJo9DEAxcO55zLffiJ4Nq47qezsLi0vLKaW8uvb2xubRd2dus6ThXDGotFrJo+1Si4xJrhRmAzUUgjX2DD71+O/MYtKs1jeW0GCXYi2pU85IwaK1VuCkW35I5B5ok3JcWL96+7g7fqt81/tIOYpRFKwwTVuuW5ielkVBnOBA7z7	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	316	6+7g7fqt81/tIOYpRFKwwTVuuW5ielkVBnOBA7z7VRjQlmfdrFlqaQR6k423nNIjqwSkDBWtqQhY/V3R0YjrQeRb5MRNT09643E/7xWasKzTsZlkhqUbDIoTAUxMRkdTQKukBkxsIQyxe2uhPWooszY1+TtE7zZk+dJ/aTkuSWv6hbL5zBBDvbhEI7Bg1MowxVUoAYMAriHR3hyuPPgPDsvk+iCM+3Zgz9wXn8AGkeQ8w==</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	317	sha1_base64="q1zM5ZsCNMJAtNUZI97B98ATlaA=">AAAB53icbVC7SgNBFL3rM8ZX1FKQwSBYhV0bLQQDNpYJmAckQWZn7yZDZmeXmVkhLCltbCwUsfUvbP0FO79BP8LJo9DEAxcO55zLffiJ4Nq47qezsLi0vLKaW8uvb2xubRd2dus6ThXDGotFrJo+1Si4xJrhRmAzUUgjX2DD71+O/MYtKs1jeW0GCXYi2pU85IwaK1VuCkW35I5B5ok3JcWL96+7g7fqt81/tIOYpRFKwwTVuuW5ielkVBnOBA7z7	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	318	6+7g7fqt81/tIOYpRFKwwTVuuW5ielkVBnOBA7z7VRjQlmfdrFlqaQR6k423nNIjqwSkDBWtqQhY/V3R0YjrQeRb5MRNT09643E/7xWasKzTsZlkhqUbDIoTAUxMRkdTQKukBkxsIQyxe2uhPWooszY1+TtE7zZk+dJ/aTkuSWv6hbL5zBBDvbhEI7Bg1MowxVUoAYMAriHR3hyuPPgPDsvk+iCM+3Zgz9wXn8AGkeQ8w==</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	319	sha1_base64="ioxb3woZF1oAlTScqds23PrgiMY=">AAAB53icbVC7SgNBFL3rM8ZX1NJmMAhWYdZGC4uAjWUE84BkkdnZ2WTI7Owyc1cIS37AxkIRW3/Jzr9xkmyhiQcGDuecy9x7wkxJi5R+e2vrG5tb25Wd6u7e/sFh7ei4Y9PccNHmqUpNL2RWKKlFGyUq0cuMYEmoRDcc38787pMwVqb6ASeZCBI21DKWnKGTWo+1Om3QOcgq8UtShxIu/zWIUp4nQiNXzNq+TzMMCmZQciWm1UFuRcb4mA1F31HNE	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	320	p4nQiNXzNq+TzMMCmZQciWm1UFuRcb4mA1F31HNEmGDYr7nlJw7JSJxatzTSObq74mCJdZOktAlE4Yju+zNxP+8fo7xdVBIneUoNF98FOeKYEpmR5NIGsFRTRxh3Ei3K+EjZhhHV03VleAvn7xKOpcNnzb8e1pv3pR1VOAUzuACfLiCJtxBC9rAIYJneIU3T3ov3rv3sYiueeXMCfyB9/kDCGmMcA==</latexit>v	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	321	<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	322	sha1_base64="HCiXjOq04H3f4Ed7vqyiRfd+2dI=">AAAB7XicbVA9SwNBEJ2LXzF+nQo2NotBsAp3NlpYBGwsI5hLIDnj3mYvWbO3e+zuRcKR/2BjoYit/8fOf+Pmo9DEBwOP92aYmRelnGnjed9OYWV1bX2juFna2t7Z3XP3DwItM0VonUguVTPCmnImaN0ww2kzVRQnEaeNaHA98RtDqjST4s6MUhomuCdYzAg2VgraaZ/dDztu2at4U6Bl4s9JuXoUPD0AQK3jfrW7kmQJFYZwrHXL91IT5lgZRjgdl	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	323	