title
string
paper_url
string
authors
list
type
string
primary_area
string
abstract
large_string
keywords
list
TL;DR
large_string
submission_number
int64
arxiv_id
string
arxiv_id_source
string
embedding
list
FractalNet: Ultra-Deep Neural Networks without Residuals
https://openreview.net/forum?id=S1VaB4cex
[ "Gustav Larsson", "Michael Maire", "Gregory Shakhnarovich" ]
Poster
null
We introduce a design strategy for neural network macro-architecture based on self-similarity. Repeated application of a simple expansion rule generates deep networks whose structural layouts are precisely truncated fractals. These networks contain interacting subpaths of different lengths, but do not include any pas...
[]
214
1605.07648
title_snapshot
[ 0.0028020611498504877, -0.04222184047102928, -0.008449847809970379, 0.03898642584681511, 0.04707934334874153, 0.05349467322230339, 0.003056851914152503, -0.012021655216813087, -0.03494293987751007, -0.06209219992160797, 0.00835649948567152, -0.013270910829305649, -0.05606921762228012, 0.00...
Deep Information Propagation
https://openreview.net/forum?id=H1W1UN9gg
[ "Samuel S. Schoenholz", "Justin Gilmer", "Surya Ganguli", "Jascha Sohl-Dickstein" ]
Poster
null
We study the behavior of untrained neural networks whose weights and biases are randomly distributed using mean field theory. We show the existence of depth scales that naturally limit the maximum depth of signal propagation through these random networks. Our main practical result is to show that random networks may be...
[ "Theory", "Deep learning" ]
We predict whether randomly initialized neural networks can be trained by studying whether or not information can travel through them.
215
1611.01232
title_snapshot
[ -0.02334374189376831, -0.00860085804015398, 0.006203089375048876, 0.04882216826081276, 0.05412111431360245, -0.003648771671578288, 0.034972500056028366, 0.027294298633933067, -0.029408052563667297, -0.03558632731437683, 0.014596599154174328, 0.0168912373483181, -0.04048809036612511, 0.0175...
Pruning Convolutional Neural Networks for Resource Efficient Inference
https://openreview.net/forum?id=SJGCiw5gl
[ "Pavlo Molchanov", "Stephen Tyree", "Tero Karras", "Timo Aila", "Jan Kautz" ]
Poster
null
We propose a new formulation for pruning convolutional kernels in neural networks to enable efficient inference. We interleave greedy criteria-based pruning with fine-tuning by backpropagation-a computationally efficient procedure that maintains good generalization in the pruned network. We propose a new criterion base...
[ "Deep learning", "Transfer Learning" ]
New approach for removing unnecessary conv neurons from network. Work is focused on how to estimate importance fast and efficiently by Taylor expantion.
427
1611.06440
title_snapshot
[ -0.016181135550141335, -0.021879585459828377, 0.005235955119132996, 0.020326729863882065, 0.03200485557317734, 0.05635978281497955, -0.00010284983000019565, 0.001322096330113709, -0.014027943834662437, -0.03339860215783119, -0.022821174934506416, 0.02215738594532013, -0.05711102485656738, ...
Recurrent Batch Normalization
https://openreview.net/forum?id=r1VdcHcxx
[ "Tim Cooijmans", "Nicolas Ballas", "César Laurent", "Çağlar Gülçehre", "Aaron Courville" ]
Poster
null
We propose a reparameterization of LSTM that brings the benefits of batch normalization to recurrent neural networks. Whereas previous works only apply batch normalization to the input-to-hidden transformation of RNNs, we demonstrate that it is both possible and beneficial to batch-normalize the hidden-to-hidden transi...
[ "Deep learning", "Optimization" ]
Make batch normalization work in recurrent neural networks
264
1603.09025
title_snapshot
[ -0.008008279837667942, -0.0391501821577549, -0.013395573012530804, 0.03159122169017792, 0.044425513595342636, 0.06350630521774292, 0.06264948099851608, 0.018175046890974045, -0.04007392004132271, -0.03075605072081089, -0.0029225782491266727, -0.016630807891488075, -0.03748273476958275, -0....
beta-VAE: Learning Basic Visual Concepts with a Constrained Variational Framework
https://openreview.net/forum?id=Sy2fzU9gl
[ "Irina Higgins", "Loic Matthey", "Arka Pal", "Christopher Burgess", "Xavier Glorot", "Matthew Botvinick", "Shakir Mohamed", "Alexander Lerchner" ]
Poster
null
Learning an interpretable factorised representation of the independent data generative factors of the world without supervision is an important precursor for the development of artificial intelligence that is able to learn and reason in the same way that humans do. We introduce beta-VAE, a new state-of-the-art framewor...
[]
We introduce beta-VAE, a new state-of-the-art framework for automated discovery of interpretable factorised latent representations from raw image data in a completely unsupervised manner.
291
null
null
[ 0.04051033779978752, 0.010797210969030857, -0.007159349974244833, 0.01957172341644764, 0.006776845548301935, 0.04706079140305519, 0.04098096862435341, -0.01248125173151493, -0.036848653107881546, -0.03551153838634491, -0.03807736560702324, 0.00703570106998086, -0.07151689380407333, 0.02527...
Words or Characters? Fine-grained Gating for Reading Comprehension
https://openreview.net/forum?id=B1hdzd5lg
[ "Zhilin Yang", "Bhuwan Dhingra", "Ye Yuan", "Junjie Hu", "William W. Cohen", "Ruslan Salakhutdinov" ]
Poster
null
Previous work combines word-level and character-level representations using concatenation or scalar weighting, which is suboptimal for high-level tasks like reading comprehension. We present a fine-grained gating mechanism to dynamically combine word-level and character-level representations based on properties of the ...
[ "Natural language processing", "Deep learning" ]
453
1611.01724
title_snapshot
[ -0.03244949132204056, 0.00630273949354887, 0.00742472754791379, 0.05760541558265686, 0.05780354142189026, -0.009384971112012863, 0.030297692865133286, 0.020800422877073288, -0.018986251205205917, -0.006954732816666365, 0.011434061452746391, 0.01880614086985588, -0.050095509737730026, -0.00...
DeepDSL: A Compilation-based Domain-Specific Language for Deep Learning
https://openreview.net/forum?id=Bks8cPcxe
[ "Tian Zhao", "Xiao Bing Huang", "Yu Cao" ]
Poster
null
In recent years, Deep Learning (DL) has found great success in domains such as multimedia understanding. However, the complex nature of multimedia data makes it difficult to develop DL-based software. The state-of-the-art tools, such as Caffe, TensorFlow, Torch7, and CNTK, while are successful in their applicable domai...
[ "Deep learning", "Applications", "Optimization" ]
DeepDSL(a DSL embedded in Scala) that compiles deep learning networks written in DeepDSL to Java source code, which runs on any GPU equipped machines with competitive efficiency as existing state-of-the-art tools (e.g. Caffe and Tensorflow)
414
1701.02284
title_snapshot
[ -0.0395638644695282, -0.017470866441726685, -0.030116582289338112, 0.022452116012573242, 0.04301020875573158, 0.012103196233510971, 0.0431685708463192, 0.012198708020150661, -0.03669032081961632, -0.034433647990226746, -0.02388487569987774, 0.008559700101613998, -0.06718263775110245, 0.025...
HyperNetworks
https://openreview.net/forum?id=rkpACe1lx
[ "David Ha", "Andrew M. Dai", "Quoc V. Le" ]
Poster
null
This work explores hypernetworks: an approach of using one network, also known as a hypernetwork, to generate the weights for another network. We apply hypernetworks to generate adaptive weights for recurrent networks. In this case, hypernetworks can be viewed as a relaxed form of weight-sharing across layers. In our ...
[ "Natural language processing", "Deep learning", "Supervised Learning" ]
We train a small RNN to generate weights for a larger RNN, and train the system end-to-end. We obtain state-of-the-art results on a variety of sequence modelling tasks.
8
1609.09106
title_snapshot
[ -0.0031016040593385696, -0.02184491790831089, 0.02617296576499939, 0.04339401796460152, 0.041929032653570175, 0.01734807901084423, 0.0127434516325593, -0.0019682690035551786, -0.04409720376133919, -0.05470503866672516, -0.007418554276227951, -0.03373544290661812, -0.05331123247742653, 0.00...
Capacity and Trainability in Recurrent Neural Networks
https://openreview.net/forum?id=BydARw9ex
[ "Jasmine Collins", "Jascha Sohl-Dickstein", "David Sussillo" ]
Poster
null
Two potential bottlenecks on the expressiveness of recurrent neural networks (RNNs) are their ability to store information about the task in their parameters, and to store information about the input history in their units. We show experimentally that all common RNN architectures achieve nearly the same per-task and pe...
[ "Deep learning" ]
447
1611.09913
title_snapshot
[ -0.026953324675559998, -0.04211743175983429, -0.021279096603393555, 0.04326792433857918, 0.04772789031267166, 0.062252216041088104, 0.04235372692346573, 0.011874412186443806, -0.04031846672296524, -0.022924037650227547, 0.005810367409139872, -0.006866415496915579, -0.05831291154026985, 0.0...
Recurrent Hidden Semi-Markov Model
https://openreview.net/forum?id=HJGODLqgx
[ "Hanjun Dai", "Bo Dai", "Yan-Ming Zhang", "Shuang Li", "Le Song" ]
Poster
null
Segmentation and labeling of high dimensional time series data has wide applications in behavior understanding and medical diagnosis. Due to the difficulty in obtaining the label information for high dimensional data, realizing this objective in an unsupervised way is highly desirable. Hidden Semi-Markov Model (HSMM) i...
