doi
stringlengths 10
10
| chunk-id
int64 0
936
| chunk
stringlengths 401
2.02k
| id
stringlengths 12
14
| title
stringlengths 8
162
| summary
stringlengths 228
1.92k
| source
stringlengths 31
31
| authors
stringlengths 7
6.97k
| categories
stringlengths 5
107
| comment
stringlengths 4
398
⌀ | journal_ref
stringlengths 8
194
⌀ | primary_category
stringlengths 5
17
| published
stringlengths 8
8
| updated
stringlengths 8
8
| references
list |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
1512.02325 | 33 | XTT XTT RWW XT RIWRWE XT REVI XT XT TT
Fig. 4: Sensitivity and impact of different object characteristics on VOC2007 test set using [21]. The plot on the left shows the effects of BBox Area per category, and the right plot shows the effect of Aspect Ratio. Key: BBox Area: XS=extra-small; S=small; M=medium; L=large; XL =extra-large. Aspect Ratio: XT=extra-tall/narrow; T=tall; M=medium; W=wide; XW =extra-wide.
9
10 Liu et al.
More default box shapes is better. As described in Sec. 2.2, by default we use 6 default boxes per location. If we remove the boxes with 1 3 and 3 aspect ratios, the performance drops by 0.6%. By further removing the boxes with 1 2 and 2 aspect ratios, the performance drops another 2.1%. Using a variety of default box shapes seems to make the task of predicting boxes easier for the network. | 1512.02325#33 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 33 | # 3.5 Frequency Convolutions
Temporal convolution is commonly used in speech recognition to efï¬ciently model temporal trans- lation invariance for variable length utterances. This type of convolution was ï¬rst proposed for neural networks in speech more than 25 years ago [67]. Many neural network speech models have a ï¬rst layer that processes input frames with some context window [16, 66]. This can be viewed as a temporal convolution with a stride of one.
Additionally, sub-sampling is essential to make recurrent neural networks computationally tractable with high sample-rate audio. The DS1 system accomplished this through the use of a spectrogram as input and temporal convolution in the ï¬rst layer with a stride parameter to reduce the number of time-steps [26].
Convolutions in frequency and time domains, when applied to the spectral input features prior to any other processing, can slightly improve ASR performance [1, 53, 60]. Convolution in frequency attempts to model spectral variance due to speaker variability more concisely than what is pos- sible with large fully connected networks. Since spectral ordering of features is removed by fully- connected and recurrent layers, frequency convolutions work better as the ï¬rst layers of the network. | 1512.02595#33 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 34 | Atrous is faster. As described in Sec. 3, we used the atrous version of a subsampled VGG16, following DeepLab-LargeFOV [17]. If we use the full VGG16, keeping pool5 s2 and not subsampling parameters from fc6 and fc7, and add conv5 3 for with 2 prediction, the result is about the same while the speed is about 20% slower.
Prediction source layers from: mAP use boundary boxes? |# Boxes conv4_3 conv7 conv8_2 conv9_2 conv10_2 conv11_2 Yes No Vv Vv Vv Vv Vv Vv 74.3 63.4 8732 Vv Vv Vv Vv Vv 74.6 63.1 8764 Vv Vv Vv Vv 73.8 68.4 8942 Vv Vv Vv 70.7 69.2 9864 Vv Vv 64.2 64.4 9025 Vv 62.4 64.0 8664
Table 3: Effects of using multiple output layers. | 1512.02325#34 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 34 | We experiment with adding between one and three layers of convolution. These are both in the time- and-frequency domain (2D invariance) and in the time-only domain (1D invariance). In all cases we use a same convolution, preserving the number of input features in both frequency and time. In some cases, we specify a stride across either dimension which reduces the size of the output. We do not explicitly control for the number of parameters, since convolutional layers add a small fraction of parameters to our networks. All networks shown in Table 4 have about 35 million parameters.
We report results on two datasetsâa development set of 2048 utterances (âRegular Devâ) and a much noisier dataset of 2048 utterances (âNoisy Devâ) randomly sampled from the CHiME 2015 development datasets [4]. We ï¬nd that multiple layers of 1D-invariant convolutions provides a very small beneï¬t. The 2D-invariant convolutions improve results substantially on noisy data, while providing a small beneï¬t on clean data. The change from one layer of 1D-invariant convolution to three layers of 2D-invariant convolution improves WER by 23.9% on the noisy development set.
9 | 1512.02595#34 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 35 | Multiple output layers at different resolutions is better. A major contribution of SSD is using default boxes of different scales on different output layers. To measure the advantage gained, we progressively remove layers and compare results. For a fair comparison, every time we remove a layer, we adjust the default box tiling to keep the total number of boxes similar to the original (8732). This is done by stacking more scales of boxes on remaining layers and adjusting scales of boxes if needed. We do not exhaustively optimize the tiling for each setting. Table 3 shows a decrease in accuracy with fewer layers, dropping monotonically from 74.3 to 62.4. When we stack boxes of multiple scales on a layer, many are on the image boundary and need to be handled carefully. We tried the strategy used in Faster R-CNN [2], ignoring boxes which are on the boundary. We observe some interesting trends. For example, it hurts the perfor- mance by a large margin if we use very coarse feature maps (e.g. conv11 2 (1 1) or conv10 2 (3 3)). The reason might be that we do not have enough large boxes to cover large objects after the pruning. When we use primarily ï¬ner resolution maps, the | 1512.02325#35 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 35 | 9
Dev no LM Dev LM Stride Unigrams Bigrams Unigrams Bigrams 2 3 4 14.93 15.01 18.86 14.56 15.60 14.84 9.52 9.65 11.92 9.66 10.06 9.93
Table 5: Comparison of WER with different amounts of striding for unigram and bigram outputs on a model with 1 layer of 1D-invariant convolution, 7 recurrent layers, and 1 fully connected layer. All models have BatchNorm, SortaGrad, and 35 million parameters. The models are compared on a development set with and without the use of a 5-gram language model.
# 3.6 Striding
In the convolutional layers, we apply a longer stride and wider context to speed up training as fewer time-steps are required to model a given utterance. Downsampling the input sound (through FFT and convolutional striding) reduces the number of time-steps and computation required in the following layers, but at the expense of reduced performance. | 1512.02595#35 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 36 | 2 (3 3)). The reason might be that we do not have enough large boxes to cover large objects after the pruning. When we use primarily ï¬ner resolution maps, the performance starts increasing again because even after pruning a sufï¬cient number of large boxes remains. If we only use conv7 for prediction, the performance is the worst, reinforcing the message that it is critical to spread boxes of different scales over dif- ferent layers. Besides, since our predictions do not rely on ROI pooling as in [6], we do not have the collapsing bins problem in low-resolution feature maps [23]. The SSD architecture combines predictions from feature maps of various resolutions to achieve comparable accuracy to Faster R-CNN, while using lower resolution input images. | 1512.02325#36 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 36 | In our Mandarin models, we employ striding in the straightforward way. However, in English, striding can reduce accuracy simply because the output of our network requires at least one time- step per output character, and the number of characters in English speech per time-step is high enough to cause problems when striding2. To overcome this, we can enrich the English alphabet with symbols representing alternate labellings like whole words, syllables or non-overlapping n- grams. In practice, we use non-overlapping bi-graphemes or bigrams, since these are simple to construct, unlike syllables, and there are few of them compared to alternatives such as whole words. We transform unigram labels into bigram labels through a simple isomorphism.
Non-overlapping bigrams shorten the length of the output transcription and thus allow for a decrease in the length of the unrolled RNN. The sentence the cat sat with non-overlapping bigrams is seg- mented as [th, e, space, ca, t, space, sa, t]. Notice that for words with odd number of characters, the last character becomes an unigram and space is treated as an unigram as well. This isomorphism ensures that the same words are always composed of the same bigram and unigram tokens. The output set of bigrams consists of all bigrams that occur in the training set. | 1512.02595#36 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 37 | SSD: Single Shot MultiBox Detector
# 3.3 PASCAL VOC2012
We use the same settings as those used for our basic VOC2007 experiments above, except that we use VOC2012 trainval and VOC2007 trainval and test (21503 images) for training, and test on VOC2012 test (10991 images). We train the models with 10â3 learning rate for 60k iterations, then 10â4 for 20k iterations. Table 4 shows the results of our SSD300 and SSD5124 model. We see the same performance trend as we observed on VOC2007 test. Our SSD300 improves accuracy over Fast/Faster R- CNN. By increasing the training and testing image size to 512 512, we are 4.5% more accurate than Faster R-CNN. Compared to YOLO, SSD is signiï¬cantly more accurate, likely due to the use of convolutional default boxes from multiple feature maps and our matching strategy during training. When ï¬ne-tuned from models trained on COCO, our SSD512 achieves 80.0% mAP, which is 4.1% higher than Faster R-CNN. | 1512.02325#37 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 37 | In Table 5 we show results for both the bigram and unigram systems for various levels of striding, with or without a language model. We observe that bigrams allow for larger strides without any sacriï¬ce in in the word error rate. This allows us to reduce the number of time-steps of the unrolled RNN beneï¬ting both computation and memory usage.
# 3.7 Row Convolution and Unidirectional Models
Bidirectional RNN models are challenging to deploy in an online, low-latency setting, because they are built to operate on an entire sample, and so it is not possible to perform the transcription process as the utterance streams from the user. We have found an unidirectional architecture that performs as well as our bidirectional models. This allows us to use unidirectional, forward-only RNN layers in our deployment system. | 1512.02595#37 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 38 | mAP aero bike Method 68.4 82.3 78.4 70.8 52.3 38.7 77.8 71.6 89.3 44.2 73.0 55.0 87.5 80.5 80.8 72.0 35.1 68.3 65.7 80.4 64.2 Fast[6] 70.4 84.9 79.8 74.3 53.9 49.8 77.5 75.9 88.5 45.6 77.1 55.3 86.9 81.7 80.9 79.6 40.1 72.6 60.9 81.2 61.5 Faster[2] Faster[2] 07++12+COCO 75.9 87.4 83.6 76.8 62.9 59.6 81.9 82.0 91.3 54.9 82.6 59.0 89.0 85.5 84.7 84.1 52.2 78.9 65.5 85.4 70.2 57.9 77.0 67.2 57.7 38.3 22.7 68.3 55.9 81.4 36.2 60.8 48.5 77.2 72.3 71.3 63.5 28.9 52.2 54.8 73.9 50.8 YOLO[5] 72.4 85.6 80.1 70.5 | 1512.02325#38 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 38 | To accomplish this, we employ a special layer that we call row convolution, shown in Figure 3. The intuition behind this layer is that we only need a small portion of future information to make an accurate prediction at the current time-step. Suppose at time-step t, we use a future context of Ï steps. We now have a feature matrix ht:t+Ï = [ht, ht+1, ..., ht+Ï ] of size d (Ï + 1). We deï¬ne a parameter matrix W of the same size as ht:t+Ï . The activations rt for the new layer at time-step t are
2Chinese characters are more similar to English syllables than English characters. This is reï¬ected in our training data, where there are on average 14.1 characters/s in English, while only 3.3 characters/s in Mandarin. Conversely, the Shannon entropy per character as calculated from occurrence in the training set, is less in English due to the smaller character setâ4.9 bits/char compared to 12.6 bits/char in Mandarin. This implies that spoken Mandarin has a lower temporal entropy density, â¼41 bits/s compared to â¼58 bits/s, and can thus more easily be temporally compressed without losing character information.
10
Row conv layer Recurrent layer
Figure 3: Row convolution architecture with future context size of 2 | 1512.02595#38 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 39 | 71.3 63.5 28.9 52.2 54.8 73.9 50.8 YOLO[5] 72.4 85.6 80.1 70.5 57.6 46.2 79.4 76.1 89.2 53.0 77.0 60.8 87.0 83.1 82.3 79.4 45.9 75.9 69.5 81.9 67.5 SSD300 07++12+COCO 77.5 90.2 83.3 76.3 63.0 53.6 83.8 82.8 92.0 59.7 82.7 63.5 89.3 87.6 85.9 84.3 52.6 82.5 74.1 88.4 74.2 SSD300 74.9 87.4 82.3 75.8 59.0 52.6 81.7 81.5 90.0 55.4 79.0 59.8 88.4 84.3 84.7 83.3 50.2 78.0 66.3 86.3 72.0 SSD512 07++12+COCO 80.0 90.7 86.8 80.5 67.8 60.8 86.3 85.5 93.5 63.2 85.7 64.4 90.9 89.0 88.9 86.8 57.2 85.1 72.8 88.4 | 1512.02325#39 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 39 | 10
Row conv layer Recurrent layer
Figure 3: Row convolution architecture with future context size of 2
TH1 rei = Wi ghipj-1a, for l <i<d. (11) j=l
Since the convolution-like operation in Eq. 11 is row oriented for both W and ht:t+Ï , we call this layer row convolution.