XoUPD0AQK3jfrW7kmQJFYZwrHXL91IT5lgZRjgdl9qZpikmA9yjLUsFTqgO8+m1Y3RqlS6KpbIlDJqqvydynGg9SiLbmWDT14veRPzPa2UmvgxzJtLMUEFmi+KMIyPR5HXUZYoSw0eWYKKYvRWRPlaYGBtQyYbgL768TILziu9V/FubxhXMUIRjOIEz8OECqnADNagDgUd4hld4c6Tz4rw7H7PWgjOfOYQ/cD5/AKnSkKA=</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	324	sha1_base64="/voSHBGyFE5xYPVKPYM/GTIarLQ=">AAAB7XicbVC7SgNBFL3rM8ZXVLCxWQyCVdi10cIiYGMZwWwCyRpmJ7PJmNmZYWY2siz5BxsLRbT0f+z8AP/DyaPQxAMXDufcy733RJJRbTzvy1laXlldWy9sFDe3tnd2S3v7gRapwqSOBROqGSFNGOWkbqhhpCkVQUnESCMaXI39xpAoTQW/NZkkYYJ6nMYUI2OloC379G7YKZW9ijeBu0j8GSlXD4MHmX2/1zqlz3ZX4DQh3GCGtG75njRhjpShm	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	325	SlXD4MHmX2/1zqlz3ZX4DQh3GCGtG75njRhjpShmJFRsZ1qIhEeoB5pWcpRQnSYT64duSdW6bqxULa4cSfq74kcJVpnSWQ7E2T6et4bi/95rdTEF2FOuUwN4Xi6KE6Za4Q7ft3tUkWwYZklCCtqb3VxHymEjQ2oaEPw519eJMFZxfcq/o1N4xKmKMARHMMp+HAOVbiGGtQBwz08wjO8OMJ5cl6dt2nrkjObOYA/cD5+AGdXkq0=</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	326	sha1_base64="/voSHBGyFE5xYPVKPYM/GTIarLQ=">AAAB7XicbVC7SgNBFL3rM8ZXVLCxWQyCVdi10cIiYGMZwWwCyRpmJ7PJmNmZYWY2siz5BxsLRbT0f+z8AP/DyaPQxAMXDufcy733RJJRbTzvy1laXlldWy9sFDe3tnd2S3v7gRapwqSOBROqGSFNGOWkbqhhpCkVQUnESCMaXI39xpAoTQW/NZkkYYJ6nMYUI2OloC379G7YKZW9ijeBu0j8GSlXD4MHmX2/1zqlz3ZX4DQh3GCGtG75njRhjpShm	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	327	SlXD4MHmX2/1zqlz3ZX4DQh3GCGtG75njRhjpShmJFRsZ1qIhEeoB5pWcpRQnSYT64duSdW6bqxULa4cSfq74kcJVpnSWQ7E2T6et4bi/95rdTEF2FOuUwN4Xi6KE6Za4Q7ft3tUkWwYZklCCtqb3VxHymEjQ2oaEPw519eJMFZxfcq/o1N4xKmKMARHMMp+HAOVbiGGtQBwz08wjO8OMJ5cl6dt2nrkjObOYA/cD5+AGdXkq0=</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	328	sha1_base64="Fc8T4ygtia14k1z/CDji4ezWDqY=">AAAB7XicbVA9SwNBEJ3zM8avqKXNYhCswp2NFhYBG8sI5gOSM+xtNsmavd1jdy4QjvwHGwtFbP0/dv4bN8kVmvhg4PHeDDPzokQKi77/7a2tb2xubRd2irt7+weHpaPjhtWpYbzOtNSmFVHLpVC8jgIlbyWG0ziSvBmNbmd+c8yNFVo94CThYUwHSvQFo+ikRicZisdxt1T2K/4cZJUEOSlDjlq39NXpaZbGXCGT1Np24CcYZtSgYJJPi53U8oSyE	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	329	lq39NXpaZbGXCGT1Np24CcYZtSgYJJPi53U8oSyER3wtqOKxtyG2fzaKTl3So/0tXGlkMzV3xMZja2dxJHrjCkO7bI3E//z2in2r8NMqCRFrthiUT+VBDWZvU56wnCGcuIIZUa4WwkbUkMZuoCKLoRg+eVV0risBH4luPfL1Zs8jgKcwhlcQABXUIU7qEEdGDzBM7zCm6e9F+/d+1i0rnn5zAn8gff5A59Hjx0=</latexit>⇢e!v	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	330	<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	331	