[ "Deep learning", "Unsupervised Learning", "Structured prediction" ]
We propose to incorporate the RNN to model the generative process in Hidden Semi-Markov Model for unsupervised segmentation and labeling.
300
null
null
[ -0.014985370449721813, -0.0326092354953289, -0.03552589192986488, 0.018192986026406288, 0.0453731007874012, 0.042257364839315414, 0.04814569279551506, -0.0008494805661030114, -0.03551585599780083, -0.035369258373975754, -0.016380492597818375, -0.006933966185897589, -0.032212890684604645, 0...
Learning Curve Prediction with Bayesian Neural Networks
https://openreview.net/forum?id=S11KBYclx
[ "Aaron Klein", "Stefan Falkner", "Jost Tobias Springenberg", "Frank Hutter" ]
Poster
null
Different neural network architectures, hyperparameters and training protocols lead to different performances as a function of time. Human experts routinely inspect the resulting learning curves to quickly terminate runs with poor hyperparameter settings and thereby considerably speed up manual hyperparameter optimizat...
[ "Deep learning", "Applications" ]
We present a general probabilistic method based on Bayesian neural networks to predit learning curves of iterative machine learning methods.
488
null
null
[ -0.010762207210063934, 0.01829088106751442, -0.020251790061593056, 0.034658052027225494, 0.03642205148935318, 0.023234358057379723, 0.033934466540813446, -0.024128830060362816, -0.022008676081895828, -0.028051016852259636, -0.0011471665930002928, 0.0284996647387743, -0.015084500424563885, ...
A Simple but Tough-to-Beat Baseline for Sentence Embeddings
https://openreview.net/forum?id=SyK00v5xx
[ "Sanjeev Arora", "Yingyu Liang", "Tengyu Ma" ]
Poster
null
The success of neural network methods for computing word embeddings has motivated methods for generating semantic embeddings of longer pieces of text, such as sentences and paragraphs. Surprisingly, Wieting et al (ICLR'16) showed that such complicated methods are outperformed, especially in out-of-domain (transfer lea...
[ "Natural language processing", "Unsupervised Learning" ]
A simple unsupervised method for sentence embedding that can get results comparable to sophisticated models like RNN's and LSTM's
448
null
null
[ -0.009285284206271172, -0.04240872338414192, -0.021255003288388252, 0.05670776590704918, 0.0463816337287426, 0.031704094260931015, 0.04120606556534767, -0.0011009029112756252, 0.013042313046753407, -0.018993236124515533, 0.006177844945341349, 0.020542575046420097, -0.051647383719682693, -0...
Learning to Repeat: Fine Grained Action Repetition for Deep Reinforcement Learning
https://openreview.net/forum?id=B1GOWV5eg
[ "Sahil Sharma", "Aravind S. Lakshminarayanan", "Balaraman Ravindran" ]
Poster
null
Reinforcement Learning algorithms can learn complex behavioral patterns for sequential decision making tasks wherein an agent interacts with an environment and acquires feedback in the form of rewards sampled from it. Traditionally, such algorithms make decisions, i.e., select actions to execute, at every single time s...
[ "Deep learning", "Reinforcement Learning" ]
Framework for temporal abstractions in policy space by learning to repeat actions
206
1702.06054
title_snapshot
[ -0.001403458183631301, -0.028492605313658714, -0.005886815022677183, 0.06618496030569077, 0.042247820645570755, 0.026696782559156418, 0.018581818789243698, 0.027407752349972725, -0.023356210440397263, -0.034119587391614914, -0.004712643101811409, 0.018208852037787437, -0.05646364390850067, ...
Improving Neural Language Models with a Continuous Cache
https://openreview.net/forum?id=B184E5qee
[ "Edouard Grave", "Armand Joulin", "Nicolas Usunier" ]
Poster
null
We propose an extension to neural network language models to adapt their prediction to the recent history. Our model is a simplified version of memory augmented networks, which stores past hidden activations as memory and accesses them through a dot product with the current hidden activation. This mechanism is very eff...
[ "Natural language processing" ]
534
1612.04426
title_snapshot
[ -0.03600415959954262, 0.006466921418905258, -0.025881370529532433, 0.04000946134328842, 0.027536923065781593, 0.03818697854876518, 0.03909873589873314, 0.040289994329214096, -0.04760206118226051, -0.019581949338316917, -0.004612192045897245, 0.01946883834898472, -0.05804569274187088, -0.01...
Making Neural Programming Architectures Generalize via Recursion
https://openreview.net/forum?id=BkbY4psgg
[ "Jonathon Cai", "Richard Shin", "Dawn Song" ]
Oral
null
Empirically, neural networks that attempt to learn programs from data have exhibited poor generalizability. Moreover, it has traditionally been difficult to reason about the behavior of these models beyond a certain level of input complexity. In order to address these issues, we propose augmenting neural architectures ...
[ "Deep learning" ]
597
1704.06611
title_snapshot
[ -0.02798757143318653, -0.0202134121209383, -0.021379049867391586, 0.03035581484436989, 0.03538363799452782, 0.04287950322031975, 0.02036716230213642, 0.01675122044980526, -0.04340814799070358, -0.021678084507584572, -0.00987180881202221, 0.023647230118513107, -0.07369501143693924, -0.00845...
Snapshot Ensembles: Train 1, Get M for Free
https://openreview.net/forum?id=BJYwwY9ll
[ "Gao Huang", "Yixuan Li", "Geoff Pleiss", "Zhuang Liu", "John E. Hopcroft", "Kilian Q. Weinberger" ]
Poster
null
Ensembles of neural networks are known to be much more robust and accurate than individual networks. However, training multiple deep networks for model averaging is computationally expensive. In this paper, we propose a method to obtain the seemingly contradictory goal of ensembling multiple neural networks at no addit...
[ "Deep learning", "Computer vision" ]
501
1704.00109
title_snapshot
[ -0.000775135587900877, -0.026747606694698334, -0.0200512632727623, 0.07063612341880798, 0.03974931687116623, 0.03542324900627136, 0.017027344554662704, -0.0053722751326859, -0.032530948519706726, -0.05959487333893776, -0.0032920034136623144, 0.007801042404025793, -0.08013886958360672, -0.0...
Soft Weight-Sharing for Neural Network Compression
https://openreview.net/forum?id=HJGwcKclx
[ "Karen Ullrich", "Edward Meeds", "Max Welling" ]
Poster
null
The success of deep learning in numerous application domains created the desire to run and train them on mobile devices. This however, conflicts with their computationally, memory and energy intense nature, leading to a growing interest in compression. Recent work by Han et al. (2016) propose a pipeline that involves r...
[ "Deep learning", "Optimization" ]
We use soft weight-sharing to compress neural network weights.
516
1702.04008
title_snapshot
[ -0.030286582186818123, -0.0453709214925766, -0.023840930312871933, 0.013293970376253128, 0.0502532534301281, 0.054117921739816666, -0.007386505603790283, -0.0025538585614413023, -0.033028751611709595, -0.05084609612822533, -0.02152002416551113, -0.006041432730853558, -0.05949435383081436, ...
Learning to Perform Physics Experiments via Deep Reinforcement Learning
https://openreview.net/forum?id=r1nTpv9eg
[ "Misha Denil", "Pulkit Agrawal", "Tejas D Kulkarni", "Tom Erez", "Peter Battaglia", "Nando de Freitas" ]
Poster
null
When encountering novel objects, humans are able to infer a wide range of physical properties such as mass, friction and deformability by interacting with them in a goal driven way. This process of active interaction is in the same spirit as a scientist performing experiments to discover hidden facts. Recent advances i...
[ "Deep learning", "Reinforcement Learning" ]
We train agents to conduct experiments in interactive simulated physical environments.
443
1611.01843
title_snapshot
[ -0.02552218548953533, -0.00921039842069149, -0.024251658469438553, 0.05581025779247284, 0.05049551650881767, 0.012096066027879715, 0.011894104070961475, -0.011544630862772465, -0.02398267388343811, -0.04032301530241966, -0.03401294723153114, 0.025524521246552467, -0.040224459022283554, 0.0...
Query-Reduction Networks for Question Answering
https://openreview.net/forum?id=B1MRcPclx
[ "Minjoon Seo", "Sewon Min", "Ali Farhadi", "Hannaneh Hajishirzi" ]
Poster
null
In this paper, we study the problem of question answering when reasoning over multiple facts is required. We propose Query-Reduction Network (QRN), a variant of Recurrent Neural Network (RNN) that effectively handles both short-term (local) and long-term (global) sequential dependencies to reason over multiple facts. Q...
[ "Natural language processing", "Deep learning" ]
421
1606.04582
title_snapshot
[ -0.015015866607427597, -0.0220854040235281, 0.00768026290461421, 0.050040166825056076, 0.032431893050670624, 0.037040021270513535, 0.005438311491161585, 0.0009256114135496318, -0.03532055765390396, -0.006237686611711979, -0.03658333793282509, 0.03502444177865982, -0.058274783194065094, -0....
Adversarial Machine Learning at Scale
https://openreview.net/forum?id=BJm4T4Kgx
[ "Alexey Kurakin", "Ian J. Goodfellow", "Samy Bengio" ]
Poster
null
Adversarial examples are malicious inputs designed to fool machine learning models. They often transfer from one model to another, allowing attackers to mount black box attacks without knowledge of the target model's parameters. Adversarial training is the process of explicitly training a model on adversarial examples,...