We place the row convolution layer above all recurrent layers. This has two advantages. First, this allows us to stream all computation below the row convolution layer on a ï¬ner granularity given little future context is needed. Second, this results in better CER compared to the best bidirectional model for Mandarin. We conjecture that the recurrent layers have learned good feature representations, so the row convolution layer simply gathers the appropriate information to feed to the classiï¬er. Results for a unidirectional Mandarin speech system with row convolution and a comparison to a bidirectional model are given in Section 7 on deployment.
# 3.8 Language Model | 1512.02595#39 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 40 | # 3.8 Language Model
We train our RNN Models over millions of unique utterances, which enables the network to learn a powerful implicit language model. Our best models are quite adept at spelling, without any external language constraints. Further, in our development datasets we ï¬nd many cases where our models can implicitly disambiguate homophonesâfor example, âhe expects the Japanese agent to sell it for two hundred seventy ï¬ve thousand dollarsâ. Nevertheless, the labeled training data is small compared to the size of unlabeled text corpora that are available. Thus we ï¬nd that WER improves when we supplement our system with a language model trained from external text. | 1512.02595#40 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 41 | # 3.4 COCO
To further validate the SSD framework, we trained our SSD300 and SSD512 architec- tures on the COCO dataset. Since objects in COCO tend to be smaller than PASCAL VOC, we use smaller default boxes for all layers. We follow the strategy mentioned in Sec. 2.2, but now our smallest default box has a scale of 0.15 instead of 0.2, and the scale of the default box on conv4 3 is 0.07 (e.g. 21 pixels for a 300
à | 1512.02325#41 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 41 | We use an n-gram language model since they scale well to large amounts of unlabeled text [26]. For English, our language model is a Kneser-Ney smoothed 5-gram model with pruning that is trained using the KenLM toolkit [28] on cleaned text from the Common Crawl Repository3. The vocabulary is the most frequently used 400,000 words from 250 million lines of text, which produces a language model with about 850 million n-grams. For Mandarin, the language model is a Kneser- Ney smoothed character level 5-gram model with pruning that is trained on an internal text corpus of 8 billion lines of text. This produces a language model with about 2 billion n-grams.
During inference we search for the transcription y that maximizes Q(y) shown in Equation 12. This is a linear combination of log probabilities from the CTC trained network and language model, along with a word insertion term [26].
Q(y) = log(pctc(y (12)
x)) + α log(plm(y)) + β word_count(y) |
The weight α controls the relative contributions of the language model and the CTC network. The weight β encourages more words in the transcription. These parameters are tuned on a development set. We use a beam search to ï¬nd the optimal transcription [27].
# 3http://commoncrawl.org | 1512.02595#41 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 42 | Ã
We use the trainval35k [24] for training. We ï¬rst train the model with 10â3 learning rate for 160k iterations, and then continue training for 40k iterations with 10â4 and 40k iterations with 10â5. Table 5 shows the results on test-dev2015. Similar to what we observed on the PASCAL VOC dataset, SSD300 is better than Fast R-CNN in both [email protected] and mAP@[0.5:0.95]. SSD300 has a similar [email protected] as ION [24] and Faster R-CNN [25], but is worse in [email protected]. By increasing the im- age size to 512 512, our SSD512 is better than Faster R-CNN [25] in both criteria. Interestingly, we observe that SSD512 is 5.3% better in [email protected], but is only 1.2% better in [email protected]. We also observe that it has much better AP (4.8%) and AR (4.6%) for large objects, but has relatively less improvement in AP (1.3%) and AR (2.0%) for | 1512.02325#42 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 42 | # 3http://commoncrawl.org
11
Language Architecture Dev no LM Dev LM English English Mandarin Mandarin 5-layer, 1 RNN 9-layer, 7 RNN 5-layer, 1 RNN 9-layer, 7 RNN 27.79 14.93 9.80 7.55 14.39 9.52 7.13 5.81
Table 6: Comparison of WER for English and CER for Mandarin with and without a language model. These are simple RNN models with only one layer of 1D invariant convolution.
Table 6 shows that an external language model helps both English and Mandarin speech systems. The relative improvement given by the language model drops from 48% to 36% in English and 27% to 23% in Mandarin, as we go from a model with 5 layers and 1 recurrent layer to a model with 9 layers and 7 recurrent layers. We hypothesize that the network builds a stronger implicit language model with more recurrent layers.
The relative performance improvement from a language model is higher in English than in Mandarin. We attribute this to the fact that a Chinese character represents a larger block of information than an English character. For example, if we output directly to syllables or words in English, the model would make fewer spelling mistakes and the language model would likely help less.
# 3.9 Adaptation to Mandarin | 1512.02595#42 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 43 | # 3.9 Adaptation to Mandarin
The techniques that we have described so far can be used to build an end-to-end Mandarin speech recognition system that outputs Chinese characters directly. This precludes the need to construct a pronunciation model, which is often a fairly involved component for porting speech systems to other languages [59]. Direct output to characters also precludes the need to explicitly model language speciï¬c pronunciation features. For example we do not need to model Mandarin tones explicitly, as some speech systems must do [59, 45].
The only architectural changes we make to our networks are due to the characteristics of the Chinese character set. Firstly, the output layer of the network outputs about 6000 characters, which includes the Roman alphabet, since hybrid Chinese-English transcripts are common. We incur an out of vocabulary error at evaluation time if a character is not contained in this set. This is not a major concern, as our test set has only 0.74% out of vocab characters. | 1512.02595#43 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 44 | Method Fast [6] Fast [24] Faster [2] ION [24] Faster [25] SSD300 SSD512 data train train trainval train trainval trainval35k trainval35k Avg. Precision, IoU: 0.5:0.95 0.5 0.75 Avg. Precision, Area: Avg. Recall, #Dets: Avg. Recall, Area: S M L 10 - - - - 1 35.9 - 39.9 19.4 4.1 20.0 35.8 21.3 29.5 30.1 7.3 32.1 52.0 42.7 - 43.2 23.6 6.4 24.1 38.3 23.2 32.7 33.5 10.1 37.7 53.6 45.3 23.5 7.7 26.4 37.1 23.8 34.0 34.6 12.0 38.5 54.4 41.2 23.4 5.3 23.2 39.6 22.5 33.2 35.3 9.6 37.6 56.5 46.5 27.8 9.0 28.9 41.9 24.8 37.5 39.8 14.0 43.5 59.0 S M L - - - 100 - 19.7 20.5 21.9 23.6 24.2 23.2 | 1512.02325#44 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 44 | We use a character level language model in Mandarin as words are not usually segmented in text. The word insertion term of Equation 12 becomes a character insertion term. In addition, we ï¬nd that the performance of the beam search during decoding levels off at a smaller beam size. This allows us to use a beam size of 200 with a negligible degradation in CER. In Section 6.2, we show that our Mandarin speech models show roughly the same improvements to architectural changes as our English speech models.
# 4 System Optimizations
Our networks have tens of millions of parameters, and the training algorithm takes tens of single- precision exaFLOPs to converge. Since our ability to evaluate hypotheses about our data and mod- els depends on the ability to train models quickly, we built a highly optimized training system. This system has two main componentsâa deep learning library written in C++, along with a high- performance linear algebra library written in both CUDA and C++. Our optimized software, running on dense compute nodes with 8 Titan X GPUs per node, allows us to sustain 24 single-precision teraFLOP/second when training a single model on one node. This is 45% of the theoretical peak computational throughput of each node. We also can scale to multiple nodes, as outlined in the next subsection.
# 4.1 Scalability and Data-Parallelism | 1512.02595#44 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 45 | 39.8 14.0 43.5 59.0 S M L - - - 100 - 19.7 20.5 21.9 23.6 24.2 23.2 26.8 - - - - - - - - - Table 5: COCO test-dev2015 detection results. small objects. Compared to ION, the improvement in AR for large and small objects is more similar (5.4% vs. 3.9%). We conjecture that Faster R-CNN is more competitive on smaller objects with SSD because it performs two box reï¬nement steps, in both the RPN part and in the Fast R-CNN part. In Fig. 5, we show some detection examples on COCO test-dev with the SSD512 model. | 1512.02325#45 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 45 | # 4.1 Scalability and Data-Parallelism
We use the standard technique of data-parallelism to train on multiple GPUs using synchronous SGD. Our most common conï¬guration uses a minibatch of 512 on 8 GPUs. Our training pipeline
12
binds one process to each GPU. These processes then exchange gradient matrices during the back- propagation by using all-reduce, which exchanges a matrix between multiple processes and sums the result so that at the end, each process has a copy of the sum of all matrices from all processes.
We ï¬nd synchronous SGD useful because it is reproducible and deterministic. We have found that the appearance of non-determinism in our system often signals a serious bug, and so having reproducibility as a goal has greatly facilitates debugging. In contrast, asynchronous methods such as asynchronous SGD with parameter servers as found in Dean et al. [18] typically do not provide reproducibility and are therefore more difï¬cult to debug. Synchronous SGD is simple to understand and implement. It scales well as we add multiple nodes to the training process.
218 â 5-3 (2560) â 9-7 (1760) Time (seconds) aa 90 2 2 Py Py Py 38 GPUs i | 1512.02595#45 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 46 | # 3.5 Preliminary ILSVRC results
We applied the same network architecture we used for COCO to the ILSVRC DET dataset [16]. We train a SSD300 model using the ILSVRC2014 DET train and val1 as used in [22]. We ï¬rst train the model with 10â3 learning rate for 320k iterations, and then continue training for 80k iterations with 10â4 and 40k iterations with 10â5. We can achieve 43.4 mAP on the val2 set [22]. Again, it validates that SSD is a general framework for high quality real-time detection.
# 3.6 Data Augmentation for Small Object Accuracy | 1512.02325#46 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 46 | 218 â 5-3 (2560) â 9-7 (1760) Time (seconds) aa 90 2 2 Py Py Py 38 GPUs i
Figure 4: Scaling comparison of two networksâa 5 layer model with 3 recurrent layers containing 2560 hidden units in each layer and a 9 layer model with 7 recurrent layers containing 1760 hidden units in each layer. The times shown are to train 1 epoch. The 5 layer model trains faster because it uses larger matrices and is more computationally efï¬cient.
Figure 4 shows that time taken to train one epoch halves as we double the number of GPUs that we train on, thus achieving near-linear weak scaling. We keep the minibatch per GPU constant at 64 during this experiment, effectively doubling the minibatch as we double the number of GPUs. Although we have the ability to scale to large minibatches, we typically use either 8 or 16 GPUs during training with a minibatch of 512 or 1024, in order to converge to the best result. | 1512.02595#46 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 47 | # 3.6 Data Augmentation for Small Object Accuracy
Without a follow-up feature resampling step as in Faster R-CNN, the classiï¬cation task for small objects is relatively hard for SSD, as demonstrated in our analysis (see Fig. 4). The data augmentation strategy described in Sec. 2.2 helps to improve the performance dramatically, especially on small datasets such as PASCAL VOC. The random crops generated by the strategy can be thought of as a âzoom inâ operation and can generate many larger training examples. To implement a âzoom outâ operation that creates more of the small training examples, we ï¬rst randomly place an image on a canvas of 16 original image size ï¬lled with mean values before we do any random crop operation. Because we have more training images by introducing this new âexpansionâ data aug- mentation trick, we have to double the training iterations. We have seen a consistent increase of 2%-3% mAP across multiple datasets, as shown in Table 6. In speciï¬c, Fig- ure 6 shows that the new augmentation trick signiï¬cantly improves the performance on small objects. This result underscores the importance of the data augmentation strategy for the ï¬nal model accuracy. | 1512.02325#47 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 47 | Since all-reduce is critical to the scalability of our training, we wrote our own implementation of the ring algorithm [48, 65] for higher performance and better stability. Our implementation avoids extraneous copies between CPU and GPU, and is fundamental to our scalability. We conï¬gure OpenMPI with the smcuda transport that can send and receive buffers residing in the memory of two different GPUs by using GPUDirect. When two GPUs are in the same PCI root complex, this avoids any unnecessary copies to CPU memory. This also takes advantage of tree-structured interconnects by running multiple segments of the ring concurrently between neighboring devices. We built our implementation using MPI send and receive, along with CUDA kernels for the element- wise operations. | 1512.02595#47 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 48 | Table 7 compares the performance of our all-reduce implementation with that provided by OpenMPI version 1.8.5. We report the time spent in all-reduce for a full training run that ran for one epoch on our English dataset using a 5 layer, 3 recurrent layer architecture with 2560 hidden units for all layers. In this table, we use a minibatch of 64 per GPU, expanding the algorithmic minibatch as we scale to more GPUs. We see that our implementation is considerably faster than OpenMPIâs when the communication is within a node (8 GPUs or less). As we increase the number of GPUs and increase the amount of inter-node communication, the gap shrinks, although our implementation is still 2-4X faster.