sha1_base64="s3/Cw/iD/Ic9TAit26LWmPV1hK0=">AAAB/nicbVBNS8NAEN3Ur1q/ouLJy2IRPJVEBD0WvXisYD+giWWznTRLN9mwu6mUUPCvePGgiFd/hzf/jds2B219MPB4b4aZeUHKmdKO822VVlbX1jfKm5Wt7Z3dPXv/oKVEJik0qeBCdgKigLMEmpppDp1UAokDDu1geDP12yOQionkXo9T8GMySFjIKNFG6tlHnozEQw6eZINIEynFIx5NenbVqTkz4GXiFqSKCjR69pfXFzSLIdGUE6W6rpNqP	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	332	nbVqTkz4GXiFqSKCjR69pfXFzSLIdGUE6W6rpNqPydSM8phUvEyBSmhQzKArqEJiUH5+ez8CT41Sh+HQppKNJ6pvydyEis1jgPTGRMdqUVvKv7ndTMdXvk5S9JMQ0Lni8KMYy3wNAvcZxKo5mNDCJXM3IppRCSh2iRWMSG4iy8vk9Z5zXVq7t1FtX5dxFFGx+gEnSEXXaI6ukUN1EQU5egZvaI368l6sd6tj3lrySpmDtEfWJ8/7hmWGA==</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	333	sha1_base64="s3/Cw/iD/Ic9TAit26LWmPV1hK0=">AAAB/nicbVBNS8NAEN3Ur1q/ouLJy2IRPJVEBD0WvXisYD+giWWznTRLN9mwu6mUUPCvePGgiFd/hzf/jds2B219MPB4b4aZeUHKmdKO822VVlbX1jfKm5Wt7Z3dPXv/oKVEJik0qeBCdgKigLMEmpppDp1UAokDDu1geDP12yOQionkXo9T8GMySFjIKNFG6tlHnozEQw6eZINIEynFIx5NenbVqTkz4GXiFqSKCjR69pfXFzSLIdGUE6W6rpNqP	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	334	nbVqTkz4GXiFqSKCjR69pfXFzSLIdGUE6W6rpNqPydSM8phUvEyBSmhQzKArqEJiUH5+ez8CT41Sh+HQppKNJ6pvydyEis1jgPTGRMdqUVvKv7ndTMdXvk5S9JMQ0Lni8KMYy3wNAvcZxKo5mNDCJXM3IppRCSh2iRWMSG4iy8vk9Z5zXVq7t1FtX5dxFFGx+gEnSEXXaI6ukUN1EQU5egZvaI368l6sd6tj3lrySpmDtEfWJ8/7hmWGA==</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	335	sha1_base64="s3/Cw/iD/Ic9TAit26LWmPV1hK0=">AAAB/nicbVBNS8NAEN3Ur1q/ouLJy2IRPJVEBD0WvXisYD+giWWznTRLN9mwu6mUUPCvePGgiFd/hzf/jds2B219MPB4b4aZeUHKmdKO822VVlbX1jfKm5Wt7Z3dPXv/oKVEJik0qeBCdgKigLMEmpppDp1UAokDDu1geDP12yOQionkXo9T8GMySFjIKNFG6tlHnozEQw6eZINIEynFIx5NenbVqTkz4GXiFqSKCjR69pfXFzSLIdGUE6W6rpNqP	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	336	nbVqTkz4GXiFqSKCjR69pfXFzSLIdGUE6W6rpNqPydSM8phUvEyBSmhQzKArqEJiUH5+ez8CT41Sh+HQppKNJ6pvydyEis1jgPTGRMdqUVvKv7ndTMdXvk5S9JMQ0Lni8KMYy3wNAvcZxKo5mNDCJXM3IppRCSh2iRWMSG4iy8vk9Z5zXVq7t1FtX5dxFFGx+gEnSEXXaI6ukUN1EQU5egZvaI368l6sd6tj3lrySpmDtEfWJ8/7hmWGA==</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	337	sha1_base64="s3/Cw/iD/Ic9TAit26LWmPV1hK0=">AAAB/nicbVBNS8NAEN3Ur1q/ouLJy2IRPJVEBD0WvXisYD+giWWznTRLN9mwu6mUUPCvePGgiFd/hzf/jds2B219MPB4b4aZeUHKmdKO822VVlbX1jfKm5Wt7Z3dPXv/oKVEJik0qeBCdgKigLMEmpppDp1UAokDDu1geDP12yOQionkXo9T8GMySFjIKNFG6tlHnozEQw6eZINIEynFIx5NenbVqTkz4GXiFqSKCjR69pfXFzSLIdGUE6W6rpNqP	