[ "Computer vision", "Supervised Learning" ]
92
1611.01236
title_snapshot
[ -0.004578341729938984, -0.03645160049200058, -0.01678018644452095, 0.04100647568702698, 0.03269682824611664, 0.008032187819480896, 0.03822479024529457, -0.027540210634469986, -0.027715791016817093, -0.02848276123404503, -0.03457560017704964, -0.010103161446750164, -0.07625837624073029, 0.0...
Variational Recurrent Adversarial Deep Domain Adaptation
https://openreview.net/forum?id=rk9eAFcxg
[ "Sanjay Purushotham", "Wilka Carvalho", "Tanachat Nilanon", "Yan Liu" ]
Poster
null
We study the problem of learning domain invariant representations for time series data while transferring the complex temporal latent dependencies between the domains. Our model termed as Variational Recurrent Adversarial Deep Domain Adaptation (VRADA) is built atop a variational recurrent neural network (VRNN) and tra...
[ "Deep learning", "Transfer Learning" ]
We propose Variational Recurrent Adversarial Deep Domain Adaptation approach to capture and transfer temporal latent dependencies in multivariate time-series data
525
null
null
[ 0.0129028195515275, 0.014134238474071026, 0.001872530672699213, 0.04149048402905464, 0.04712314531207085, 0.04429631680250168, 0.05715150758624077, -0.003984498325735331, -0.022113079205155373, -0.01407534722238779, -0.02501703053712845, 0.0088103748857975, -0.055515363812446594, 0.0350772...
Discrete Variational Autoencoders
https://openreview.net/forum?id=ryMxXPFex
[ "Jason Tyler Rolfe" ]
Poster
null
Probabilistic models with discrete latent variables naturally capture datasets composed of discrete classes. However, they are difficult to train efficiently, since backpropagation through discrete variables is generally not possible. We present a novel method to train a class of probabilistic models with discrete late...
[ "Deep learning", "Unsupervised Learning" ]
We present a novel method to train a class of probabilistic models with discrete latent variables using the variational autoencoder framework, including backpropagation through the discrete latent variables.
103
1609.02200
title_snapshot
[ 0.015457363799214363, -0.005240717437118292, -0.018622271716594696, 0.06715838611125946, 0.04204848036170006, 0.053734924644231796, 0.04273129999637604, 0.005684537347406149, -0.04862487316131592, -0.043862245976924896, -0.030589329078793526, -0.0036819945089519024, -0.05539237707853317, 0...
Towards Principled Methods for Training Generative Adversarial Networks
https://openreview.net/forum?id=Hk4_qw5xe
[ "Martin Arjovsky", "Leon Bottou" ]
Oral
null
The goal of this paper is not to introduce a single algorithm or method, but to make theoretical steps towards fully understanding the training dynamics of gen- erative adversarial networks. In order to substantiate our theoretical analysis, we perform targeted experiments to verify our assumptions, illustrate our clai...
[]
We introduce a theory about generative adversarial networks and their issues.
415
1701.04862
title_snapshot
[ -0.012729175388813019, -0.04041435196995735, -0.004548934753984213, 0.05219271406531334, 0.01680326648056507, 0.037562295794487, 0.0069841439835727215, 0.00929250381886959, -0.01008650939911604, -0.048306435346603394, -0.025547372177243233, 0.007903244346380234, -0.0705731064081192, 0.0066...
Sparsely-Connected Neural Networks: Towards Efficient VLSI Implementation of Deep Neural Networks
https://openreview.net/forum?id=r1fYuytex
[ "Arash Ardakani", "Carlo Condo", "Warren J. Gross" ]
Poster
null
Recently deep neural networks have received considerable attention due to their ability to extract and represent high-level abstractions in data sets. Deep neural networks such as fully-connected and convolutional neural networks have shown excellent performance on a wide range of recognition and classification tasks. ...
[ "Deep learning", "Applications", "Optimization" ]
We show that the number of connections in fully-connected networks can be reduced by up to 90% while improving the accuracy performance.
71
1611.01427
title_snapshot
[ 0.004229061771184206, -0.018427183851599693, -0.03019680082798004, 0.04873514920473099, 0.05596398562192917, 0.03274135664105415, 0.014242874458432198, 0.010833094827830791, -0.03488276153802872, -0.04031495377421379, 0.026721671223640442, -0.03452891856431961, -0.051507703959941864, 0.017...
Recurrent Mixture Density Network for Spatiotemporal Visual Attention
https://openreview.net/forum?id=SJRpRfKxx
[ "Loris Bazzani", "Hugo Larochelle", "Lorenzo Torresani" ]
Poster
null
In many computer vision tasks, the relevant information to solve the problem at hand is mixed to irrelevant, distracting information. This has motivated researchers to design attentional models that can dynamically focus on parts of images or videos that are salient, e.g., by down-weighting irrelevant pixels. In this w...
[ "Computer vision", "Deep learning", "Applications" ]
83
1603.08199
title_snapshot
[ 0.04512457177042961, 0.004068369045853615, 0.011751151643693447, 0.03936369717121124, 0.011843468062579632, 0.032625723630189896, 0.025837522000074387, 0.05551409721374512, -0.026605859398841858, -0.024460626766085625, -0.02470860816538334, 0.001871817046776414, -0.05007380619645119, -0.00...
Efficient Representation of Low-Dimensional Manifolds using Deep Networks
https://openreview.net/forum?id=BJ3filKll
[ "Ronen Basri", "David W. Jacobs" ]
Poster
null
We consider the ability of deep neural networks to represent data that lies near a low-dimensional manifold in a high-dimensional space. We show that deep networks can efficiently extract the intrinsic, low-dimensional coordinates of such data. Specifically we show that the first two layers of a deep network can exac...
[ "Theory", "Deep learning" ]
We show constructively that deep networks can learn to represent manifold data efficiently
75
1602.04723
title_snapshot
[ -0.029029982164502144, -0.019187530502676964, 0.001791620859876275, 0.03552861884236336, 0.028599843382835388, 0.04800887405872345, 0.017093544825911522, -0.00032732257386669517, -0.020924588665366173, -0.05161271244287491, -0.00382341630756855, -0.026583248749375343, -0.06952325254678726, ...
Reinforcement Learning through Asynchronous Advantage Actor-Critic on a GPU
https://openreview.net/forum?id=r1VGvBcxl
[ "Mohammad Babaeizadeh", "Iuri Frosio", "Stephen Tyree", "Jason Clemons", "Jan Kautz" ]
Poster
null
We introduce a hybrid CPU/GPU version of the Asynchronous Advantage Actor-Critic (A3C) algorithm, currently the state-of-the-art method in reinforcement learning for various gaming tasks. We analyze its computational traits and concentrate on aspects critical to leveraging the GPU's computational power. We introduce a ...
[ "Reinforcement Learning" ]
Implementation and analysis of the computational aspect of a GPU version of the Asynchronous Advantage Actor-Critic (A3C) algorithm
255
1611.06256
title_snapshot
[ -0.02263174019753933, -0.035026393830776215, -0.001745461137033999, 0.041088417172431946, 0.006849992088973522, 0.030656889081001282, -0.001247609150595963, 0.025758808478713036, -0.03563860431313515, -0.049539413303136826, -0.005177087150514126, 0.022347137331962585, -0.07400650531053543, ...
Topology and Geometry of Half-Rectified Network Optimization
https://openreview.net/forum?id=Bk0FWVcgx
[ "C. Daniel Freeman", "Joan Bruna" ]
Poster
null
The loss surface of deep neural networks has recently attracted interest in the optimization and machine learning communities as a prime example of high-dimensional non-convex problem. Some insights were recently gained using spin glass models and mean-field approximations, but at the expense of strongly simplifying...
[ "Theory", "Deep learning" ]
We provide theoretical, algorithmical and experimental results concerning the optimization landscape of deep neural networks
207
1611.01540
title_snapshot
[ -0.014354401268064976, -0.03302352502942085, -0.002913429168984294, 0.02312021143734455, 0.02083822712302208, 0.025946088135242462, 0.02377471886575222, 0.022402092814445496, -0.02268163673579693, -0.050825607031583786, 0.001427177689038217, -0.02735699526965618, -0.06619090586900711, 0.00...
Tighter bounds lead to improved classifiers
https://openreview.net/forum?id=HyAbMKwxe
[ "Nicolas Le Roux" ]
Poster
null
The standard approach to supervised classification involves the minimization of a log-loss as an upper bound to the classification error. While this is a tight bound early on in the optimization, it overemphasizes the influence of incorrectly classified examples far from the decision boundary. Updating the upper bound ...
[]
46
1606.09202
title_snapshot
[ -0.01979924365878105, 0.01996639557182789, -0.0213386919349432, 0.036089107394218445, 0.07278487831354141, 0.011421958915889263, 0.010965228080749512, -0.06637303531169891, -0.04881169646978378, -0.00854208879172802, -0.031760070472955704, 0.010588408447802067, -0.08861850202083588, 0.0106...
Zoneout: Regularizing RNNs by Randomly Preserving Hidden Activations
https://openreview.net/forum?id=rJqBEPcxe
[ "David Krueger", "Tegan Maharaj", "Janos Kramar", "Mohammad Pezeshki", "Nicolas Ballas", "Nan Rosemary Ke", "Anirudh Goyal", "Yoshua Bengio", "Aaron Courville", "Christopher Pal" ]
Poster
null
We propose zoneout, a novel method for regularizing RNNs. At each timestep, zoneout stochastically forces some hidden units to maintain their previous values. Like dropout, zoneout uses random noise to train a pseudo-ensemble, improving generalization. But by preserving instead of dropping hidden units, gradient inform...