All of our training runs use either 8 or 16 GPUs, and in this regime, our all-reduce implementation results in 2.5 faster training for the full training run, compared to using OpenMPI directly. Opti- mizing all-reduce has thus resulted in important productivity beneï¬ts for our experiments, and has made our simple synchronous SGD approach scalable.
13 | 1512.02595#48 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 49 | Fig. 5: Detection examples on COCO test-dev with SSD512 model. We show detections with scores higher than 0.6. Each color corresponds to an object category.
Method VOC2007 test VOC2012 test 07+12 07+12+COCO 07++12 07++12+COCO COCO test-dev2015 trainval35k SSD300 SSD512 SSD300* SSD512* 0.5 74.3 76.8 77.2 79.8 0.5 79.6 81.6 81.2 83.2 0.5 72.4 74.9 75.8 78.5 0.5 77.5 80.0 79.3 82.2 0.5:0.95 23.2 26.8 25.1 28.8 0.5 41.2 46.5 43.1 48.5 0.75 23.4 27.8 25.8 30.3
13
14 Liu et al. | 1512.02325#49 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 49 | 13
GPU OpenMPI all-reduce Our all-reduce Performance Gain 4 8 16 32 64 128 55359.1 48881.6 21562.6 8191.8 1395.2 1602.1 2587.4 2470.9 1393.7 1339.6 611.0 422.6 21.4 19.8 15.5 6.1 2.3 3.8
Table 7: Comparison of two different all-reduce implementations. All times are in seconds. Performance gain is the ratio of OpenMPI all-reduce time to our all-reduce time.
Language Architecture CPU CTC Time GPU CTC Time Speedup English Mandarin 5-layer, 3 RNN 5-layer, 3 RNN 5888.12 1688.01 203.56 135.05 28.9 12.5
Table 8: Comparison of time spent in seconds in computing the CTC loss function and gradient in one epoch for two different implementations. Speedup is the ratio of CPU CTC time to GPU CTC time.
# 4.2 GPU implementation of CTC loss function | 1512.02595#49 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 50 | 13
14 Liu et al.
â880300: BBox Are â880512: B80) SSW SNA SSW KEM SWIM TSW SEN " âSSHDN MOSWLAL SSHLE SSEMIBL DSSWLDL XSSWLN XSSHTM en ' or ye ba mary Bie Tal gt Beat os age _ fe os 0 XSSMUXL XSSMUXL XSSMLXL XSSMLXL XSSMLXL XSSMLXL XSSMLXL XSSMUXL XSSMLXL XSSMUXL XSSMUXL XSSMUXL XSSMUXL XSSMUXL
XSSMUXL XSSMUXL XSSMLXL XSSMLXL XSSMLXL XSSMLXL XSSMLXL
XSSMUXL XSSMLXL XSSMUXL XSSMUXL XSSMUXL XSSMUXL XSSMUXL | 1512.02325#50 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 50 | # 4.2 GPU implementation of CTC loss function
Calculating the CTC loss function is more complicated than performing forward and back prop- agation on our RNN architectures. Originally, we transferred activations from the GPUs to the CPU, where we calculated the loss function using an OpenMP parallelized implementation of CTC. However, this implementation limited our scalability rather signiï¬cantly, for two reasons. Firstly, it became computationally more signiï¬cant as we improved efï¬ciency and scalability of the RNN itself. Secondly, transferring large activation matrices between CPU and GPU required us to spend interconnect bandwidth for CTC, rather than on transferring gradient matrices to allow us to scale using data parallelism to more processors.
To overcome this, we wrote a GPU implementation of the CTC loss function. Our parallel imple- mentation relies on a slight refactoring to simplify the dependences in the CTC calculation, as well as the use of optimized parallel sort implementations from ModernGPU [5]. We give more details of this parallelization in the Appendix.
Table 8 compares the performance of two CTC implementations. The GPU implementation saves us 95 minutes per epoch in English, and 25 minutes in Mandarin. This reduces overall training time by 10-20%, which is also an important productivity beneï¬t for our experiments.
# 4.3 Memory allocation | 1512.02595#50 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 51 | XSSMUXL XSSMLXL XSSMUXL XSSMUXL XSSMUXL XSSMUXL XSSMUXL
Fig. 6: Sensitivity and impact of object size with new data augmentation on VOC2007 test set using [21]. The top row shows the effects of BBox Area per cat- egory for the original SSD300 and SSD512 model, and the bottom row corresponds to the SSD300* and SSD512* model trained with the new data augmentation trick. It is obvious that the new data augmentation trick helps detecting small objects signiï¬cantly.
# 3.7 Inference time
Considering the large number of boxes generated from our method, it is essential to perform non-maximum suppression (nms) efï¬ciently during inference. By using a con- ï¬dence threshold of 0.01, we can ï¬lter out most boxes. We then apply nms with jaccard overlap of 0.45 per class and keep the top 200 detections per image. This step costs about 1.7 msec per image for SSD300 and 20 VOC classes, which is close to the total time (2.4 msec) spent on all newly added layers. We measure the speed with batch size 8 using Titan X and cuDNN v4 with Intel Xeon [email protected]. | 1512.02325#51 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 51 | # 4.3 Memory allocation
Our system makes frequent use of dynamic memory allocations to GPU and CPU memory, mainly to store activation data for variable length utterances, and for intermediate results. Individual al- locations can be very large; over 1 GB for the longest utterances. For these very large allocations we found that CUDAâs memory allocator and even std::malloc introduced signiï¬cant overhead into our applicationâover a 2x slowdown from using std::malloc in some cases. This is because both cudaMalloc and std::malloc forward very large allocations to the operating system or GPU driver to update the system page tables. This is a good optimization for systems running multiple applications, all sharing memory resources, but editing page tables is pure overhead for our system where nodes are dedicated entirely to running a single model. To get around this limitation, we wrote our own memory allocator for both CPU and GPU allocations. Our implementation follows the approach of the last level shared allocator in jemalloc: all allocations are carved out of contigu- ous memory blocks using the buddy algorithm [34]. To avoid fragmentation, we preallocate all of GPU memory at the start of training and subdivide individual allocations from this block. Simi- larly, we set the CPU memory block size that we forward to mmap to be substantially larger than std::malloc, at 12GB.
14 | 1512.02595#51 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 52 | Table 7 shows the comparison between SSD, Faster R-CNN[2], and YOLO[5]. Both our SSD300 and SSD512 method outperforms Faster R-CNN in both speed and accu- racy. Although Fast YOLO[5] can run at 155 FPS, it has lower accuracy by almost 22% mAP. To the best of our knowledge, SSD300 is the ï¬rst real-time method to achieve above 70% mAP. Note that about 80% of the forward time is spent on the base network (VGG16 in our case). Therefore, using a faster base network could even further improve the speed, which can possibly make the SSD512 model real-time as well.
# 4 Related Work
There are two established classes of methods for object detection in images, one based on sliding windows and the other based on region proposal classiï¬cation. Before the advent of convolutional neural networks, the state of the art for those two approaches â Deformable Part Model (DPM) [26] and Selective Search [1] â had comparable performance. However, after the dramatic improvement brought on by R-CNN [22], which combines selective search region proposals and convolutional network based post-classiï¬cation, region proposal object detection methods became prevalent. | 1512.02325#52 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 52 | 14
Dataset Speech Type Hours WSJ Switchboard Fisher LibriSpeech Baidu Baidu read conversational conversational read read mixed 80 300 2000 960 5000 3600 Total 11940
Table 9: Summary of the datasets used to train DS2 in English. The Wall Street Journal (WSJ), Switchboard and Fisher [13] corpora are all published by the Linguistic Data Consortium. The LibriSpeech dataset [46] is available free on-line. The other datasets are internal Baidu corpora. | 1512.02595#52 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 53 | The original R-CNN approach has been improved in a variety of ways. The ï¬rst set of approaches improve the quality and speed of post-classiï¬cation, since it requires
SSD: Single Shot MultiBox Detector
Method Faster R-CNN (VGG16) Fast YOLO YOLO (VGG16) SSD300 SSD512 SSD300 SSD512 mAP 73.2 52.7 66.4 74.3 76.8 74.3 76.8 FPS 7 155 21 46 19 59 22 batch size 1 1 1 1 1 8 8 # Boxes Input resolution â¼ 6000 â¼ 1000 Ã 600 98 98 8732 24564 8732 24564 448 Ã 448 448 Ã 448 300 Ã 300 512 Ã 512 300 Ã 300 512 Ã 512
Table 7: Results on Pascal VOC2007 test. SSD300 is the only real-time detection method that can achieve above 70% mAP. By using a larger input image, SSD512 out- performs all methods on accuracy while maintaining a close to real-time speed. | 1512.02325#53 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 53 | Most of the memory required for training deep recurrent networks is used to store activations through each layer for use by back propagation, not to store the parameters of the network. For example, storing the weights for a 70M parameter network with 9 layers requires approximately 280 MB of memory, but storing the activations for a batch of 64, seven-second utterances requires 1.5 GB of memory. TitanX GPUs include 12GB of GDDR5 RAM, and sometimes very deep networks can exceed the GPU memory capacity when processing long utterances. This can happen unpredictably, especially when the distribution of utterance lengths includes outliers, and it is desirable to avoid a catastrophic failure when this occurs. When a requested memory allocation exceeds available GPU memory, we allocate page-locked GPU-memory-mapped CPU memory using cudaMallocHost in- stead. This memory can be accessed directly by the GPU by forwarding individual memory trans- actions over PCIe at reduced bandwidth, and it allows a model to continue to make progress even after encountering an outlier.
The combination of fast memory allocation with a fallback mechanism that allows us to slightly overï¬ow available GPU memory in exceptional cases makes the system signiï¬cantly simpler, more robust, and more efï¬cient.
# 5 Training Data | 1512.02595#53 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 54 | the classiï¬cation of thousands of image crops, which is expensive and time-consuming. SPPnet [9] speeds up the original R-CNN approach signiï¬cantly. It introduces a spatial pyramid pooling layer that is more robust to region size and scale and allows the classi- ï¬cation layers to reuse features computed over feature maps generated at several image resolutions. Fast R-CNN [6] extends SPPnet so that it can ï¬ne-tune all layers end-to- end by minimizing a loss for both conï¬dences and bounding box regression, which was ï¬rst introduced in MultiBox [7] for learning objectness. | 1512.02325#54 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 54 | # 5 Training Data
Large-scale deep learning systems require an abundance of labeled training data. We have collected an extensive training dataset for both English and Mandarin speech models, in addition to augment- ing our training with publicly available datasets. In English we use 11,940 hours of labeled speech data containing 8 million utterances summarized in Table 9. For the Mandarin system we use 9,400 hours of labeled audio containing 11 million utterances. The Mandarin speech data consists of in- ternal Baidu corpora, representing a mix of read speech and spontaneous speech, in both standard Mandarin and accented Mandarin.
# 5.1 Dataset Construction
Some of the internal English (3,600 hours) and Mandarin (1,400 hours) datasets were created from raw data captured as long audio clips with noisy transcriptions. The length of these clips ranged from several minutes to more than hour, making it impractical to unroll them in time in the RNN during training. To solve this problem, we developed an alignment, segmentation and ï¬ltering pipeline that can generate a training set with shorter utterances and few erroneous transcriptions.