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	338	nbVqTkz4GXiFqSKCjR69pfXFzSLIdGUE6W6rpNqPydSM8phUvEyBSmhQzKArqEJiUH5+ez8CT41Sh+HQppKNJ6pvydyEis1jgPTGRMdqUVvKv7ndTMdXvk5S9JMQ0Lni8KMYy3wNAvcZxKo5mNDCJXM3IppRCSh2iRWMSG4iy8vk9Z5zXVq7t1FtX5dxFFGx+gEnSEXXaI6ukUN1EQU5egZvaI368l6sd6tj3lrySpmDtEfWJ8/7hmWGA==</latexit>e	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	339	<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	340	sha1_base64="gRKFy+QFytmwqWy0cvo5FmmPz8I=">AAAB7XicbVA9SwNBEJ3zM8avqGBjsxgEq3Bno4VFwMYygrkEkjPubeaSNXu3x+6eEo78BxsLRWz9P3b+GzcfhSY+GHi8N8PMvDAVXBvX/XaWlldW19YLG8XNre2d3dLevq9lphjWmRRSNUOqUfAE64Ybgc1UIY1DgY1wcDX2G4+oNJfJrRmmGMS0l/CIM2qs5LfTPr/DTqnsVtwJyCLxZqRcPfSf7gGg1il9tbuSZTEmhgmqdctzUxPkVBnOBI6K7	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	341	fSf7gGg1il9tbuSZTEmhgmqdctzUxPkVBnOBI6K7UxjStmA9rBlaUJj1EE+uXZETqzSJZFUthJDJurviZzGWg/j0HbG1PT1vDcW//NamYkugpwnaWYwYdNFUSaIkWT8OulyhcyIoSWUKW5vJaxPFWXGBlS0IXjzLy8S/6ziuRXvxqZxCVMU4AiO4RQ8OIcqXEMN6sDgAZ7hFd4c6bw4787HtHXJmc0cwB84nz+QDpCP</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	342	sha1_base64="74MJShuZzxGyM2nLY3EY87InuhI=">AAAB7XicbVC7SgNBFJ2NrxhfUcHGZjAIVmHXRguLgI1lBLMJJGuYndxNxszODjOzyrLkH2wsFNHS/7HzA/wPJ49CEw9cOJxzL/feE0rOtHHdL6ewtLyyulZcL21sbm3vlHf3fJ2kikKDJjxRrZBo4ExAwzDDoSUVkDjk0AyHl2O/eQ9Ks0TcmExCEJO+YBGjxFjJ78gBu4VuueJW3QnwIvFmpFI78B9k9v1e75Y/O72EpjEIQznRuu250gQ5UYZRD	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	343	FI78B9k9v1e75Y/O72EpjEIQznRuu250gQ5UYZRDqNSJ9UgCR2SPrQtFSQGHeSTa0f42Co9HCXKljB4ov6eyEmsdRaHtjMmZqDnvbH4n9dOTXQe5EzI1ICg00VRyrFJ8Ph13GMKqOGZJYQqZm/FdEAUocYGVLIhePMvLxL/tOq5Ve/apnGBpiiiQ3SETpCHzlANXaE6aiCK7tAjekYvTuI8Oa/O27S14Mxm9tEfOB8/TZOSnA==</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	344	sha1_base64="74MJShuZzxGyM2nLY3EY87InuhI=">AAAB7XicbVC7SgNBFJ2NrxhfUcHGZjAIVmHXRguLgI1lBLMJJGuYndxNxszODjOzyrLkH2wsFNHS/7HzA/wPJ49CEw9cOJxzL/feE0rOtHHdL6ewtLyyulZcL21sbm3vlHf3fJ2kikKDJjxRrZBo4ExAwzDDoSUVkDjk0AyHl2O/eQ9Ks0TcmExCEJO+YBGjxFjJ78gBu4VuueJW3QnwIvFmpFI78B9k9v1e75Y/O72EpjEIQznRuu250gQ5UYZRD	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	345	FI78B9k9v1e75Y/O72EpjEIQznRuu250gQ5UYZRDqNSJ9UgCR2SPrQtFSQGHeSTa0f42Co9HCXKljB4ov6eyEmsdRaHtjMmZqDnvbH4n9dOTXQe5EzI1ICg00VRyrFJ8Ph13GMKqOGZJYQqZm/FdEAUocYGVLIhePMvLxL/tOq5Ve/apnGBpiiiQ3SETpCHzlANXaE6aiCK7tAjekYvTuI8Oa/O27S14Mxm9tEfOB8/TZOSnA==</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	