[ "Deep learning" ]
Zoneout is like dropout (for RNNs) but uses identity masks instead of zero masks
378
1606.01305
title_snapshot
[ 0.02198285423219204, -0.041039321571588516, -0.020146381109952927, 0.04668615013360977, 0.047431718558073044, 0.05758223310112953, 0.054429538547992706, 0.01355683896690607, -0.05495799705386162, -0.03439493104815483, -0.009777083061635494, -0.010686369612812996, -0.03035227209329605, -0.0...
Learning End-to-End Goal-Oriented Dialog
https://openreview.net/forum?id=S1Bb3D5gg
[ "Antoine Bordes", "Y-Lan Boureau", "Jason Weston" ]
Oral
null
Traditional dialog systems used in goal-oriented applications require a lot of domain-specific handcrafting, which hinders scaling up to new domains. End- to-end dialog systems, in which all components are trained from the dialogs themselves, escape this limitation. But the encouraging success recently obtained in chit...
[]
A new open dataset and testbed for training and evaluating end-to-end dialog systems in goal-oriented scenarios.
428
1605.07683
title_snapshot
[ -0.011975504457950592, -0.029004542157053947, -0.0007519219652749598, 0.0035811495035886765, 0.04417033866047859, 0.005128578282892704, 0.02009429596364498, 0.02254198119044304, -0.001997424988076091, -0.032994259148836136, -0.060798078775405884, 0.04040662944316864, -0.06685785204172134, ...
An Information-Theoretic Framework for Fast and Robust Unsupervised Learning via Neural Population Infomax
https://openreview.net/forum?id=SkYbF1slg
[ "Wentao Huang", "Kechen Zhang" ]
Poster
null
A framework is presented for unsupervised learning of representations based on infomax principle for large-scale neural populations. We use an asymptotic approximation to the Shannon's mutual information for a large neural population to demonstrate that a good initial approximation to the global information-theoretic o...
[ "Unsupervised Learning", "Theory", "Deep learning" ]
We present a novel information-theoretic framework for fast and robust unsupervised Learning via information maximization for neural population coding.
549
1611.01886
title_snapshot
[ -0.02141772396862507, -0.023859020322561264, 0.005112148821353912, 0.02916884422302246, 0.017556536942720413, 0.03480641171336174, 0.0047037736512720585, -0.0066785551607608795, -0.020259464159607887, -0.02636410854756832, -0.008947578258812428, 0.011607752181589603, -0.07286562770605087, ...
Third Person Imitation Learning
https://openreview.net/forum?id=B16dGcqlx
[ "Bradly C Stadie", "Pieter Abbeel", "Ilya Sutskever" ]
Poster
null
Reinforcement learning (RL) makes it possible to train agents capable of achieving sophisticated goals in complex and uncertain environments. A key difficulty in reinforcement learning is specifying a reward function for the agent to optimize. Traditionally, imitation learning in RL has been used to overcome this probl...
[]
Agent watches another agent at a different camera angle completing the task and learns via raw pixels how to imitate.
531
1703.01703
title_snapshot
[ -0.03642874211072922, -0.018766921013593674, -0.0026679763104766607, 0.039096053689718246, 0.07129007577896118, -0.0023901243694126606, 0.015480812638998032, 0.0059929355047643185, -0.045799918472766876, -0.022141899913549423, -0.026426978409290314, 0.01992361806333065, -0.06348481774330139,...
Deep Variational Bayes Filters: Unsupervised Learning of State Space Models from Raw Data
https://openreview.net/forum?id=HyTqHL5xg
[ "Maximilian Karl", "Maximilian Soelch", "Justin Bayer", "Patrick van der Smagt" ]
Poster
null
We introduce Deep Variational Bayes Filters (DVBF), a new method for unsupervised learning and identification of latent Markovian state space models. Leveraging recent advances in Stochastic Gradient Variational Bayes, DVBF can overcome intractable inference distributions via variational inference. Thus, it can handle ...
[ "Deep learning", "Unsupervised Learning" ]
296
1605.06432
title_snapshot
[ 0.011129962280392647, 0.013354146853089333, -0.013816110789775848, 0.02684675343334675, 0.06679201871156693, 0.034612350165843964, 0.014980468899011612, 0.014026962220668793, -0.04433475434780121, -0.046864986419677734, -0.0048430440947413445, 0.01291930116713047, -0.07525666058063507, 0.0...
Learning Invariant Representations Of Planar Curves
https://openreview.net/forum?id=BymIbLKgl
[ "Gautam Pai", "Aaron Wetzler", "Ron Kimmel" ]
Poster
null
We propose a metric learning framework for the construction of invariant geometric functions of planar curves for the Euclidean and Similarity group of transformations. We leverage on the representational power of convolutional neural networks to compute these geometric quantities. In comparison with axiomatic construc...
[ "Computer vision", "Deep learning", "Supervised Learning", "Applications" ]
97
1611.07807
title_snapshot
[ 0.002347154077142477, -0.020089559257030487, 0.014456418342888355, 0.02051318809390068, 0.02940157614648342, 0.054442793130874634, -0.006129163783043623, 0.0020399801433086395, -0.028245145455002785, -0.03272290155291557, -0.04003356769680977, -0.007479135878384113, -0.07459751516580582, 0...
Density estimation using Real NVP
https://openreview.net/forum?id=HkpbnH9lx
[ "Laurent Dinh", "Jascha Sohl-Dickstein", "Samy Bengio" ]
Poster
null
Unsupervised learning of probabilistic models is a central yet challenging problem in machine learning. Specifically, designing models with tractable learning, sampling, inference and evaluation is crucial in solving this task. We extend the space of such models using real-valued non-volume preserving (real NVP) transf...
[ "Deep learning", "Unsupervised Learning" ]
Efficient invertible neural networks for density estimation and generation
269
1605.08803
title_snapshot
[ 0.0006893584504723549, -0.029946761205792427, -0.0062360623851418495, 0.03980756923556328, 0.057993751019239426, 0.03693603724241257, 0.008373918011784554, -0.005333638284355402, -0.03720288351178169, -0.05637889727950096, -0.01287204772233963, -0.002768076490610838, -0.06626217067241669, ...
Metacontrol for Adaptive Imagination-Based Optimization
https://openreview.net/forum?id=Bk8BvDqex
[ "Jessica B. Hamrick", "Andrew J. Ballard", "Razvan Pascanu", "Oriol Vinyals", "Nicolas Heess", "Peter W. Battaglia" ]
Poster
null
Many machine learning systems are built to solve the hardest examples of a particular task, which often makes them large and expensive to run---especially with respect to the easier examples, which might require much less computation. For an agent with a limited computational budget, this "one-size-fits-all" approach m...
[ "Deep learning", "Reinforcement Learning", "Optimization" ]
We present a "metacontroller" neural architecture which can adaptively decide how long to run an model-based online optimization procedure for, and which models to use during the optimization.
392
1705.02670
title_snapshot
[ -0.030017312616109848, -0.014189819805324078, -0.012055113911628723, 0.037825535982847214, 0.05839778110384941, 0.01714496500790119, 0.01396077498793602, -0.005601359531283379, -0.03496842458844185, -0.039710883051157, -0.03253583610057831, 0.017107859253883362, -0.0680980309844017, -0.002...
The Neural Noisy Channel
https://openreview.net/forum?id=SJ25-B5eg
[ "Lei Yu", "Phil Blunsom", "Chris Dyer", "Edward Grefenstette", "Tomas Kocisky" ]
Poster
null
We formulate sequence to sequence transduction as a noisy channel decoding problem and use recurrent neural networks to parameterise the source and channel models. Unlike direct models which can suffer from explaining-away effects during training, noisy channel models must produce outputs that explain their inputs, and...
[ "Natural language processing", "Deep learning", "Semi-Supervised Learning" ]
We formulate sequence to sequence transduction as a noisy channel decoding problem and use recurrent neural networks to parameterise the source and channel models.
240
1611.02554
title_snapshot
[ 0.013090447522699833, 0.003571406239643693, -0.029193362221121788, 0.032371047884225845, 0.02665780484676361, 0.04433203116059303, 0.05517467483878136, 0.01627173461019993, -0.009594911709427834, -0.030858946964144707, -0.02269505336880684, 0.03991491347551346, -0.053986482322216034, -0.00...
A Compare-Aggregate Model for Matching Text Sequences
https://openreview.net/forum?id=HJTzHtqee
[ "Shuohang Wang", "Jing Jiang" ]
Poster
null
Many NLP tasks including machine comprehension, answer selection and text entailment require the comparison between sequences. Matching the important units between sequences is a key to solve these problems. In this paper, we present a general "compare-aggregate" framework that performs word-level matching followed by ...
[ "Natural language processing", "Deep learning" ]
A general "compare-aggregate" framework that performs word-level matching followed by aggregation using Convolutional Neural Networks
481
1611.01747
title_snapshot
[ -0.012961289845407009, -0.025770245119929314, -0.015939710661768913, 0.04508171230554581, 0.031539563089609146, 0.028404394164681435, 0.023393146693706512, 0.04163196310400963, -0.008145947940647602, -0.012098630890250206, 0.012461354024708271, 0.04953652247786522, -0.06658904254436493, -0...
Calibrating Energy-based Generative Adversarial Networks
https://openreview.net/forum?id=SyxeqhP9ll
[ "Zihang Dai", "Amjad Almahairi", "Philip Bachman", "Eduard Hovy", "Aaron Courville" ]
Poster
null
In this paper, we propose to equip Generative Adversarial Networks with the ability to produce direct energy estimates for samples. Specifically, we propose a flexible adversarial training framework, and prove this framework not only ensures the generator converges to the true data distribution, but also enables the di...