The ï¬rst step in the pipeline is to use an existing bidirectional RNN model trained with CTC to align the transcription to the frames of audio. For a given audio-transcript pair, (x, y), we ï¬nd the alignment that maximizes | 1512.02595#54 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 55 | The second set of approaches improve the quality of proposal generation using deep neural networks. In the most recent works like MultiBox [7,8], the Selective Search region proposals, which are based on low-level image features, are replaced by pro- posals generated directly from a separate deep neural network. This further improves the detection accuracy but results in a somewhat complex setup, requiring the training of two neural networks with a dependency between them. Faster R-CNN [2] replaces selective search proposals by ones learned from a region proposal network (RPN), and introduces a method to integrate the RPN with Fast R-CNN by alternating between ï¬ne- tuning shared convolutional layers and prediction layers for these two networks. This way region proposals are used to pool mid-level features and the ï¬nal classiï¬cation step is less expensive. Our SSD is very similar to the region proposal network (RPN) in Faster R-CNN in that we also use a ï¬xed set of (default) boxes for prediction, similar to the anchor boxes in the RPN. But instead of using these to pool features and evaluate another classiï¬er, we simultaneously produce a score for each object category in each box. Thus, our approach avoids the complication of merging RPN with Fast R-CNN and is easier to train, faster, and straightforward to integrate in other tasks. | 1512.02325#55 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 55 | T é* = argmax J [Pee (Cel; 8)- (13) ¢eAlign(x,y) ~ |
# t
This is essentially a Viterbi alignment found using a RNN model trained with CTC. Since Equation 9 integrates over the alignment, the CTC loss function is never explicitly asked to produce an accurate alignment. In principle, CTC could choose to emit all the characters of the transcription after some
15
ï¬xed delay and this can happen with unidirectional RNNs [54]. However, we found that CTC produces an accurate alignment when trained with a bidirectional RNN.
Following the alignment is a segmentation step that splices the audio and the corresponding aligned transcription whenever it encounters a long series of consecutive blank labels occurs, since this usually denotes a stretch of silence. By tuning the number of consecutive blanks, we can tune the length of the utterances generated. For the English speech data, we also require a space token to be within the stretch of blanks in order to segment only on word boundaries. We tune the segmentation to generate utterances that are on average 7 seconds long. | 1512.02595#55 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 56 | Another set of methods, which are directly related to our approach, skip the proposal step altogether and predict bounding boxes and conï¬dences for multiple categories di- rectly. OverFeat [4], a deep version of the sliding window method, predicts a bounding box directly from each location of the topmost feature map after knowing the conï¬- dences of the underlying object categories. YOLO [5] uses the whole topmost feature map to predict both conï¬dences for multiple categories and bounding boxes (which are shared for these categories). Our SSD method falls in this category because we do not have the proposal step but use the default boxes. However, our approach is more ï¬exible than the existing methods because we can use default boxes of different aspect
15
16 Liu et al.
ratios on each feature location from multiple feature maps at different scales. If we only use one default box per location from the topmost feature map, our SSD would have similar architecture to OverFeat [4]; if we use the whole topmost feature map and add a fully connected layer for predictions instead of our convolutional predictors, and do not explicitly consider multiple aspect ratios, we can approximately reproduce YOLO [5].
# 5 Conclusions | 1512.02325#56 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 56 | The ï¬nal step in the pipeline removes erroneous examples that arise from a failed alignment. We crowd source the ground truth transcriptions for several thousand examples. The word level edit distance between the ground truth and the aligned transcription is used to produce a good or bad label. The threshold for the word level edit distance is chosen such that the resulting WER of the good portion of the development set is less than 5%. We then train a linear classiï¬er to accurately predict bad examples given the input features generated from the speech recognizer. We ï¬nd the following features useful: the raw CTC cost, the CTC cost normalized by the sequence length, the CTC cost normalized by the transcript length, the ratio of the sequence length to the transcript length, the number of words in the transcription and the number of characters in the transcription. For the English dataset, we ï¬nd that the ï¬ltering pipeline reduces the WER from 17% to 5% while retaining more than 50% of the examples.
# 5.2 Data Augmentation | 1512.02595#56 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 57 | This paper introduces SSD, a fast single-shot object detector for multiple categories. A key feature of our model is the use of multi-scale convolutional bounding box outputs attached to multiple feature maps at the top of the network. This representation allows us to efï¬ciently model the space of possible box shapes. We experimentally validate that given appropriate training strategies, a larger number of carefully chosen default bounding boxes results in improved performance. We build SSD models with at least an order of magnitude more box predictions sampling location, scale, and aspect ratio, than existing methods [5,7]. We demonstrate that given the same VGG-16 base architecture, SSD compares favorably to its state-of-the-art object detector counterparts in terms of both accuracy and speed. Our SSD512 model signiï¬cantly outperforms the state-of-the- art Faster R-CNN [2] in terms of accuracy on PASCAL VOC and COCO, while being faster. Our real time SSD300 model runs at 59 FPS, which is faster than the current 3 à real time YOLO [5] alternative, while producing markedly superior detection accuracy. Apart from its standalone utility, | 1512.02325#57 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 57 | # 5.2 Data Augmentation
We augment our training data by adding noise to increase the effective size of our training data and to improve our robustness to noisy speech [26]. Although the training data contains some intrinsic noise, we can increase the quantity and variety of noise through augmentation. Too much noise augmentation tends to make optimization difï¬cult and can lead to worse results, and too little noise augmentation makes the system less robust to low signal-to-noise speech. We ï¬nd that a good balance is to add noise to 40% of the utterances that are chosen at random. The noise source consists of several thousand hours of randomly selected audio clips combined to produce hundreds of hours of noise.
# 5.3 Scaling Data | 1512.02595#57 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 58 | which is faster than the current 3 Ã real time YOLO [5] alternative, while producing markedly superior detection accuracy. Apart from its standalone utility, we believe that our monolithic and relatively sim- ple SSD model provides a useful building block for larger systems that employ an object detection component. A promising future direction is to explore its use as part of a sys- tem using recurrent neural networks to detect and track objects in video simultaneously. | 1512.02325#58 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 58 | # 5.3 Scaling Data
Our English and Mandarin corpora are substantially larger than those commonly reported in speech recognition literature. In Table 10, we show the effect of increasing the amount of labeled training data on WER. This is done by randomly sampling the full dataset before training. For each dataset, the model was trained for up to 20 epochs though usually early-stopped based on the error on a held out development set. We note that the WER decreases with a power law for both the regular and 40% relative for each factor of 10 increase in noisy development sets. The WER decreases by training set size. We also observe a consistent gap in WER ( 60% relative) between the regular and noisy datasets, implying that more data beneï¬ts both cases equally.
This implies that a speech system will continue to improve with more labeled training data. We hypothesize that equally as important as increasing raw number of hours is increasing the number of speech contexts that are captured in the dataset. A context can be any property that makes speech unique including different speakers, background noise, environment, and microphone hardware. While we do not have the labels needed to validate this claim, we suspect that measuring WER as a function of speakers in the dataset would lead to much larger relative gains than simple random sampling.
# 6 Results | 1512.02595#58 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 59 | # 6 Acknowledgment
This work was started as an internship project at Google and continued at UNC. We would like to thank Alex Toshev for helpful discussions and are indebted to the Im- age Understanding and DistBelief teams at Google. We also thank Philip Ammirato and Patrick Poirson for helpful comments. We thank NVIDIA for providing GPUs and acknowledge support from NSF 1452851, 1446631, 1526367, 1533771.
# References
1. Uijlings, J.R., van de Sande, K.E., Gevers, T., Smeulders, A.W.: Selective search for object recognition. IJCV (2013)
2. Ren, S., He, K., Girshick, R., Sun, J.: Faster R-CNN: Towards real-time object detection with region proposal networks. In: NIPS. (2015)
3. He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. In: CVPR. (2016)
4. Sermanet, P., Eigen, D., Zhang, X., Mathieu, M., Fergus, R., LeCun, Y.: Overfeat: Integrated recognition, localization and detection using convolutional networks. In: ICLR. (2014)
SSD: Single Shot MultiBox Detector | 1512.02325#59 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 59 | # 6 Results
To better assess the real-world applicability of our speech system, we evaluate on a wide range of test sets. We use several publicly available benchmarks and several test sets collected internally. Together these test sets represent a wide range of challenging speech environments including low signal-to-noise ratios (noisy and far-ï¬eld), accented, read, spontaneous and conversational speech.
16
Fraction of Data Hours Regular Dev Noisy Dev 1% 10% 20% 50% 100% 120 1200 2400 6000 12000 29.23 13.80 11.65 9.51 8.46 50.97 22.99 20.41 15.90 13.59
Table 10: Comparison of English WER for Regular and Noisy development sets on increasing training dataset size. The architecture is a 9-layer model with 2 layers of 2D-invariant convolution and 7 recurrent layers with 68M parameters. | 1512.02595#59 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 60 | SSD: Single Shot MultiBox Detector
5. Redmon, J., Divvala, S., Girshick, R., Farhadi, A.: You only look once: Uniï¬ed, real-time object detection. In: CVPR. (2016)
6. Girshick, R.: Fast R-CNN. In: ICCV. (2015) 7. Erhan, D., Szegedy, C., Toshev, A., Anguelov, D.: Scalable object detection using deep
neural networks. In: CVPR. (2014)
8. Szegedy, C., Reed, S., Erhan, D., Anguelov, D.: Scalable, high-quality object detection. arXiv preprint arXiv:1412.1441 v3 (2015)
9. He, K., Zhang, X., Ren, S., Sun, J.: Spatial pyramid pooling in deep convolutional networks for visual recognition. In: ECCV. (2014)
10. Long, J., Shelhamer, E., Darrell, T.: Fully convolutional networks for semantic segmentation. In: CVPR. (2015) | 1512.02325#60 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 60 | All models are trained for 20 epochs on either the full English dataset, described in Table 9, or the full Mandarin dataset described in Section 5. We use stochastic gradient descent with Nesterov momentum [61] along with a minibatch of 512 utterances. If the norm of the gradient exceeds a threshold of 400, it is rescaled to 400 [47]. The model which performs the best on a held-out development set during training is chosen for evaluation. The learning rate is chosen from [1 10â4, 6 epoch. We use a momentum of 0.99 for all models.
The language models used are those described in Section 3.8. The decoding parameters from Equa- tion 12 are tuned on a held-out development set. We use a beam size of 500 for the English decoder and a beam size of 200 for the Mandarin decoder.
# 6.1 English | 1512.02595#60 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 61 | 10. Long, J., Shelhamer, E., Darrell, T.: Fully convolutional networks for semantic segmentation. In: CVPR. (2015)
11. Hariharan, B., Arbel´aez, P., Girshick, R., Malik, J.: Hypercolumns for object segmentation and ï¬ne-grained localization. In: CVPR. (2015)
12. Liu, W., Rabinovich, A., Berg, A.C.: ParseNet: Looking wider to see better. In: ILCR. (2016) 13. Zhou, B., Khosla, A., Lapedriza, A., Oliva, A., Torralba, A.: Object detectors emerge in deep
scene cnns. In: ICLR. (2015)
14. Howard, A.G.: Some improvements on deep convolutional neural network based image classiï¬cation. arXiv preprint arXiv:1312.5402 (2013)
15. Simonyan, K., Zisserman, A.: Very deep convolutional networks for large-scale image recog- nition. In: NIPS. (2015) | 1512.02325#61 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 61 | # 6.1 English
The best DS2 model has 11 layers with 3 layers of 2D convolution, 7 bidirectional recurrent layers, a fully-connected output layer along with Batch Normalization. The ï¬rst layer outputs to bigrams with a temporal stride of 3. By comparison the DS1 model has 5 layers with a single bidirectional recurrent layer and it outputs to unigrams with a temporal stride of 2 in the ï¬rst layer. We report results on several test sets for both the DS2 and DS1 model. We do not tune or adapt either model to any of the speech conditions in the test sets. Language model decoding parameters are set once on a held-out development set. | 1512.02595#61 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 62 | 15. Simonyan, K., Zisserman, A.: Very deep convolutional networks for large-scale image recog- nition. In: NIPS. (2015)
16. Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., Berg, A.C., Fei-Fei, L.: Imagenet large scale visual recognition challenge. IJCV (2015)
17. Chen, L.C., Papandreou, G., Kokkinos, I., Murphy, K., Yuille, A.L.: Semantic image seg- mentation with deep convolutional nets and fully connected crfs. In: ICLR. (2015)
18. Holschneider, M., Kronland-Martinet, R., Morlet, J., Tchamitchian, P.: A real-time algorithm for signal analysis with the help of the wavelet transform. In: Wavelets. Springer (1990) 286â297 | 1512.02325#62 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 62 | To put the performance of our system in context, we benchmark most of our results against human workers, since speech recognition is an audio perception and language understanding problem that humans excel at. We obtain a measure of human level performance by paying workers from Amazon Mechanical Turk to hand-transcribe all of our test sets. Two workers transcribe the same audio clip, that is typically about 5 seconds long, and we use the better of the two transcriptions for the ï¬nal WER calculation. They are free to listen to the audio clip as many times as they like. These workers are mostly based in the United States, and on average spend about 27 seconds per transcription. The hand-transcribed results are compared to the existing ground truth to produce a WER. While the existing ground truth transcriptions do have some label error, this is rarely more than 1%. This implies that disagreement between the ground truth transcripts and the human level transcripts is a good heuristic for human level performance.