346	sha1_base64="pLq6KB/1S9uyUeWp/G4byg43mK0=">AAAB7XicbVA9SwNBEJ2LXzF+RS1tFoNgFe5stLAI2FhGMB+QnGFvM5es2ds9dveEEPIfbCwUsfX/2Plv3CRXaOKDgcd7M8zMi1LBjfX9b6+wtr6xuVXcLu3s7u0flA+PmkZlmmGDKaF0O6IGBZfYsNwKbKcaaRIJbEWjm5nfekJtuJL3dpximNCB5DFn1Dqp2U2H/AF75Ypf9ecgqyTISQVy1Hvlr25fsSxBaZmgxnQCP7XhhGrLmcBpqZsZTCkb0	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	347	Hvlr25fsSxBaZmgxnQCP7XhhGrLmcBpqZsZTCkb0QF2HJU0QRNO5tdOyZlT+iRW2pW0ZK7+npjQxJhxErnOhNqhWfZm4n9eJ7PxVTjhMs0sSrZYFGeCWEVmr5M+18isGDtCmebuVsKGVFNmXUAlF0Kw/PIqaV5UA78a3PmV2nUeRxFO4BTOIYBLqMEt1KEBDB7hGV7hzVPei/fufSxaC14+cwx/4H3+AIWDjww=</latexit>E<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	348	sha1_base64="iJ/x8cSgmmYNbMN8WtCvsNrlH/U=">AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0IOHgggeW7Af0Iay2U7atZtN2N0IJfQXePGgiFd/kjf/jds2B219MPB4b4aZeUEiuDau++0U1tY3NreK26Wd3b39g/LhUUvHqWLYZLGIVSegGgWX2DTcCOwkCmkUCGwH49uZ335CpXksH8wkQT+iQ8lDzqixUuOuX664VXcOskq8nFQgR71f/uoNYpZGKA0TVOuu5ybGz6gynAmclnqpxoSyMR1i1	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	349	uoNYpZGKA0TVOuu5ybGz6gynAmclnqpxoSyMR1i11JJI9R+Nj90Ss6sMiBhrGxJQ+bq74mMRlpPosB2RtSM9LI3E//zuqkJr/2MyyQ1KNliUZgKYmIy+5oMuEJmxMQSyhS3txI2oooyY7Mp2RC85ZdXSeui6rlVr3FZqd3kcRThBE7hHDy4ghrcQx2awADhGV7hzXl0Xpx352PRWnDymWP4A+fzB5cfjMM=</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	350	sha1_base64="iJ/x8cSgmmYNbMN8WtCvsNrlH/U=">AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0IOHgggeW7Af0Iay2U7atZtN2N0IJfQXePGgiFd/kjf/jds2B219MPB4b4aZeUEiuDau++0U1tY3NreK26Wd3b39g/LhUUvHqWLYZLGIVSegGgWX2DTcCOwkCmkUCGwH49uZ335CpXksH8wkQT+iQ8lDzqixUuOuX664VXcOskq8nFQgR71f/uoNYpZGKA0TVOuu5ybGz6gynAmclnqpxoSyMR1i1	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	351	uoNYpZGKA0TVOuu5ybGz6gynAmclnqpxoSyMR1i11JJI9R+Nj90Ss6sMiBhrGxJQ+bq74mMRlpPosB2RtSM9LI3E//zuqkJr/2MyyQ1KNliUZgKYmIy+5oMuEJmxMQSyhS3txI2oooyY7Mp2RC85ZdXSeui6rlVr3FZqd3kcRThBE7hHDy4ghrcQx2awADhGV7hzXl0Xpx352PRWnDymWP4A+fzB5cfjMM=</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	352	sha1_base64="iJ/x8cSgmmYNbMN8WtCvsNrlH/U=">AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0IOHgggeW7Af0Iay2U7atZtN2N0IJfQXePGgiFd/kjf/jds2B219MPB4b4aZeUEiuDau++0U1tY3NreK26Wd3b39g/LhUUvHqWLYZLGIVSegGgWX2DTcCOwkCmkUCGwH49uZ335CpXksH8wkQT+iQ8lDzqixUuOuX664VXcOskq8nFQgR71f/uoNYpZGKA0TVOuu5ybGz6gynAmclnqpxoSyMR1i1	