[ "Deep learning" ]
434
1702.01691
title_snapshot
[ -0.0013052296126261353, -0.018521735444664955, -0.006747453007847071, 0.057548586279153824, 0.0199624914675951, 0.024557216092944145, 0.002023606328293681, 0.0024389747995883226, -0.020682798698544502, -0.05346806347370148, 0.012334254570305347, -0.006630267482250929, -0.09139565378427505, ...
Learning a Natural Language Interface with Neural Programmer
https://openreview.net/forum?id=ry2YOrcge
[ "Arvind Neelakantan", "Quoc V. Le", "Martin Abadi", "Andrew McCallum", "Dario Amodei" ]
Poster
null
Learning a natural language interface for database tables is a challenging task that involves deep language understanding and multi-step reasoning. The task is often approached by mapping natural language queries to logical forms or programs that provide the desired response when executed on the database. To our knowle...
[ "Natural language processing", "Deep learning" ]
To our knowledge, this paper presents the first weakly supervised, end-to-end neural network model to induce programs on a real-world dataset.
260
1611.08945
title_snapshot
[ -0.016580140218138695, -0.026825644075870514, -0.014923554845154285, 0.031488001346588135, 0.04122506082057953, 0.02931729145348072, 0.011268840171396732, 0.009888746775686741, -0.015855461359024048, 0.003987855277955532, -0.03613397851586342, 0.061380207538604736, -0.0627630278468132, -0....
Learning to superoptimize programs
https://openreview.net/forum?id=r1rz6U5lg
[ "Rudy Bunel", "Alban Desmaison", "M. Pawan Kumar", "Philip H.S. Torr", "Pushmeet Kohli" ]
Poster
null
Code super-optimization is the task of transforming any given program to a more efficient version while preserving its input-output behaviour. In some sense, it is similar to the paraphrase problem from natural language processing where the intention is to change the syntax of an utterance without changing its semant...
[]
329
1611.01787
title_snapshot
[ -0.03007480315864086, -0.0050798882730305195, -0.024445734918117523, 0.020688027143478394, 0.057354673743247986, 0.039817970246076584, 0.03510047122836113, -0.019342394545674324, -0.02226659655570984, -0.024957701563835144, -0.03391652554273605, 0.01563998870551586, -0.05873984098434448, -...
Optimal Binary Autoencoding with Pairwise Correlations
https://openreview.net/forum?id=ryelgY5eg
[ "Akshay Balsubramani" ]
Poster
null
We formulate learning of a binary autoencoder as a biconvex optimization problem which learns from the pairwise correlations between encoded and decoded bits. Among all possible algorithms that use this information, ours finds the autoencoder that reconstructs its inputs with worst-case optimal loss. The optimal decode...
[ "Theory", "Unsupervised Learning", "Games" ]
Efficient biconvex learning of binary autoencoders, using pairwise correlations between encodings and decodings, is strongly optimal.
469
1611.02268
title_snapshot
[ 0.0026092645712196827, -0.010223538614809513, -0.010668701492249966, 0.05701250955462456, 0.024104798212647438, 0.054090309888124466, 0.037395935505628586, 0.009876862168312073, -0.03250541538000107, -0.030420836061239243, -0.030518047511577606, -0.011122471652925014, -0.06929703056812286, ...
Neural Program Lattices
https://openreview.net/forum?id=HJjiFK5gx
[ "Chengtao Li", "Daniel Tarlow", "Alexander L. Gaunt", "Marc Brockschmidt", "Nate Kushman" ]
Poster
null
We propose the Neural Program Lattice (NPL), a neural network that learns to perform complex tasks by composing low-level programs to express high-level programs. Our starting point is the recent work on Neural Programmer-Interpreters (NPI), which can only learn from strong supervision that contains the whole hierarchy...
[ "Deep learning", "Semi-Supervised Learning" ]
513
null
null
[ -0.0034975274465978146, -0.00019779743161052465, -0.05695883184671402, 0.036324113607406616, 0.02370787225663662, 0.044769156724214554, 0.016369208693504333, -0.009342921897768974, -0.03478635102510452, -0.021848764270544052, -0.012821843847632408, 0.016245035454630852, -0.06950867921113968,...
Adversarially Learned Inference
https://openreview.net/forum?id=B1ElR4cgg
[ "Vincent Dumoulin", "Ishmael Belghazi", "Ben Poole", "Alex Lamb", "Martin Arjovsky", "Olivier Mastropietro", "Aaron Courville" ]
Poster
null
We introduce the adversarially learned inference (ALI) model, which jointly learns a generation network and an inference network using an adversarial process. The generation network maps samples from stochastic latent variables to the data space while the inference network maps training examples in data space to the sp...
[ "Computer vision", "Deep learning", "Unsupervised Learning", "Semi-Supervised Learning" ]
We present and adverserially trained generative model with an inference network. Samples quality is high. Competitive semi-supervised results are achieved.
222
1606.00704
title_snapshot
[ -0.005163659807294607, -0.01375341322273016, -0.008444327861070633, 0.06254495680332184, 0.018796231597661972, 0.0046888054348528385, 0.027887016534805298, -0.02033914253115654, -0.02026638388633728, -0.020061075687408447, -0.011734829284250736, 0.011272587813436985, -0.055928997695446014, ...
Frustratingly Short Attention Spans in Neural Language Modeling
https://openreview.net/forum?id=ByIAPUcee
[ "Michał Daniluk", "Tim Rocktäschel", "Johannes Welbl", "Sebastian Riedel" ]
Poster
null
Current language modeling architectures often use recurrent neural networks. Recently, various methods for incorporating differentiable memory into these architectures have been proposed. When predicting the next token, these models query information from a memory of the recent history and thus can facilitate learning ...
[ "Natural language processing", "Deep learning" ]
We investigate various memory-augmented neural language models and compare them against state-of-the-art architectures.
301
1702.04521
title_snapshot
[ -0.027916887775063515, -0.005707049276679754, -0.01806819811463356, 0.02648567222058773, 0.022673090919852257, 0.03519554063677788, 0.03315079212188721, 0.04095929116010666, -0.053042106330394745, 0.00316837546415627, -0.02195555530488491, 0.033120691776275635, -0.04464979097247124, 0.0153...
Steerable CNNs
https://openreview.net/forum?id=rJQKYt5ll
[ "Taco S. Cohen", "Max Welling" ]
Poster
null
It has long been recognized that the invariance and equivariance properties of a representation are critically important for success in many vision tasks. In this paper we present Steerable Convolutional Neural Networks, an efficient and flexible class of equivariant convolutional networks. We show that steerable CNNs ...
[]
511
1612.08498
title_snapshot
[ 0.01732438988983631, -0.011047802865505219, 0.0422079898416996, 0.01256733201444149, 0.023032592609524727, 0.02466350980103016, 0.003048564074561, 0.03965871408581734, -0.0047945003025233746, -0.03937910124659538, -0.024521619081497192, 0.011891485191881657, -0.04860340431332588, 0.0219077...
Multi-view Recurrent Neural Acoustic Word Embeddings
https://openreview.net/forum?id=rJxDkvqee
[ "Wanjia He", "Weiran Wang", "Karen Livescu" ]
Poster
null
Recent work has begun exploring neural acoustic word embeddings–fixed dimensional vector representations of arbitrary-length speech segments corresponding to words. Such embeddings are applicable to speech retrieval and recognition tasks, where reasoning about whole words may make it possible to avoid ambiguous sub-wor...
[]
347
1611.04496
title_snapshot
[ -0.025309832766652107, -0.003332686610519886, -0.0072549255564808846, 0.03901251032948494, 0.013241980224847794, 0.05707455426454544, 0.05331602692604065, 0.01883142627775669, -0.02148820087313652, -0.03489295393228531, -0.006438673008233309, 0.01625780574977398, -0.059811774641275406, -0....
Transfer Learning for Sequence Tagging with Hierarchical Recurrent Networks
https://openreview.net/forum?id=ByxpMd9lx
[ "Zhilin Yang", "Ruslan Salakhutdinov", "William W. Cohen" ]
Poster
null
Recent papers have shown that neural networks obtain state-of-the-art performance on several different sequence tagging tasks. One appealing property of such systems is their generality, as excellent performance can be achieved with a unified architecture and without task-specific feature engineering. However, it is u...
[ "Natural language processing", "Deep learning", "Transfer Learning" ]
454
1703.06345
title_snapshot
[ -0.026845140382647514, -0.03389361873269081, -0.0032137215603142977, 0.014860711991786957, 0.0390581451356411, 0.019562555477023125, 0.01676064357161522, 0.017606642097234726, -0.002892513060942292, -0.013722955249249935, -0.004986974876374006, 0.010158385150134563, -0.049779944121837616, ...
Training Agent for First-Person Shooter Game with Actor-Critic Curriculum Learning
https://openreview.net/forum?id=Hk3mPK5gg
[ "Yuxin Wu", "Yuandong Tian" ]
Poster
null
In this paper, we propose a novel framework for training vision-based agent for First-Person Shooter (FPS) Game, in particular Doom. Our framework combines the state-of-the-art reinforcement learning approach (Asynchronous Advantage Actor-Critic (A3C) model) with curriculum learning. Our model is simple in design and o...
[ "Reinforcement Learning", "Applications", "Games" ]
We propose a novel framework for training vision-based agent for First-Person Shooter (FPS) Game, Doom, using actor-critic model and curriculum training.
499
null
null
[ 0.003619636408984661, -0.0066416081972420216, 0.006162987090647221, 0.036646418273448944, 0.037345223128795624, 0.007676164153963327, 0.017781225964426994, 0.009625344537198544, -0.03353998064994812, -0.029604904353618622, -0.01987750269472599, 0.05087518319487572, -0.08183466643095016, -0...