# 6.1.1 Model Size | 1512.02595#62 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02325 | 63 | 19. Jia, Y., Shelhamer, E., Donahue, J., Karayev, S., Long, J., Girshick, R., Guadarrama, S., Darrell, T.: Caffe: Convolutional architecture for fast feature embedding. In: MM. (2014) 20. Glorot, X., Bengio, Y.: Understanding the difï¬culty of training deep feedforward neural
networks. In: AISTATS. (2010)
21. Hoiem, D., Chodpathumwan, Y., Dai, Q.: Diagnosing error in object detectors. In: ECCV 2012. (2012)
22. Girshick, R., Donahue, J., Darrell, T., Malik, J.: Rich feature hierarchies for accurate object detection and semantic segmentation. In: CVPR. (2014)
23. Zhang, L., Lin, L., Liang, X., He, K.: Is faster r-cnn doing well for pedestrian detection. In: ECCV. (2016)
24. Bell, S., Zitnick, C.L., Bala, K., Girshick, R.: Inside-outside net: Detecting objects in context with skip pooling and recurrent neural networks. In: CVPR. (2016) | 1512.02325#63 | SSD: Single Shot MultiBox Detector | We present a method for detecting objects in images using a single deep
neural network. Our approach, named SSD, discretizes the output space of
bounding boxes into a set of default boxes over different aspect ratios and
scales per feature map location. At prediction time, the network generates
scores for the presence of each object category in each default box and
produces adjustments to the box to better match the object shape. Additionally,
the network combines predictions from multiple feature maps with different
resolutions to naturally handle objects of various sizes. Our SSD model is
simple relative to methods that require object proposals because it completely
eliminates proposal generation and subsequent pixel or feature resampling stage
and encapsulates all computation in a single network. This makes SSD easy to
train and straightforward to integrate into systems that require a detection
component. Experimental results on the PASCAL VOC, MS COCO, and ILSVRC datasets
confirm that SSD has comparable accuracy to methods that utilize an additional
object proposal step and is much faster, while providing a unified framework
for both training and inference. Compared to other single stage methods, SSD
has much better accuracy, even with a smaller input image size. For $300\times
300$ input, SSD achieves 72.1% mAP on VOC2007 test at 58 FPS on a Nvidia Titan
X and for $500\times 500$ input, SSD achieves 75.1% mAP, outperforming a
comparable state of the art Faster R-CNN model. Code is available at
https://github.com/weiliu89/caffe/tree/ssd . | http://arxiv.org/pdf/1512.02325 | Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg | cs.CV | ECCV 2016 | null | cs.CV | 20151208 | 20161229 | [] |
1512.02595 | 63 | # 6.1.1 Model Size
Our English speech training set is substantially larger than the size of commonly used speech datasets. Furthermore, the data is augmented with noise synthesis. To get the best generalization error, we expect that the model size must increase to fully exploit the patterns in the data. In Sec- tion 3.2 we explored the effect of model depth while ï¬xing the number of parameters. In contrast, here we show the effect of varying model size on the performance of the speech system. We only vary the size of each layer, while keeping the depth and other architectural parameters constant. We evaluate the models on the same Regular and Noisy development sets that we use in Section 3.5.
The models in Table 11 differ from those in Table 3 in that we increase the the stride to 3 and output to bigrams. Because we increase the model size to as many as 100 million parameters, we ï¬nd that an increase in stride is necessary for fast computation and memory constraints. However, in this regime we note that the performance advantage of the GRU networks appears to diminish over the
17
Model size Model type Regular Dev Noisy Dev 18 38 70 70 100 100 106 106 106 106 106 106 GRU GRU GRU RNN GRU RNN 10.59 9.06 8.54 8.44 7.78 7.73 21.38 17.07 15.98 15.09 14.17 13.06 | 1512.02595#63 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 64 | Table 11: Comparing the effect of model size on the WER of the English speech system on both the regular and noisy development sets. We vary the number of hidden units in all but the convolutional layers. The GRU model has 3 layers of bidirectional GRUs with 1 layer of 2D-invariant convolution. The RNN model has 7 layers of bidirectional simple recurrence with 3 layers of 2D-invariant convolution. Both models output bigrams with a temporal stride of 3. All models contain approximately 35 million parameters and are trained with BatchNorm and SortaGrad.
Test set DS1 DS2 Baidu Test 24.01 13.59
Table 12: Comparison of DS1 and DS2 WER on an internal test set of 3,300 examples. The test set contains a wide variety of speech including accents, low signal-to-noise speech, spontaneous and conversational speech.
simple RNN. In fact, for the 100 million parameter networks the simple RNN performs better than the GRU network and is faster to train despite the 2 extra layers of convolution.
Table 11 shows that the performance of the system improves consistently up to 100 million parame- ters. All further English DS2 results are reported with this same 100 million parameter RNN model since it achieves the lowest generalization errors. | 1512.02595#64 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 65 | Table 12 shows that the 100 million parameter RNN model (DS2) gives a 43.4% relative improve- ment over the 5-layer model with 1 recurrent layer (DS1) on an internal Baidu dataset of 3,300 utterances that contains a wide variety of speech including challenging accents, low signal-to-noise ratios from far-ï¬eld or background noise, spontaneous and conversational speech.
# 6.1.2 Read Speech
Read speech with high signal-to-noise ratio is arguably the easiest large vocabulary for a continuous speech recognition task. We benchmark our system on two test sets from the Wall Street Journal (WSJ) corpus of read news articles. These are available in the LDC catalog as LDC94S13B and LDC93S6B. We also take advantage of the recently developed LibriSpeech corpus constructed using audio books from the LibriVox project [46].
Table 13 shows that the DS2 system outperforms humans in 3 out of the 4 test sets and is competitive on the fourth. Given this result, we suspect that there is little room for a generic speech system to further improve on clean read speech without further domain adaptation. | 1512.02595#65 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 66 | Read Speech Test set DS1 DS2 Human WSJ evalâ92 WSJ evalâ93 LibriSpeech test-clean LibriSpeech test-other 4.94 6.94 7.89 21.74 3.60 4.98 5.33 13.25 5.03 8.08 5.83 12.69
Table 13: Comparison of WER for two speech systems and human level performance on read speech.
18
Accented Speech Test set DS1 DS2 Human VoxForge American-Canadian VoxForge Commonwealth VoxForge European VoxForge Indian 15.01 28.46 31.20 45.35 7.55 13.56 17.55 22.44 4.85 8.15 12.76 22.15
Table 14: Comparing WER of the DS1 system to the DS2 system on accented speech.
Noisy Speech Test set DS1 DS2 Human CHiME eval clean CHiME eval real CHiME eval sim 6.30 67.94 80.27 3.34 21.79 45.05 3.46 11.84 31.33 | 1512.02595#66 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 67 | Table 15: Comparison of DS1 and DS2 system on noisy speech. âCHiME eval cleanâ is a noise-free baseline. The âCHiME eval realâ dataset is collected in real noisy environments and the âCHiME eval simâ dataset has similar noise synthetically added to clean speech. Note that we use only one of the six channels to test each utterance.
# 6.1.3 Accented Speech
Our source for accented speech is the publicly available VoxForge (http://www.voxforge.org) dataset, which has clean speech read from speakers with many different accents. We group these accents into four categories. The American-Canadian and Indian groups are self-explanatory. The Commonwealth accent denotes speakers with British, Irish, South African, Australian and New Zealand accents. The European group contains speakers with accents from countries in Europe that do not have English as a ï¬rst language. We construct a test set from the VoxForge data with 1024 examples from each accent group for a total of 4096 examples.
Performance on these test sets is to some extent a measure of the breadth and quality of our training data. Table 14 shows that our performance improved on all the accents when we include more accented training data and use an architecture that can effectively train on that data. However human level performance is still notably better than that of DS2 for all but the Indian accent. | 1512.02595#67 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 68 | # 6.1.4 Noisy Speech
We test our performance on noisy speech using the publicly available test sets from the recently completed third CHiME challenge [4]. This dataset has 1320 utterances from the WSJ test set read in various noisy environments, including a bus, a cafe, a street and a pedestrian area. The CHiME set also includes 1320 utterances with simulated noise from the same environments as well as the control set containing the same utterances delivered by the same speakers in a noise-free environment. Differences between results on the control set and the noisy sets provide a measure of the networkâs ability to handle a variety of real and synthetic noise conditions. The CHiME audio has 6 channels and using all of them can provide substantial performance improvements [69]. We use a single channel for all our results, since multi-channel audio is not pervasive on most devices. Table 15 shows that DS2 substantially improves upon DS1, however DS2 is worse than human level performance on noisy data. The relative gap between DS2 and human level performance is larger when the data comes from a real noisy environment instead of synthetically adding noise to clean speech.
# 6.2 Mandarin | 1512.02595#68 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 69 | # 6.2 Mandarin
In Table 16 we compare several architectures trained on the Mandarin Chinese speech, on a develop- ment set of 2000 utterances as well as a test set of 1882 examples of noisy speech. This development set was also used to tune the decoding parameters We see that the deepest model with 2D-invariant
19
convolution and BatchNorm outperforms the shallow RNN by 48% relative, thus continuing the trend that we saw with the English systemâmultiple layers of bidirectional recurrence improves performance substantially.
Architecture Dev Test 5-layer, 1 RNN 5-layer, 3 RNN 5-layer, 3 RNN + BatchNorm 9-layer, 7 RNN + BatchNorm + 2D conv 7.13 6.49 6.22 5.81 15.41 11.85 9.39 7.93
Table 16: Comparison of the improvements in DeepSpeech with architectural improvements. The development and test sets are Baidu internal corpora. All the models in the table have about 80 million parameters each | 1512.02595#69 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 70 | Table 16: Comparison of the improvements in DeepSpeech with architectural improvements. The development and test sets are Baidu internal corpora. All the models in the table have about 80 million parameters each
We ï¬nd that our best Mandarin Chinese speech system transcribes short voice-query like utterances better than a typical Mandarin Chinese speaker. To benchmark against humans we ran a test with 100 randomly selected utterances and had a group of 5 humans label all of them together. The group of humans had an error rate of 4.0% as compared to the speech systems performance of 3.7%. We also compared a single human transcriber to the speech system on 250 randomly selected utterances. In this case the speech system performs much better: 9.7% for the human compared to 5.7% for the speech model.
# 7 Deployment | 1512.02595#70 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 71 | # 7 Deployment
Real-world applications usually require a speech system to transcribe in real time or with relatively low latency. The system used in Section 6.1 is not well-designed for this task, for several reasons. First, since the RNN has several bidirectional layers, transcribing the ï¬rst part of an utterance re- quires the entire utterance to be presented to the RNN. Second, since we use a wide beam when decoding with a language model, beam search can be expensive, particularly in Mandarin where the number of possible next characters is very large (around 6000). Third, as described in Section 3, we normalize power across an entire utterance, which again requires the entire utterance to be available in advance.
We solve the power normalization problem by using some statistics from our training set to perform an adaptive normalization of speech inputs during online transcription. We can solve the other problems by modifying our network and decoding procedure to produce a model that performs almost as well while having much lower latency. We focus on our Mandarin system since some aspects of that system are more challenging to deploy (e.g. the large character set), but the same techniques could also be applied in English. | 1512.02595#71 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 72 | In this section, latency refers to the computational latency of our speech system as measured from the end of an utterance until the transcription is produced. This latency does not include data trans- mission over the internet, and does not measure latency from the beginning of an utterance until the ï¬rst transcription is produced. We focus on latency from end of utterance to transcription because it is important to applications using speech recognition.