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	353	uoNYpZGKA0TVOuu5ybGz6gynAmclnqpxoSyMR1i11JJI9R+Nj90Ss6sMiBhrGxJQ+bq74mMRlpPosB2RtSM9LI3E//zuqkJr/2MyyQ1KNliUZgKYmIy+5oMuEJmxMQSyhS3txI2oooyY7Mp2RC85ZdXSeui6rlVr3FZqd3kcRThBE7hHDy4ghrcQx2awADhGV7hzXl0Xpx352PRWnDymWP4A+fzB5cfjMM=</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	354	sha1_base64="iJ/x8cSgmmYNbMN8WtCvsNrlH/U=">AAAB6HicbVBNS8NAEJ3Ur1q/qh69LBbBU0lE0IOHgggeW7Af0Iay2U7atZtN2N0IJfQXePGgiFd/kjf/jds2B219MPB4b4aZeUEiuDau++0U1tY3NreK26Wd3b39g/LhUUvHqWLYZLGIVSegGgWX2DTcCOwkCmkUCGwH49uZ335CpXksH8wkQT+iQ8lDzqixUuOuX664VXcOskq8nFQgR71f/uoNYpZGKA0TVOuu5ybGz6gynAmclnqpxoSyMR1i1	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	355	uoNYpZGKA0TVOuu5ybGz6gynAmclnqpxoSyMR1i11JJI9R+Nj90Ss6sMiBhrGxJQ+bq74mMRlpPosB2RtSM9LI3E//zuqkJr/2MyyQ1KNliUZgKYmIy+5oMuEJmxMQSyhS3txI2oooyY7Mp2RC85ZdXSeui6rlVr3FZqd3kcRThBE7hHDy4ghrcQx2awADhGV7hzXl0Xpx352PRWnDymWP4A+fzB5cfjMM=</latexit>V<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	356	sha1_base64="xc4uzoZiBSxZUZkArgltxczS6nM=">AAAB6HicbVA9SwNBEJ2LXzF+RS1tFoNgFe5EiIVFwMYyAfMByRH2NnPJmr29Y3dPCEd+gY2FIrb+JDv/jZvkCk18MPB4b4aZeUEiuDau++0UNja3tneKu6W9/YPDo/LxSVvHqWLYYrGIVTegGgWX2DLcCOwmCmkUCOwEk7u533lCpXksH8w0QT+iI8lDzqixUrM9KFfcqrsAWSdeTiqQozEof/WHMUsjlIYJqnXPcxPjZ1QZzgTOSv1UY0LZhI6wZ	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	357	/WHMUsjlIYJqnXPcxPjZ1QZzgTOSv1UY0LZhI6wZ6mkEWo/Wxw6IxdWGZIwVrakIQv190RGI62nUWA7I2rGetWbi/95vdSEN37GZZIalGy5KEwFMTGZf02GXCEzYmoJZYrbWwkbU0WZsdmUbAje6svrpH1V9dyq17yu1G/zOIpwBudwCR7UoA730IAWMEB4hld4cx6dF+fd+Vi2Fpx85hT+wPn8AbDjjNQ=</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	358	sha1_base64="xc4uzoZiBSxZUZkArgltxczS6nM=">AAAB6HicbVA9SwNBEJ2LXzF+RS1tFoNgFe5EiIVFwMYyAfMByRH2NnPJmr29Y3dPCEd+gY2FIrb+JDv/jZvkCk18MPB4b4aZeUEiuDau++0UNja3tneKu6W9/YPDo/LxSVvHqWLYYrGIVTegGgWX2DLcCOwmCmkUCOwEk7u533lCpXksH8w0QT+iI8lDzqixUrM9KFfcqrsAWSdeTiqQozEof/WHMUsjlIYJqnXPcxPjZ1QZzgTOSv1UY0LZhI6wZ	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	359	/WHMUsjlIYJqnXPcxPjZ1QZzgTOSv1UY0LZhI6wZ6mkEWo/Wxw6IxdWGZIwVrakIQv190RGI62nUWA7I2rGetWbi/95vdSEN37GZZIalGy5KEwFMTGZf02GXCEzYmoJZYrbWwkbU0WZsdmUbAje6svrpH1V9dyq17yu1G/zOIpwBudwCR7UoA730IAWMEB4hld4cx6dF+fd+Vi2Fpx85hT+wPn8AbDjjNQ=</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	