Lossy Image Compression with Compressive Autoencoders
https://openreview.net/forum?id=rJiNwv9gg
[ "Lucas Theis", "Wenzhe Shi", "Andrew Cunningham", "Ferenc Huszár" ]
Poster
null
We propose a new approach to the problem of optimizing autoencoders for lossy image compression. New media formats, changing hardware technology, as well as diverse requirements and content types create a need for compression algorithms which are more flexible than existing codecs. Autoencoders have the potential to ad...
[ "Computer vision", "Deep learning", "Applications" ]
A simple approach to train autoencoders to compress images as well or better than JPEG 2000.
390
1703.00395
title_snapshot
[ 0.0017107718158513308, -0.04885624721646309, -0.01985245756804943, 0.05687730386853218, 0.04674140363931656, 0.058847010135650635, 0.009828168898820877, -0.017680300399661064, -0.04801179841160774, -0.058136068284511566, -0.0029347888194024563, -0.01115450356155634, -0.039699576795101166, ...
Learning and Policy Search in Stochastic Dynamical Systems with Bayesian Neural Networks
https://openreview.net/forum?id=H1fl8S9ee
[ "Stefan Depeweg", "José Miguel Hernández-Lobato", "Finale Doshi-Velez", "Steffen Udluft" ]
Poster
null
We present an algorithm for policy search in stochastic dynamical systems using model-based reinforcement learning. The system dynamics are described with Bayesian neural networks (BNNs) that include stochastic input variables. These input variables allow us to capture complex statistical patterns in the transition dy...
[ "Deep learning", "Reinforcement Learning" ]
249
1605.07127
title_snapshot
[ -0.043974682688713074, -0.0004573248734232038, -0.008185633458197117, 0.04480699077248573, 0.027241528034210205, 0.027665533125400543, 0.015418186783790588, -0.0009547359659336507, -0.030586322769522667, -0.035462912172079086, 0.002176071982830763, 0.013468662276864052, -0.046031732112169266...
Neural Photo Editing with Introspective Adversarial Networks
https://openreview.net/forum?id=HkNKFiGex
[ "Andrew Brock", "Theodore Lim", "J.M. Ritchie", "Nick Weston" ]
Poster
null
The increasingly photorealistic sample quality of generative image models suggests their feasibility in applications beyond image generation. We present the Neural Photo Editor, an interface that leverages the power of generative neural networks to make large, semantically coherent changes to existing images. To tackle...
[ "Computer vision", "Unsupervised Learning", "Applications" ]
An interface for editing photos using generative image models.
11
1609.07093
title_snapshot
[ 0.006376661825925112, -0.017307259142398834, -0.015020502731204033, 0.04682333767414093, 0.03434082865715027, 0.029283152893185616, 0.018749210983514786, -0.00493080448359251, -0.055280592292547226, -0.07780211418867111, -0.04500579833984375, -0.018704939633607864, -0.05913005396723747, -0...
Neural Architecture Search with Reinforcement Learning
https://openreview.net/forum?id=r1Ue8Hcxg
[ "Barret Zoph", "Quoc Le" ]
Oral
null
Neural networks are powerful and flexible models that work well for many difficult learning tasks in image, speech and natural language understanding. Despite their success, neural networks are still hard to design. In this paper, we use a recurrent network to generate the model descriptions of neural networks and trai...
[]
251
1611.01578
title_snapshot
[ -0.02098808065056801, -0.017847532406449318, -0.014825178310275078, 0.057482942938804626, 0.04440467804670334, 0.05215844139456749, 0.020007584244012833, 0.02522769384086132, -0.021194344386458397, -0.03950300067663193, -0.025558603927493095, -0.0027124788612127304, -0.05681189149618149, -...
Batch Policy Gradient Methods for Improving Neural Conversation Models
https://openreview.net/forum?id=rJfMusFll
[ "Kirthevasan Kandasamy", "Yoram Bachrach", "Ryota Tomioka", "Daniel Tarlow", "David Carter" ]
Poster
null
We study reinforcement learning of chat-bots with recurrent neural network architectures when the rewards are noisy and expensive to obtain. For instance, a chat-bot used in automated customer service support can be scored by quality assurance agents, but this process can be expensive, time consuming and noisy. Previo...
[ "Natural language processing", "Reinforcement Learning" ]
124
1702.03334
title_snapshot
[ 0.00040267256554216146, -0.04857183247804642, -0.016961487010121346, 0.04782136529684067, 0.01636418141424656, 0.06637801229953766, 0.0378301627933979, 0.014047438278794289, -0.019456658512353897, -0.016567282378673553, -0.03416569158434868, 0.049236901104450226, -0.08022486418485641, -0.0...
Quasi-Recurrent Neural Networks
https://openreview.net/forum?id=H1zJ-v5xl
[ "James Bradbury", "Stephen Merity", "Caiming Xiong", "Richard Socher" ]
Poster
null
Recurrent neural networks are a powerful tool for modeling sequential data, but the dependence of each timestep’s computation on the previous timestep’s output limits parallelism and makes RNNs unwieldy for very long sequences. We introduce quasi-recurrent neural networks (QRNNs), an approach to neural sequence modelin...
[ "Natural language processing", "Deep learning" ]
QRNNs, composed of convolutions and a recurrent pooling function, outperform LSTMs on a variety of sequence tasks and are up to 16 times faster.
355
1611.01576
title_snapshot
[ -0.005509678740054369, -0.033644791692495346, -0.011847239919006824, 0.03711101785302162, 0.03600284084677696, 0.05429566279053688, 0.01720806397497654, 0.030073776841163635, -0.04331202432513237, -0.021254656836390495, -0.00564353633671999, -0.00561508908867836, -0.06325218081474304, -0.0...
Amortised MAP Inference for Image Super-resolution
https://openreview.net/forum?id=S1RP6GLle
[ "Casper Kaae Sønderby", "Jose Caballero", "Lucas Theis", "Wenzhe Shi", "Ferenc Huszár" ]
Oral
null
Image super-resolution (SR) is an underdetermined inverse problem, where a large number of plausible high resolution images can explain the same downsampled image. Most current single image SR methods use empirical risk minimisation, often with a pixel-wise mean squared error (MSE) loss. However, the outputs from such ...
[ "Theory", "Computer vision", "Deep learning" ]
Probabilisticly motivated image superresolution using a projection to the subspace of valid solutions
29
1610.04490
title_snapshot
[ 0.006394284777343273, 0.010226654820144176, 0.009618348442018032, 0.04127078503370285, 0.028270462527871132, 0.0566839724779129, 0.02015337161719799, -0.0010786207858473063, -0.022481832653284073, -0.04234422370791435, -0.028211887925863266, -0.003642217954620719, -0.05797410383820534, 0.0...
Sigma Delta Quantized Networks
https://openreview.net/forum?id=HkNRsU5ge
[ "Peter O'Connor", "Max Welling" ]
Poster
null
Deep neural networks can be obscenely wasteful. When processing video, a convolutional network expends a fixed amount of computation for each frame with no regard to the similarity between neighbouring frames. As a result, it ends up repeatedly doing very similar computations. To put an end to such waste, we introduce ...
[ "Computer vision", "Deep learning", "Applications" ]
A deep neural network that saves computation on temporal data by using neurons that only communicate their changes in activation
322
1611.02024
title_snapshot
[ 0.01857421174645424, 0.004125108476728201, -0.0278939176350832, 0.04997376725077629, 0.06182653084397316, 0.04174106568098068, 0.005525505170226097, -0.007796625141054392, -0.008421970531344414, -0.031053539365530014, 0.019366271793842316, -0.00795312412083149, -0.04705435037612915, 0.0181...
The Concrete Distribution: A Continuous Relaxation of Discrete Random Variables
https://openreview.net/forum?id=S1jE5L5gl
[ "Chris J. Maddison", "Andriy Mnih", "Yee Whye Teh" ]
Poster
null
The reparameterization trick enables optimizing large scale stochastic computation graphs via gradient descent. The essence of the trick is to refactor each stochastic node into a differentiable function of its parameters and a random variable with fixed distribution. After refactoring, the gradients of the loss propag...
[ "Deep learning", "Unsupervised Learning", "Structured prediction" ]
Relaxed reparameterization trick for discrete stochastic units.
311
1611.00712
title_snapshot
[ 0.009289263747632504, 0.002580889966338873, -0.021160470321774483, 0.06042853370308876, 0.04759007692337036, 0.05967395752668381, 0.032631296664476395, 0.017218006774783134, -0.017785105854272842, -0.04051709920167923, 0.014944948256015778, -0.05627625435590744, -0.07426876574754715, -0.01...
Highway and Residual Networks learn Unrolled Iterative Estimation
https://openreview.net/forum?id=Skn9Shcxe
[ "Klaus Greff", "Rupesh K. Srivastava", "Jürgen Schmidhuber" ]
Poster
null
The past year saw the introduction of new architectures such as Highway networks and Residual networks which, for the first time, enabled the training of feedforward networks with dozens to hundreds of layers using simple gradient descent. While depth of representation has been posited as a primary reason for their suc...
[ "Theory", "Deep learning", "Supervised Learning" ]
539
1612.07771
title_snapshot
[ 0.025497635826468468, -0.0323091559112072, 0.007423484232276678, 0.0492660254240036, 0.047946151345968246, 0.030920930206775665, 0.03953856602311134, 0.025977006182074547, 0.005612980108708143, -0.04412737488746643, 0.018329938873648643, 0.023153357207775116, -0.04754909873008728, -0.00645...