# 7.1 Batch Dispatch
In order to deploy our relatively large deep neural networks at low latency, we have paid special at- tention to efï¬ciency during deployment. Most internet applications process requests individually as they arrive in the data center. This makes for a straightforward implementation where each request can be managed by one thread. However, processing requests individually is inefï¬cient computa- tionally, for two main reasons. Firstly, when processing requests individually, the processor must load all the weights of the network for each request. This lowers the arithmetic intensity of the work- load, and tends to make the computation memory bandwidth bound, as it is difï¬cult to effectively use on-chip caches when requests are presented individually. Secondly, the amount of parallelism that can be exploited to classify one request is limited, making it difï¬cult to exploit SIMD or multi- core parallelism. RNNs are especially challenging to deploy because evaluating RNNs sample by
20 | 1512.02595#72 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 73 | 20
0.4 1) 10 streams GH 20 streams CC) 320 streams pm 0.3 B FA 3 202 = 7 - | | | | | 0.0 i I 1 2 3 4 5 6 7 8 9 10 11 Batch size
Figure 5: Probability that a request is processed in a batch of given size
sample relies on sequential matrix vector multiplications, which are bandwidth bound and difï¬cult to parallelize.
To overcome these issues, we built a batching scheduler called Batch Dispatch that assembles streams of data from user requests into batches before performing forward propagation on these batches. In this case, there is a tradeoff between increased batch size, and consequently improved efï¬ciency, and increased latency. The more we buffer user requests to assemble a large batch, the longer users must wait for their results. This places constraints on the amount of batching we can perform.
We use an eager batching scheme that processes each batch as soon as the previous batch is com- pleted, regardless of how much work is ready by that point. This scheduling algorithm has proved to be the best at reducing end-user latency, despite the fact that it is less efï¬cient computationally, since it does not attempt to maximize batch size. | 1512.02595#73 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 74 | Figure 5 shows the probability that a request is processed in a batch of given size for our production system running on a single NVIDIA Quadro K1200 GPU, with 10-30 concurrent user requests. As expected, batching works best when the server is heavily loaded: as load increases, the distribution shifts to favor processing requests in larger batches. However, even with a light load of only 10 concurrent user requests, our system performs more than half the work in batches with at least 2 samples.
GH 50%ile 100} |) 98%ile - latency (ms) 0 10 20 30 40 Number of concurrent streams
Figure 6: Median and 98 percentile latencies as a function of server load
We see in Figure 6, that our system achieves a median latency of 44 ms, and a 98 percentile latency of 70 ms when loaded with 10 concurrent streams. As the load increases on the server, the batching scheduler shifts work to more efï¬cient batches, which keeps latency low. This shows that Batch Dispatch makes it possible to deploy these large models at high throughput and low latency.
21
â_ nervana 0.4 â _baidu 2 a fo) 0.3 aq a é 0.2 O01 0.0 1 2 3 4 5 6 7 8 9 10 Batch size | 1512.02595#74 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 75 | Figure 7: Comparison of kernels that compute Ax = b where A is a matrix with dimension 2560 Ã 2560, and x is a matrix with dimension 2560 Ã Batch size, where Batch size â [1, 10]. All matrices are in half-precision format.
# 7.2 Deployment Optimized Matrix Multiply Kernels
We have found that deploying our models using half-precision (16-bit) ï¬oating-point arithmetic does not measurably change recognition accuracy. Because deployment does not require any updates to the network weights, it is far less sensitive to numerical precision than training. Using half-precision arithmetic saves memory space and bandwidth, which is especially useful for deployment, since RNN evaluation is dominated by the cost of caching and streaming the weight matrices. | 1512.02595#75 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 76 | As seen in Section 7.1, the batch size during deployment is much smaller than in training. We found that standard BLAS libraries are inefï¬cient at this batch size. To overcome this, we wrote our own half-precision matrix-matrix multiply kernel. For 10 simultaneous streams over 90 percent of 4, a regime where the matrix multiply will be bandwidth bound. We store the A batches are for N matrix transposed to maximize bandwidth by using the widest possible vector loads while avoiding transposition after loading. Each warp computes four rows of output for all N output columns. Note 4 the B matrix ï¬ts entirely in the L1 cache. This scheme achieves 90 percent of peak that for N 4 but starts to lose efï¬ciency for larger N as the B matrix stops ï¬tting into the bandwidth for N L1 cache. Nonetheless, it continues to provide improved performance over existing libraries up to N = 10.
Figure 7 shows that our deployment kernel sustains a higher computational throughput than those from Nervana Systems [44] on the K1200 GPU, across the entire range of batch sizes that we use in deployment. Both our kernels and the Nervana kernels are signiï¬cantly faster than NVIDIA CUBLAS version 7.0, more details are found here [20].
# 7.3 Beam Search | 1512.02595#76 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 77 | # 7.3 Beam Search
Performing the beam search involves repeated lookups in the n-gram language model, most of which translate to uncached reads from memory. The direct implementation of beam search means that each time-step dispatches one lookup per character for each beam. In Mandarin, this results in over 1M lookups per 40ms stride of speech data, which is too slow for deployment. To deal with this problem, we use a heuristic to further prune the beam search. Rather than considering all characters as viable additions to the beam, we only consider the fewest number of characters whose cumulative probability is at least p. In practice, we have found that p = 0.99 works well. Additionally, we limit ourselves to no more than 40 characters. This speeds up the Mandarin language model lookup time by a factor of 150x, and has a negligible effect on the CER (0.1-0.3% relative).
# 7.4 Results
We can deploy our system at low latency and high throughput without sacriï¬cing much accuracy. On a held-out set of 2000 utterances, our research system achieves 5.81 character error rate whereas the deployed system achieves 6.10 character error rate. This is only a 5% relative degradation for
22 | 1512.02595#77 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 78 | 22
the deployed system. In order to accomplish this, we employ a neural network architecture with low deployment latency, reduce the precision of our network to 16-bit, built a batching scheduler to more efï¬ciently evaluate RNNs, and ï¬nd a simple heuristic to reduce beam search cost. The model has ï¬ve forward-only recurrent layers with 2560 hidden units, one row convolution layer (Section 3.7) with Ï = 19, and one fully-connected layer with 2560 hidden units. These techniques allow us to deploy Deep Speech at low cost to interactive applications.
# 8 Conclusion
End-to-end deep learning presents the exciting opportunity to improve speech recognition systems continually with increases in data and computation. Indeed, our results show that, compared to the previous incarnation, Deep Speech has signiï¬cantly closed the gap in transcription performance with human workers by leveraging more data and larger models. Further, since the approach is highly generic, weâve shown that it can quickly be applied to new languages. Creating high-performing recognizers for two very different languages, English and Mandarin, required essentially no expert knowledge of the languages. Finally, we have also shown that this approach can be efï¬ciently deployed by batching user requests together on a GPU server, paving the way to deliver end-to-end Deep Learning technologies to users. | 1512.02595#78 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 79 | To achieve these results, we have explored various network architectures, ï¬nding several effective techniques: enhancements to numerical optimization through SortaGrad and Batch Normalization, evaluation of RNNs with larger strides with bigram outputs for English, searching through both bidirectional and unidirectional models. This exploration was powered by a well optimized, High Performance Computing inspired training system that allows us to train new, full-scale models on our large datasets in just a few days.
Overall, we believe our results conï¬rm and exemplify the value of end-to-end Deep Learning meth- ods for speech recognition in several settings. In those cases where our system is not already com- parable to humans, the difference has fallen rapidly, largely because of application-agnostic Deep Learning techniques. We believe these techniques will continue to scale, and thus conclude that the vision of a single speech system that outperforms humans in most scenarios is imminently achiev- able.
# Acknowledgments
We are grateful to Baiduâs speech technology group for help with data preparation and useful con- versations. We would like to thank Scott Gray, Amir Khosrowshahi and all of Nervana Systems for their excellent matrix multiply routines and useful discussions. We would also like to thank Natalia Gimelshein of NVIDIA for useful discussions and thoughts on implementing our fast deployment matrix multiply.
# References | 1512.02595#79 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 80 | # References
[1] O. Abdel-Hamid, A.-r. Mohamed, H. Jang, and G. Penn. Applying convolutional neural networks con- cepts to hybrid nn-hmm model for speech recognition. In ICASSP, 2012.
[2] D. Bahdanau, K. Cho, and Y. Bengio. Neural machine translation by jointly learning to align and translate. In ICLR, 2015.
[3] D. Bahdanau, J. Chorowski, D. Serdyuk, P. Brakel, and Y. Bengio. End-to-end attention-based large vocabulary speech recognition. abs/1508.04395, 2015. http://arxiv.org/abs/1508.04395.
[4] J. Barker, E. Marxer, Ricard Vincent, and S. Watanabe. The third âCHiMEâ speech separation and recogni- tion challenge: Dataset, task and baselines. 2015. Submitted to IEEE 2015 Automatic Speech Recognition and Understanding Workshop (ASRU).
[5] S. Baxter. Modern GPU. https://nvlabs.github.io/moderngpu/. | 1512.02595#80 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 81 | [5] S. Baxter. Modern GPU. https://nvlabs.github.io/moderngpu/.
[6] Y. Bengio, J. Louradour, R. Collobert, and J. Weston. Curriculum learning. In International Conference on Machine Learning, 2009.
[7] H. Bourlard and N. Morgan. Connectionist Speech Recognition: A Hybrid Approach. Kluwer Academic Publishers, Norwell, MA, 1993.
23
[8] W. Chan, N. Jaitly, Q. Le, and O. Vinyals. Listen, attend, and spell. abs/1508.01211, 2015. http://arxiv.org/abs/1508.01211.
[9] S. Chetlur, C. Woolley, P. Vandermersch, J. Cohen, J. Tran, B. Catanzaro, and E. Shelhamer. cuDNN: Efï¬cient primitives for deep learning.
[10] T. Chilimbi, Y. Suzue, J. Apacible, and K. Kalyanaraman. Project adam: Building an efï¬cient and scalable deep learning training system. In USENIX Symposium on Operating Systems Design and Implementation, 2014. | 1512.02595#81 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 82 | [11] K. Cho, B. Van Merrienboer, C. Gulcehre, D. Bahdanau, F. Bougares, H. Schwenk, and Y. Bengio. Learn- ing phrase representations using rnn encoder-decoder for statistical machine translation. In EMNLP, 2014.
[12] J. Chorowski, D. Bahdanau, K. Cho, and Y. Bengio. End-to-end continuous speech recognition using attention-based recurrent nn: First results. abs/1412.1602, 2015. http://arxiv.org/abs/1412.1602.
[13] C. Cieri, D. Miller, and K. Walker. The Fisher corpus: a resource for the next generations of speech-to- text. In LREC, volume 4, pages 69â71, 2004.
[14] A. Coates, B. Carpenter, C. Case, S. Satheesh, B. Suresh, T. Wang, D. J. Wu, and A. Y. Ng. Text detection and character recognition in scene images with unsupervised feature learning. In International Conference on Document Analysis and Recognition, 2011. | 1512.02595#82 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 83 | [15] A. Coates, B. Huval, T. Wang, D. J. Wu, A. Y. Ng, and B. Catanzaro. Deep learning with COTS HPC. In International Conference on Machine Learning, 2013.
[16] G. Dahl, D. Yu, and L. Deng. Large vocabulary continuous speech recognition with context-dependent DBN-HMMs. In Proc. ICASSP, 2011.
[17] G. Dahl, D. Yu, L. Deng, and A. Acero. Context-dependent pre-trained deep neural networks for large vocabulary speech recognition. IEEE Transactions on Audio, Speech, and Language Processing, 2011.
[18] J. Dean, G. S. Corrado, R. Monga, K. Chen, M. Devin, Q. Le, M. Mao, M. Ranzato, A. Senior, P. Tucker, K. Yang, and A. Ng. Large scale distributed deep networks. In Advances in Neural Information Processing Systems 25, 2012.
[19] D. Ellis and N. Morgan. Size matters: An empirical study of neural network training for large vocabulary continuous speech recognition. In ICASSP, volume 2, pages 1013â1016. IEEE, 1999. | 1512.02595#83 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 84 | [20] E. Elsen. Optimizing RNN performance. http://svail.github.io/rnn_perf. Accessed: 2015-11-24.
[21] M. J. F. Gales, A. Ragni, H. Aldamarki, and C. Gautier. Support vector machines for noise robust ASR. In ASRU, pages 205â2010, 2009.