360	sha1_base64="xc4uzoZiBSxZUZkArgltxczS6nM=">AAAB6HicbVA9SwNBEJ2LXzF+RS1tFoNgFe5EiIVFwMYyAfMByRH2NnPJmr29Y3dPCEd+gY2FIrb+JDv/jZvkCk18MPB4b4aZeUEiuDau++0UNja3tneKu6W9/YPDo/LxSVvHqWLYYrGIVTegGgWX2DLcCOwmCmkUCOwEk7u533lCpXksH8w0QT+iI8lDzqixUrM9KFfcqrsAWSdeTiqQozEof/WHMUsjlIYJqnXPcxPjZ1QZzgTOSv1UY0LZhI6wZ	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	361	/WHMUsjlIYJqnXPcxPjZ1QZzgTOSv1UY0LZhI6wZ6mkEWo/Wxw6IxdWGZIwVrakIQv190RGI62nUWA7I2rGetWbi/95vdSEN37GZZIalGy5KEwFMTGZf02GXCEzYmoJZYrbWwkbU0WZsdmUbAje6svrpH1V9dyq17yu1G/zOIpwBudwCR7UoA730IAWMEB4hld4cx6dF+fd+Vi2Fpx85hT+wPn8AbDjjNQ=</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	362	sha1_base64="xc4uzoZiBSxZUZkArgltxczS6nM=">AAAB6HicbVA9SwNBEJ2LXzF+RS1tFoNgFe5EiIVFwMYyAfMByRH2NnPJmr29Y3dPCEd+gY2FIrb+JDv/jZvkCk18MPB4b4aZeUEiuDau++0UNja3tneKu6W9/YPDo/LxSVvHqWLYYrGIVTegGgWX2DLcCOwmCmkUCOwEk7u533lCpXksH8w0QT+iI8lDzqixUrM9KFfcqrsAWSdeTiqQozEof/WHMUsjlIYJqnXPcxPjZ1QZzgTOSv1UY0LZhI6wZ	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	363	/WHMUsjlIYJqnXPcxPjZ1QZzgTOSv1UY0LZhI6wZ6mkEWo/Wxw6IxdWGZIwVrakIQv190RGI62nUWA7I2rGetWbi/95vdSEN37GZZIalGy5KEwFMTGZf02GXCEzYmoJZYrbWwkbU0WZsdmUbAje6svrpH1V9dyq17yu1G/zOIpwBudwCR7UoA730IAWMEB4hld4cx6dF+fd+Vi2Fpx85hT+wPn8AbDjjNQ=</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	364	(c) Message-passing neural network Edge blockNode blockGlobal blockV0	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	365	<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	366	sha1_base64="gAQ7qdt3IKvK5oBqK3uN1PHYi1k=">AAAB6XicbVA9SwNBEJ2LXzF+RS1tFoNoFe4koIVFwMYyivmA5Ah7m7lkyd7esbsnhCP/wMZCEVv/kZ3/xk1yhSY+GHi8N8PMvCARXBvX/XYKa+sbm1vF7dLO7t7+QfnwqKXjVDFssljEqhNQjYJLbBpuBHYShTQKBLaD8e3Mbz+h0jyWj2aSoB/RoeQhZ9RY6aF13i9X3Ko7B1klXk4qkKPRL3/1BjFLI5SGCap113MT42dUGc4ETku9VGNC2ZgOs	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	367	3/1BjFLI5SGCap113MT42dUGc4ETku9VGNC2ZgOsWuppBFqP5tfOiVnVhmQMFa2pCFz9fdERiOtJ1FgOyNqRnrZm4n/ed3UhNd+xmWSGpRssShMBTExmb1NBlwhM2JiCWWK21sJG1FFmbHhlGwI3vLLq6R1WfXcqndfq9Rv8jiKcAKncAEeXEEd7qABTWAQwjO8wpszdl6cd+dj0Vpw8plj+APn8wcRSI0F</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	[]

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.