Mode Regularized Generative Adversarial Networks
https://openreview.net/forum?id=HJKkY35le
[ "Tong Che", "Yanran Li", "Athul Jacob", "Yoshua Bengio", "Wenjie Li" ]
Poster
null
Although Generative Adversarial Networks achieve state-of-the-art results on a variety of generative tasks, they are regarded as highly unstable and prone to miss modes. We argue that these bad behaviors of GANs are due to the very particular functional shape of the trained discriminators in high dimensional spaces, wh...
[ "Deep learning", "Unsupervised Learning" ]
541
1612.02136
title_snapshot
[ 0.007914653979241848, -0.051599498838186264, -0.022036978974938393, 0.0364103838801384, 0.017883291468024254, 0.014143363572657108, -0.004660603124648333, -0.0004895555321127176, -0.04358317330479622, -0.06281547993421555, -0.02063470520079136, -0.0019939690828323364, -0.07446391880512238, ...
Introspection:Accelerating Neural Network Training By Learning Weight Evolution
https://openreview.net/forum?id=Hkg8bDqee
[ "Abhishek Sinha", "Aahitagni Mukherjee", "Mausoom Sarkar", "Balaji Krishnamurthy" ]
Poster
null
Neural Networks are function approximators that have achieved state-of-the-art accuracy in numerous machine learning tasks. In spite of their great success in terms of accuracy, their large training time makes it difficult to use them for various tasks. In this paper, we explore the idea of learning weight evolution pa...
[ "Computer vision", "Deep learning", "Optimization" ]
Acceleration of training by performing weight updates, using knowledge obtained from training other neural networks.
360
1704.04959
title_snapshot
[ -0.02596934325993061, -0.0469031035900116, 0.0007088854908943176, 0.017942072823643684, 0.036688704043626785, 0.05135121941566467, 0.0012251598527655005, 0.0018100838642567396, -0.0554504431784153, -0.053769804537296295, 0.00717027485370636, 0.013994189910590649, -0.058899834752082825, 0.0...
Improving Generative Adversarial Networks with Denoising Feature Matching
https://openreview.net/forum?id=S1X7nhsxl
[ "David Warde-Farley", "Yoshua Bengio" ]
Poster
null
We propose an augmented training procedure for generative adversarial networks designed to address shortcomings of the original by directing the generator towards probable configurations of abstract discriminator features. We estimate and track the distribution of these features, as computed from data, with a denoising...
[ "Deep learning", "Unsupervised Learning" ]
Use a denoiser trained on discriminator features to train better generators.
580
null
null
[ 0.008255649358034134, -0.017092106863856316, -0.0280712079256773, 0.0677173063158989, 0.01825854182243347, 0.05124174430966377, 0.027631867676973343, -0.018422827124595642, -0.012535437941551208, -0.061403363943099976, -0.045044805854558945, 0.0036413969937711954, -0.06970994174480438, 0.0...
Outrageously Large Neural Networks: The Sparsely-Gated Mixture-of-Experts Layer
https://openreview.net/forum?id=B1ckMDqlg
[ "Noam Shazeer", "*Azalia Mirhoseini", "*Krzysztof Maziarz", "Andy Davis", "Quoc Le", "Geoffrey Hinton", "Jeff Dean" ]
Poster
null
The capacity of a neural network to absorb information is limited by its number of parameters. Conditional computation, where parts of the network are active on a per-example basis, has been proposed in theory as a way of dramatically increasing model capacity without a proportional increase in computation. In practi...
[ "Deep learning" ]
364
1701.06538
title_snapshot
[ -0.030154496431350708, -0.04016288369894028, -0.008092168718576431, 0.03411052003502846, 0.02671666257083416, 0.03429040312767029, 0.0169070977717638, 0.005715370178222656, -0.027901507914066315, -0.009789048694074154, -0.01441102847456932, 0.013460853137075901, -0.05405830964446068, 0.003...
Normalizing the Normalizers: Comparing and Extending Network Normalization Schemes
https://openreview.net/forum?id=rk5upnsxe
[ "Mengye Ren", "Renjie Liao", "Raquel Urtasun", "Fabian H. Sinz", "Richard S. Zemel" ]
Poster
null
Normalization techniques have only recently begun to be exploited in supervised learning tasks. Batch normalization exploits mini-batch statistics to normalize the activations. This was shown to speed up training and result in better models. However its success has been very limited when dealing with recurrent neural n...
[]
584
1611.04520
title_snapshot
[ -0.0017770454287528992, -0.04002398997545242, -0.00790411327034235, 0.01743527129292488, 0.04245036467909813, 0.033198852092027664, 0.030900251120328903, 0.0031358373817056417, -0.037174392491579056, -0.03765296936035156, 0.010084579698741436, -0.01096818596124649, -0.058046359568834305, -...
Towards Deep Interpretability (MUS-ROVER II): Learning Hierarchical Representations of Tonal Music
https://openreview.net/forum?id=ryhqQFKgl
[ "Haizi Yu", "Lav R. Varshney" ]
Poster
null
Music theory studies the regularity of patterns in music to capture concepts underlying music styles and composers' decisions. This paper continues the study of building \emph{automatic theorists} (rovers) to learn and represent music concepts that lead to human interpretable knowledge and further lead to materials for...
[]
109
null
null
[ -0.01420888677239418, 0.021300356835126877, -0.006379574537277222, 0.02801883965730667, 0.049998994916677475, 0.003515413496643305, 0.04967021569609642, -0.0125078484416008, -0.040315479040145874, -0.050794266164302826, -0.03432416170835495, 0.023661009967327118, -0.03278215602040291, 0.03...
Incorporating long-range consistency in CNN-based texture generation
https://openreview.net/forum?id=HyGTuv9eg
[ "Guillaume Berger", "Roland Memisevic" ]
Poster
null
Gatys et al. (2015) showed that pair-wise products of features in a convolutional network are a very effective representation of image textures. We propose a simple modification to that representation which makes it possible to incorporate long-range structure into image generation, and to render images that satisfy va...
[ "Computer vision", "Deep learning" ]
We propose a simple extension to the Gatys et al. algorithm which makes it possible to incorporate long-range structure into texture generation.
409
1606.01286
title_snapshot
[ 0.02870122343301773, -0.031803473830223083, -0.021767741069197655, 0.009929037652909756, 0.04805135726928711, 0.030988238751888275, 0.002379545010626316, 0.03722450137138367, -0.02385944314301014, -0.0759318619966507, -0.030581118538975716, -0.029244031757116318, -0.06428804993629456, 0.03...
Support Regularized Sparse Coding and Its Fast Encoder
https://openreview.net/forum?id=HkljfjFee
[ "Yingzhen Yang", "Jiahui Yu", "Pushmeet Kohli", "Jianchao Yang", "Thomas S. Huang" ]
Poster
null
Sparse coding represents a signal by a linear combination of only a few atoms of a learned over-complete dictionary. While sparse coding exhibits compelling performance for various machine learning tasks, the process of obtaining sparse code with fixed dictionary is independent for each data point without considering t...
[]
We present Support Regularized Sparse Coding (SRSC) to improve the regular sparse coding, and propose a feed-forward neural network termed Deep Support Regularized Sparse Coding (Deep-SRSC) as its fast encoder.
122
null
null
[ -0.003101157722994685, -0.03213938698172569, -0.007872667163610458, 0.04057998210191727, 0.05533543601632118, 0.05322066694498062, 0.01248861476778984, 0.004810325335711241, -0.052385054528713226, -0.06849046051502228, 0.003199284430593252, -0.006622096057981253, -0.06106998398900032, 0.00...
Transfer of View-manifold Learning to Similarity Perception of Novel Objects
https://openreview.net/forum?id=B1gtu5ilg
[ "Xingyu Lin", "Hao Wang", "Zhihao Li", "Yimeng Zhang", "Alan Yuille", "Tai Sing Lee" ]
Poster
null
We develop a model of perceptual similarity judgment based on re-training a deep convolution neural network (DCNN) that learns to associate different views of each 3D object to capture the notion of object persistence and continuity in our visual experience. The re-training process effectively performs distance metric ...
[ "Deep learning", "Transfer Learning" ]
DCNN trained with multiple views of the same object can develop human-like perpetual similarity judgment that can transfer to novel objects
566
1704.00033
title_snapshot
[ 0.005500770173966885, 0.01463374961167574, -0.008974981494247913, 0.027747543528676033, 0.023262564092874527, 0.011122784577310085, 0.013748707249760628, 0.023208867758512497, -0.020840298384428024, -0.0481218658387661, -0.017271501943469048, 0.013373613357543945, -0.08796804398298264, 0.0...
Variable Computation in Recurrent Neural Networks
https://openreview.net/forum?id=S1LVSrcge
[ "Yacine Jernite", "Edouard Grave", "Armand Joulin", "Tomas Mikolov" ]
Poster
null
Recurrent neural networks (RNNs) have been used extensively and with increasing success to model various types of sequential data. Much of this progress has been achieved through devising recurrent units and architectures with the flexibility to capture complex statistics in the data, such as long range dependency or l...
[ "Natural language processing", "Deep learning" ]
We show that an RNN can learn to control the amount of computation it does at each time step, leading to better efficiency and performance as well as discovering time patterns of interest.
246
1611.06188
title_snapshot
[ 0.007100685965269804, -0.011413056403398514, -0.013766762800514698, 0.03260844573378563, 0.05827636271715164, 0.06530965864658356, 0.06461019068956375, 0.028812112286686897, -0.053561195731163025, -0.014808072708547115, -0.014853041619062424, -0.017506174743175507, -0.04870469123125076, 0....