[22] A. Graves, S. Fernández, F. Gomez, and J. Schmidhuber. Connectionist temporal classiï¬cation: Labelling unsegmented sequence data with recurrent neural networks. In ICML, pages 369â376. ACM, 2006.
[23] A. Graves and N. Jaitly. Towards end-to-end speech recognition with recurrent neural networks. In ICML, 2014.
[24] A. Graves, A.-r. Mohamed, and G. Hinton. Speech recognition with deep recurrent neural networks. In ICASSP, 2013.
[25] H. H. Sak, A. Senior, and F. Beaufays. Long short-term memory recurrent neural network architectures for large scale acoustic modeling. In Interspeech, 2014. | 1512.02595#84 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 85 | [26] A. Hannun, C. Case, J. Casper, B. Catanzaro, G. Diamos, E. Elsen, R. Prenger, S. Satheesh, S. Sengupta, A. Coates, and A. Y. Ng. Deep speech: Scaling up end-to-end speech recognition. 1412.5567, 2014. http://arxiv.org/abs/1412.5567.
[27] A. Y. Hannun, A. L. Maas, D. Jurafsky, and A. Y. Ng. First-pass large vocabulary continuous speech recognition using bi-directional recurrent DNNs. abs/1408.2873, 2014. http://arxiv.org/abs/1408.2873.
[28] K. Heaï¬eld, I. Pouzyrevsky, J. H. Clark, and P. Koehn. Scalable modiï¬ed Kneser-Ney language model estimation. In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics, Soï¬a, Bulgaria, 8 2013. | 1512.02595#85 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 86 | [29] G. Hinton, L. Deng, D. Yu, G. Dahl, A. Mohamed, N. Jaitly, A. Senior, V. Vanhoucke, P. Nguyen, T. Sainath, and B. Kingsbury. Deep neural networks for acoustic modeling in speech recognition. IEEE Signal Processing Magazine, 29(November):82â97, 2012.
[30] S. Hochreiter and J. Schmidhuber. Long short-term memory. Neural Computation, 9(8):1735â1780, 1997.
[31] N. Jaitly and G. Hinton. Vocal tract length perturbation (VTLP) improves speech recognition. In ICML Workshop on Deep Learning for Audio, Speech, and Language Processing, 2013.
[32] R. Jozefowicz, W. Zaremba, and I. Sutskever. An empirical exploration of recurrent network architectures. In ICML, 2015.
24
[33] O. Kapralova, J. Alex, E. Weinstein, P. Moreno, and O. Siohan. A big data approach to acoustic model training corpus selection. In Interspeech, 2014. | 1512.02595#86 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 87 | [34] K. C. Knowlton. A fast storage allocator. Commun. ACM, 8(10):623â624, Oct. 1965. [35] T. Ko, V. Peddinti, D. Povey, and S. Khudanpur. Audio augmentation for speech recognition. In Interspeech, 2015.
[36] A. Krizhevsky, I. Sutskever, and G. Hinton. Imagenet classiï¬cation with deep convolutional neural net- works. In Advances in Neural Information Processing Systems 25, pages 1106â1114, 2012.
[37] C. Laurent, G. Pereyra, P. Brakel, Y. Zhang, and Y. Bengio. Batch normalized recurrent neural networks. abs/1510.01378, 2015. http://arxiv.org/abs/1510.01378.
[38] Q. Le, M. Ranzato, R. Monga, M. Devin, K. Chen, G. Corrado, J. Dean, and A. Ng. Building high-level features using large scale unsupervised learning. In International Conference on Machine Learning, 2012. | 1512.02595#87 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 88 | [39] Y. LeCun, F. J. Huang, and L. Bottou. Learning methods for generic object recognition with invariance to pose and lighting. In Computer Vision and Pattern Recognition, volume 2, pages 97â104, 2004.
[40] A. Maas, Z. Xie, D. Jurafsky, and A. Ng. Lexicon-free conversational speech recognition with neural networks. In NAACL, 2015.
[41] Y. Miao, M. Gowayyed, and F. Metz. EESEN: End-to-end speech recognition using deep rnn models and wfst-based decoding. In ASRU, 2015.
[42] A. Mohamed, G. Dahl, and G. Hinton. Acoustic modeling using deep belief networks. IEEE Transactions on Audio, Speech, and Language Processing, (99), 2011.
[43] A. S. N. Jaitly, P. Nguyen and V. Vanhoucke. Application of pretrained deep neural networks to large vocabulary speech recognition. In Interspeech, 2012.
[44] Nervana Systems. Nervana GPU. https://github.com/NervanaSystems/nervanagpu. Accessed: 2015-11-06. | 1512.02595#88 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 89 | [44] Nervana Systems. Nervana GPU. https://github.com/NervanaSystems/nervanagpu. Accessed: 2015-11-06.
[45] J. Niu, L. Xie, L. Jia, and N. Hu. Context-dependent deep neural networks for commercial mandarin speech recognition applications. In APSIPA, 2013.
[46] V. Panayotov, G. Chen, D. Povey, and S. Khudanpur. Librispeech: an asr corpus based on public domain audio books. In ICASSP, 2015.
[47] R. Pascanu, T. Mikolov, and Y. Bengio. On the difï¬culty of training recurrent neural networks. abs/1211.5063, 2012. http://arxiv.org/abs/1211.5063.
[48] P. Patarasuk and X. Yuan. Bandwidth optimal all-reduce algorithms for clusters of workstations. J. Parallel Distrib. Comput., 69(2):117â124, Feb. 2009.
[49] R. Raina, A. Madhavan, and A. Ng. Large-scale deep unsupervised learning using graphics processors. In 26th International Conference on Machine Learning, 2009. | 1512.02595#89 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 90 | [50] S. Renals, N. Morgan, H. Bourlard, M. Cohen, and H. Franco. Connectionist probability estimators in HMM speech recognition. IEEE Transactions on Speech and Audio Processing, 2(1):161â174, 1994.
[51] T. Robinson, M. Hochberg, and S. Renals. The use of recurrent neural networks in continuous speech recognition. pages 253â258, 1996.
[52] T. Sainath, O. Vinyals, A. Senior, and H. Sak. Convolutional, long short-term memory, fully connected deep neural networks. In ICASSP, 2015.
[53] T. N. Sainath, A. rahman Mohamed, B. Kingsbury, and B. Ramabhadran. Deep convolutional neural networks for LVCSR. In ICASSP, 2013.
[54] H. Sak, A. Senior, K. Rao, and F. Beaufays. Fast and accurate recurrent neural network acoustic models for speech recognition. abs/1507.06947, 2015. http://arxiv.org/abs/1507.06947. | 1512.02595#90 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 91 | [55] H. Sak, O. Vinyals, G. Heigold, A. Senior, E. McDermott, R. Monga, and M. Mao. Sequence discrimina- tive distributed training of long shortterm memory recurrent neural networks. In Interspeech, 2014.
[56] B. Sapp, A. Saxena, and A. Ng. A fast data collection and augmentation procedure for object recognition. In AAAI Twenty-Third Conference on Artiï¬cial Intelligence, 2008.
[57] M. Schuster and K. K. Paliwal. Bidirectional recurrent neural networks. IEEE Transactions on Signal Processing, 45(11):2673â2681, 1997.
[58] F. Seide, G. Li, and D. Yu. Conversational speech transcription using context-dependent deep neural networks. In Interspeech, pages 437â440, 2011.
[59] J. Shan, G. Wu, Z. Hu, X. Tang, M. Jansche, and P. Moreno. Search by voice in mandarin chinese. In Interspeech, 2010.
[60] H. Soltau, G. Saon, and T. Sainath. Joint training of convolutional and non-convolutional neural networks. In ICASSP, 2014.
25 | 1512.02595#91 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 92 | 25
[61] I. Sutskever, J. Martens, G. Dahl, and G. Hinton. On the importance of momentum and initialization in deep learning. In 30th International Conference on Machine Learning, 2013.
[62] I. Sutskever, O. Vinyals, and Q. V. Le. Sequence to sequence learning with neural networks. 2014. http://arxiv.org/abs/1409.3215.
[63] C. Szegedy and S. Ioffe. Batch normalization: Accelerating deep network training by reducing internal covariate shift. abs/1502.03167, 2015. http://arxiv.org/abs/1502.03167.
[64] C. Szegedy, W. Liu, Y. Jia, P. Sermanet, S. Reed, D. Anguelov, D. Erhan, V. Vanhoucke, and A. Rabi- novich. Going deeper with convolutions. 2014.
[65] R. Thakur and R. Rabenseifner. Optimization of collective communication operations in mpich. Interna- tional Journal of High Performance Computing Applications, 19:49â66, 2005. | 1512.02595#92 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 93 | [66] K. Vesely, A. Ghoshal, L. Burget, and D. Povey. Sequence-discriminative training of deep neural net- works. In Interspeech, 2013.
[67] A. Waibel, T. Hanazawa, G. Hinton, K. Shikano, and K. Lang. Phoneme recognition using time-delay neural networks,â ËAËI acoustics speech and signal processing. IEEE Transactions on Acoustics, Speech and Signal Processing, 37(3):328â339, 1989.
[68] R. Williams and J. Peng. An efï¬cient gradient-based algorithm for online training of recurrent network trajectories. Neural computation, 2:490â501, 1990.
[69] T. Yoshioka, N. Ito, M. Delcroix, A. Ogawa, K. Kinoshita, M. F. C. Yu, W. J. Fabian, M. Espi, T. Higuchi, S. Araki, and T. Nakatani. The ntt chime-3 system: Advances in speech enhancement and recognition for mobile multi-microphone devices. In IEEE ASRU, 2015. | 1512.02595#93 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 94 | [70] W. Zaremba and I. Sutskever. Learning to execute. abs/1410.4615, 2014. http://arxiv.org/abs/1410.4615.
# A Scalability improvements
In this section, we discuss some of our scalability improvements in more detail.
# A.1 Node and cluster architecture
The software stack runs on a compute dense node built from 2 Intel CPUs and 8 NVIDIA Titan X GPUs, with peak single-precision computational throughput of 53 teraFLOP/second. Each node also has 384 GB of CPU memory and an 8 TB storage volume built from two 4 TB hard disks in RAID-0 conï¬guration. We use the CPU memory to cache our input data so that we are not directly exposed to the low bandwidth and high latency of spinning disks. We replicate our English and Mandarin datasets on each nodeâs local hard disk. This allows us to use our network only for weight updates and avoids having to rely on centralized ï¬le servers.
: cpu }}â+] cpu 3 PLX PLX]! : | PLX PLX :|@Pu]|Gpu}|apu}}a@pu]: :]|apu}]apu} |a@pu| | qu] : | 1512.02595#94 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 95 | Figure 8: Schematic of our training node where PLX indicates a PCI switch and the dotted box includes all devices that are connected by the same PCI root complex.
Figure 8 shows a schematic diagram of one our nodes, where all devices connected by the same PCI root complex are encapsulated in a dotted box. We have tried to maximize the number of GPUs within the root complex for faster communication between GPUs using GPUDirect. This allows us to use an efï¬cient communication mechanism to transfer gradient matrices between GPUs.
26
All the nodes in our cluster are connected through Fourteen Data Rate (FDR) Inï¬niband which is primarily used for gradient transfer during back-propagation.
# A.2 GPU Implementation of CTC Loss Function
The CTC loss function that we use to train our models has two passes: forward and backward, and the gradient computation involves element-wise addition of two matrices, α and β, generated during the forward and backward passes respectively. Finally, we sum the gradients using the character in the utterance label as the key, to generate one gradient per character. These gradients are then back- propagated through the network. The input to the CTC loss function are probabilities calculated by the softmax function which can be very small, so we compute in log probability space for better numerical stability. | 1512.02595#95 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 96 | The forward pass of the CTC algorithm calculates the α matrix, which has S rows and T columns, where S = 2(L + 1). The variable L is the number of characters in the label and T is the number of time-steps in the utterance. Our CPU-based implementation of the CTC algorithm assigns one thread to each utterance label in a minibatch, performing the CTC calculation for the utterances in parallel. Each thread calculates the relevant entries of the matrix sequentially. This is inefï¬cient for two reasons.
Firstly, since the remainder of our network is computed on the GPU, the output of the softmax function has to be copied to the CPU for CTC calculation. The gradient matrices from the CTC function then has to be copied back to the GPU for backpropagation. For languages like Mandarin with large character sets, these matrices have hundreds of millions of entries, making this copy expensive. Furthermore, we need as much interconnect bandwidth as possible for synchronizing the gradient updates with data parallelism, so this copy incurs a substantial opportunity cost.
Secondly, although entries in each column of the α matrix can be computed in parallel, the number of entries to calculate in each column depends both on the column and the number of repeated characters in the utterance label. Due to this complexity, the CPU implementation does not use SIMD parallelism optimally, making the computation inefï¬cient. | 1512.02595#96 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 97 | We wrote a GPU-based implementation of CTC in order to overcome these two problems. The key insight behind our implementation is that we can compute all elements in each column of the α matrix, rather than just the valid entries. If we do so, Figure 9 shows that invalid elements either (I), when we use a special summation function that adds contain a ï¬nite garbage value (G), or probabilities in log space that discards inputs that are . This summation is shown in Figure 9 where arrows incident on a circle are inputs and the result is stored in the circle. However, when we compute the ï¬nal gradient by element-wise summing α and β, all ï¬nite garbage values will be added with a corresponding , effectively ignoring the ââ garbage value and computing the correct result. One important observation is that this element-wise sum of α and β is a simple sum and does not use our summation function. | 1512.02595#97 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 98 | To compute the gradient, we take each column of the matrix generated from element-wise addition of α and β matrices, and do a key-value reduction using the character as key, using the ModernGPU library [5]. This means elements of the column corresponding to the same character will sum up , is the only repeated character their values. In the example shown in Figure 9, the blank character, B and at some columns, say for t = 1 of t = 2, both valid elements (gray) and correspond to it. Since our summation function in log space effectively ignores the elements, only the valid elements are combined in the reduction.
In our GPU implementation, we map each utterance in the minibatch to a CUDA thread block. Since there are no dependencies between the elements of a column, all of them can be computed in parallel by the threads in a threadblock. There are dependencies between columns, since the column corresponding to time-step t + 1 cannot be computed before the column corresponding to time-step t. The reverse happens when computing the β matrix, when column corresponding to time-step t cannot be computed before the column corresponding to time-step t + 1. Thus, in both cases, columns are processed sequentially by the thread block. | 1512.02595#98 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 99 | Mapping the forward and backward passes to corresponding CUDA kernels is straightforward since there are no data dependencies between elements of a column. The kernel that does the backward pass also computes the gradient. However, since the gradients must be summed up based on the label
27
T3) T-2 TA T B fetfetfet_fo) c © © B A Qa B T B B Cc B A B T B ââââ
Figure 9: Forward and backward pass for GPU implementation of CTC. Gray circles contain valid values, circle with I contain ââ and circle with G contain garbage values that are ï¬nite. B stand for the blank character that the CTC algorithm adds to the input utterance label. Column labels on top show different time-steps going from 1 to T. | 1512.02595#99 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 100 | values, with each character as key, we must deal with data dependencies due to repeated characters in an utterance label. For languages with small character sets like English, this happens with high probability. Even if there are no repeated characters, the CTC algorithm adds L + 1 blank characters to the utterance label. We solve this problem by performing a key-value sort, where the keys are the characters in the utterance label, and the values are the indices of each character in the utterance. After sorting, all occurrences of a given character are arranged in contiguous segments. We only need to do the sort once for each utterance. The indices generated by the sort are then used to sequentially sum up the gradients for each character. This sum is done once per column and in parallel over all characters in the utterance. Amortizing the cost of key-value sort over T columns is a key insight that makes the gradient calculation fast. | 1512.02595#100 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02595 | 101 | Our GPU implementation uses fast shared memory and registers to achieve high performance when performing this task. Both forward and backward kernels store the α matrix in shared memory. Since shared memory is a limited resource, it is not possible to store the entire β matrix. However, as we go backward in time, we only need to keep one column of the β matrix as we compute the gradient, adding element-wise the column of the β matrix with the corresponding column of the α matrix. Due to on-chip memory space constraints, we read the output of the softmax function directly from off-chip global memory.
Due to inaccuracies in ï¬oating-point arithmetic, especially in transcendental functions, our GPU and CPU implementation are not bit-wise identical. This is not an impediment in practice, since both implementations train models equally well when coupled with the technique of sorting utterances by length mentioned in Section 3.3.
28 | 1512.02595#101 | Deep Speech 2: End-to-End Speech Recognition in English and Mandarin | We show that an end-to-end deep learning approach can be used to recognize
either English or Mandarin Chinese speech--two vastly different languages.
Because it replaces entire pipelines of hand-engineered components with neural
networks, end-to-end learning allows us to handle a diverse variety of speech
including noisy environments, accents and different languages. Key to our
approach is our application of HPC techniques, resulting in a 7x speedup over
our previous system. Because of this efficiency, experiments that previously
took weeks now run in days. This enables us to iterate more quickly to identify
superior architectures and algorithms. As a result, in several cases, our
system is competitive with the transcription of human workers when benchmarked
on standard datasets. Finally, using a technique called Batch Dispatch with
GPUs in the data center, we show that our system can be inexpensively deployed
in an online setting, delivering low latency when serving users at scale. | http://arxiv.org/pdf/1512.02595 | Dario Amodei, Rishita Anubhai, Eric Battenberg, Carl Case, Jared Casper, Bryan Catanzaro, Jingdong Chen, Mike Chrzanowski, Adam Coates, Greg Diamos, Erich Elsen, Jesse Engel, Linxi Fan, Christopher Fougner, Tony Han, Awni Hannun, Billy Jun, Patrick LeGresley, Libby Lin, Sharan Narang, Andrew Ng, Sherjil Ozair, Ryan Prenger, Jonathan Raiman, Sanjeev Satheesh, David Seetapun, Shubho Sengupta, Yi Wang, Zhiqian Wang, Chong Wang, Bo Xiao, Dani Yogatama, Jun Zhan, Zhenyao Zhu | cs.CL | null | null | cs.CL | 20151208 | 20151208 | [] |
1512.02167 | 0 | 5 1 0 2 c e D 5 1
] V C . s c [
2 v 7 6 1 2 0 . 2 1 5 1 : v i X r a
# Simple Baseline for Visual Question Answering
Bolei Zhou1, Yuandong Tian2, Sainbayar Sukhbaatar2, Arthur Szlam2, and Rob Fergus2
1Massachusetts Institute of Technology 2Facebook AI Research
# Abstract
We describe a very simple bag-of-words baseline for visual question answering. This baseline concatenates the word features from the question and CNN features from the image to predict the answer. When evaluated on the challenging VQA dataset [2], it shows comparable performance to many recent approaches using recurrent neural networks. To explore the strength and weakness of the trained model, we also provide an interactive web demo1, and open-source code2.
# Introduction
Combining Natural Language Processing with Computer Vision for high-level scene interpretation is a recent trend, e.g., image captioning [10, 15, 7, 4]. These works have beneï¬ted from the rapid development of deep learning for visual recognition (object recognition [8] and scene recognition [20]), and have been made possible by the emergence of large image datasets and text corpus (e.g., [9]). Beyond image captioning, a natural next step is visual question answering (QA) [12, 2, 5]. | 1512.02167#0 | Simple Baseline for Visual Question Answering | We describe a very simple bag-of-words baseline for visual question
answering. This baseline concatenates the word features from the question and
CNN features from the image to predict the answer. When evaluated on the
challenging VQA dataset [2], it shows comparable performance to many recent
approaches using recurrent neural networks. To explore the strength and
weakness of the trained model, we also provide an interactive web demo and
open-source code. . | http://arxiv.org/pdf/1512.02167 | Bolei Zhou, Yuandong Tian, Sainbayar Sukhbaatar, Arthur Szlam, Rob Fergus | cs.CV, cs.CL | One comparison method's scores are put into the correct column, and a
new experiment of generating attention map is added | null | cs.CV | 20151207 | 20151215 | [
{
"id": "1511.05234"
},
{
"id": "1505.05612"
},
{
"id": "1511.07394"
},
{
"id": "1511.05960"
},
{
"id": "1511.05676"
},
{
"id": "1511.02799"
},
{
"id": "1511.02274"
},
{
"id": "1511.05756"
},
{
"id": "1511.06973"
},
{
"id": "1505.00468"
},
{
"id": "1512.04150"
},
{
"id": "1505.04467"
}
] |
1512.02167 | 1 | Compared with the image captioning task, in which an algorithm is required to generate free-form text description for a given image, visual QA can involve a wider range of knowledge and reasoning skills. A captioning algorithm has the liberty to pick the easiest relevant descriptions of the image, whereas as responding to a question needs to ï¬nd the correct answer for *that* question. Further- more, the algorithms for visual QA are required to answer all kinds of questions people might ask about the image, some of which might be relevant to the image contents, such as âwhat books are under the televisionâ and âwhat is the color of the boatâ, while others might require knowledge or reasoning beyond the image content, such as âwhy is the baby crying?â and âwhich chair is the most expensive?â. Building robust algorithms for visual QA that perform at near human levels would be an important step towards solving AI. | 1512.02167#1 | Simple Baseline for Visual Question Answering | We describe a very simple bag-of-words baseline for visual question
answering. This baseline concatenates the word features from the question and
CNN features from the image to predict the answer. When evaluated on the
challenging VQA dataset [2], it shows comparable performance to many recent
approaches using recurrent neural networks. To explore the strength and
weakness of the trained model, we also provide an interactive web demo and
open-source code. . | http://arxiv.org/pdf/1512.02167 | Bolei Zhou, Yuandong Tian, Sainbayar Sukhbaatar, Arthur Szlam, Rob Fergus | cs.CV, cs.CL | One comparison method's scores are put into the correct column, and a
new experiment of generating attention map is added | null | cs.CV | 20151207 | 20151215 | [
{
"id": "1511.05234"
},
{
"id": "1505.05612"
},
{
"id": "1511.07394"
},
{
"id": "1511.05960"
},
{
"id": "1511.05676"
},
{
"id": "1511.02799"
},
{
"id": "1511.02274"
},
{
"id": "1511.05756"
},
{
"id": "1511.06973"
},
{
"id": "1505.00468"
},
{
"id": "1512.04150"
},
{
"id": "1505.04467"
}
] |
1512.02167 | 2 | Recently, several papers have appeared on arXiv (after CVPRâ16 submission deadline) proposing neural network architectures for visual question answering, such as [13, 17, 5, 18, 16, 3, 11, 1]. Some of them are derived from the image captioning framework, in which the output of a recurrent neural network (e.g., LSTM [16, 11, 1]) applied to the question sentence is concatenated with visual features from VGG or other CNNs to feed a classiï¬er to predict the answer. Other models integrate visual attention mechanisms [17, 13, 3] and visualize how the network learns to attend the local image regions relevant to the content of the question.
Interestingly, we notice that in one of the earliest VQA papers [12], the simple baseline Bag-of- words + image feature (referred to as BOWIMG baseline) outperforms the LSTM-based models on a synthesized visual QA dataset built up on top of the image captions of COCO dataset [9]. For the recent much larger COCO VQA dataset [2], the BOWIMG baseline performs worse than the LSTM-based models [2].
1http://visualqa.csail.mit.edu 2https://github.com/metalbubble/VQAbaseline
1 | 1512.02167#2 | Simple Baseline for Visual Question Answering | We describe a very simple bag-of-words baseline for visual question
answering. This baseline concatenates the word features from the question and
CNN features from the image to predict the answer. When evaluated on the
challenging VQA dataset [2], it shows comparable performance to many recent
approaches using recurrent neural networks. To explore the strength and
weakness of the trained model, we also provide an interactive web demo and
open-source code. . | http://arxiv.org/pdf/1512.02167 | Bolei Zhou, Yuandong Tian, Sainbayar Sukhbaatar, Arthur Szlam, Rob Fergus | cs.CV, cs.CL | One comparison method's scores are put into the correct column, and a
new experiment of generating attention map is added | null | cs.CV | 20151207 | 20151215 | [
{
"id": "1511.05234"
},
{
"id": "1505.05612"
},
{
"id": "1511.07394"
},
{
"id": "1511.05960"
},
{
"id": "1511.05676"
},
{
"id": "1511.02799"
},
{
"id": "1511.02274"
},
{
"id": "1511.05756"
},
{
"id": "1511.06973"
},
{
"id": "1505.00468"
},
{
"id": "1512.04150"
},
{
"id": "1505.04467"
}
] |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.