Learning Visual Servoing with Deep Features and Fitted Q-Iteration
https://openreview.net/forum?id=r1YNw6sxg
[ "Alex X. Lee", "Sergey Levine", "Pieter Abbeel" ]
Poster
null
Visual servoing involves choosing actions that move a robot in response to observations from a camera, in order to reach a goal configuration in the world. Standard visual servoing approaches typically rely on manually designed features and analytical dynamics models, which limits their generalization capability and of...
[ "Computer vision", "Deep learning", "Reinforcement Learning" ]
We use deep semantic features, learned predictive dynamics, and reinforcement learning to efficiently learn a visual servoing policy that is robust to visual variations.
606
1703.11000
title_snapshot
[ -0.006071778479963541, -0.009235790930688381, -0.015645582228899002, 0.03494655713438988, 0.026900222525000572, 0.042618557810783386, -0.0011992952786386013, 0.013303895480930805, -0.029924992471933365, -0.030543163418769836, -0.04072798416018486, 0.013551712967455387, -0.07886829972267151, ...
LR-GAN: Layered Recursive Generative Adversarial Networks for Image Generation
https://openreview.net/forum?id=HJ1kmv9xx
[ "Jianwei Yang", "Anitha Kannan", "Dhruv Batra", "Devi Parikh" ]
Poster
null
We present LR-GAN: an adversarial image generation model which takes scene structure and context into account. Unlike previous generative adversarial networks (GANs), the proposed GAN learns to generate image background and foregrounds separately and recursively, and stitch the foregrounds on the background in a contex...
[ "Computer vision", "Deep learning", "Unsupervised Learning" ]
A layered recursive GAN for image generation, which considers the structure in images and can disentangle the foreground objects from background well in unsupervised manner.
373
1703.01560
title_snapshot
[ 0.022419538348913193, -0.002597623039036989, -0.0010735135292634368, 0.030974891036748886, 0.028417449444532394, 0.023402556777000427, -0.006899544037878513, 0.006610236130654812, -0.008826789446175098, -0.06064978986978531, -0.038718245923519135, -0.009670456871390343, -0.06001095101237297,...
Learning Recurrent Representations for Hierarchical Behavior Modeling
https://openreview.net/forum?id=BkLhzHtlg
[ "Eyrun Eyjolfsdottir", "Kristin Branson", "Yisong Yue", "Pietro Perona" ]
Poster
null
We propose a framework for detecting action patterns from motion sequences and modeling the sensory-motor relationship of animals, using a generative recurrent neural network. The network has a discriminative part (classifying actions) and a generative part (predicting motion), whose recurrent cells are laterally conne...
[ "Unsupervised Learning", "Semi-Supervised Learning", "Reinforcement Learning", "Applications" ]
94
1611.00094
title_snapshot
[ 0.015333927236497402, 0.00010257876419927925, -0.011289759539067745, 0.039569322019815445, 0.051636889576911926, 0.01646377705037594, 0.04708279296755791, 0.025995714589953423, -0.03703295439481735, -0.03222787752747536, -0.0012873552041128278, -0.018026312813162804, -0.06042192503809929, ...
Episodic Exploration for Deep Deterministic Policies for StarCraft Micromanagement
https://openreview.net/forum?id=r1LXit5ee
[ "Nicolas Usunier", "Gabriel Synnaeve", "Zeming Lin", "Soumith Chintala" ]
Poster
null
We consider scenarios from the real-time strategy game StarCraft as benchmarks for reinforcement learning algorithms. We focus on micromanagement, that is, the short-term, low-level control of team members during a battle. We propose several scenarios that are challenging for reinforcement learning algorithms because t...
[ "Deep learning", "Reinforcement Learning", "Games" ]
We propose a new reinforcement learning algorithm based on zero order optimization, that we evaluate on StarCraft micromanagement scenarios.
519
1609.02993
title_judge
[ -0.03882922977209091, -0.007465342991054058, -0.014010925777256489, 0.03289821743965149, 0.032276611775159836, 0.01905663125216961, 0.01577143743634224, -0.0034968166146427393, -0.029474811628460884, -0.030224446207284927, -0.00621037557721138, 0.0026913434267044067, -0.0555710569024086, -...
Learning to Navigate in Complex Environments
https://openreview.net/forum?id=SJMGPrcle
[ "Piotr Mirowski", "Razvan Pascanu", "Fabio Viola", "Hubert Soyer", "Andy Ballard", "Andrea Banino", "Misha Denil", "Ross Goroshin", "Laurent Sifre", "Koray Kavukcuoglu", "Dharshan Kumaran", "Raia Hadsell" ]
Poster
null
Learning to navigate in complex environments with dynamic elements is an important milestone in developing AI agents. In this work we formulate the navigation question as a reinforcement learning problem and show that data efficiency and task performance can be dramatically improved by relying on additional auxiliary t...
[ "Deep learning", "Reinforcement Learning" ]
We proposed a deep RL method, augmented with memory and auxiliary learning targets, for training agents to navigate within large and visually rich environments that include frequently changing start and goal locations
254
1611.03673
title_snapshot
[ -0.0228703822940588, 0.018669862300157547, 0.014047141186892986, 0.011833908967673779, 0.05056079849600792, 0.0000768530007917434, 0.028119055554270744, 0.019352762028574944, -0.02874041348695755, -0.0370524525642395, -0.02480819821357727, 0.014237853698432446, -0.05673830956220627, -0.018...
A recurrent neural network without chaos
https://openreview.net/forum?id=S1dIzvclg
[ "Thomas Laurent", "James von Brecht" ]
Poster
null
We introduce an exceptionally simple gated recurrent neural network (RNN) that achieves performance comparable to well-known gated architectures, such as LSTMs and GRUs, on the word-level language modeling task. We prove that our model has simple, predicable and non-chaotic dynamics. This stands in stark contrast to ...
[]
368
1612.06212
title_snapshot
[ -0.037196654826402664, -0.016932819038629532, -0.010052897036075592, 0.04758370667695999, 0.029435599222779274, 0.05026424676179886, 0.04184975102543831, 0.03241109475493431, -0.04765195772051811, -0.021362492814660072, 0.00342762959189713, 0.0005664175259880722, -0.060449253767728806, 0.0...
Energy-based Generative Adversarial Networks
https://openreview.net/forum?id=ryh9pmcee
[ "Junbo Zhao", "Michael Mathieu", "Yann LeCun" ]
Poster
null
We introduce the "Energy-based Generative Adversarial Network" model (EBGAN) which views the discriminator as an energy function that attributes low energies to the regions near the data manifold and higher energies to other regions. Similar to the probabilistic GANs, a generator is seen as being trained to produce con...
[ "Deep learning", "Unsupervised Learning", "Semi-Supervised Learning" ]
We introduce the "Energy-based Generative Adversarial Network" (EBGAN) model.
200
null
null
[ -0.004456463269889355, -0.021311236545443535, -0.005615478381514549, 0.05078904703259468, 0.020230673253536224, 0.003248395398259163, -0.0066093336790800095, 0.002917317906394601, -0.016945073381066322, -0.06821098178625107, 0.0032420626375824213, 0.00207420252263546, -0.0856175348162651, ...
Designing Neural Network Architectures using Reinforcement Learning
https://openreview.net/forum?id=S1c2cvqee
[ "Bowen Baker", "Otkrist Gupta", "Nikhil Naik", "Ramesh Raskar" ]
Poster
null
At present, designing convolutional neural network (CNN) architectures requires both human expertise and labor. New architectures are handcrafted by careful experimentation or modified from a handful of existing networks. We introduce MetaQNN, a meta-modeling algorithm based on reinforcement learning to automatically g...
[ "Deep learning", "Reinforcement Learning" ]
A Q-learning algorithm for automatically generating neural nets
419
1611.02167
title_snapshot
[ -0.008212137036025524, -0.028635486960411072, -0.023018229752779007, 0.046389415860176086, 0.037608347833156586, 0.020355017855763435, 0.0002180352748837322, 0.01149358507245779, -0.0346238911151886, -0.020774072036147118, -0.0238087959587574, 0.012957431375980377, -0.07711014896631241, -0...
Structured Attention Networks
https://openreview.net/forum?id=HkE0Nvqlg
[ "Yoon Kim", "Carl Denton", "Luong Hoang", "Alexander M. Rush" ]
Poster
null
Attention networks have proven to be an effective approach for embedding categorical inference within a deep neural network. However, for many tasks we may want to model richer structural dependencies without abandoning end-to-end training. In this work, we experiment with incorporating richer structural distributions,...
[]
Use a graphical model as a hidden layer to perform attention over latent structures
379
1702.00887
title_snapshot
[ 0.027582470327615738, 0.007599827367812395, -0.0008892092737369239, 0.05227307975292206, 0.01863192580640316, 0.03439219295978546, 0.024327171966433525, 0.01896248757839203, -0.007416653912514448, -0.011912484653294086, -0.006415847223252058, 0.01364705990999937, -0.05925581976771355, -0.0...
A STRUCTURED SELF-ATTENTIVE SENTENCE EMBEDDING
https://openreview.net/forum?id=BJC_jUqxe
[ "Zhouhan Lin", "Minwei Feng", "Cicero Nogueira dos Santos", "Mo Yu", "Bing Xiang", "Bowen Zhou", "Yoshua Bengio" ]
Poster
null
This paper proposes a new model for extracting an interpretable sentence embedding by introducing self-attention. Instead of using a vector, we use a 2-D matrix to represent the embedding, with each row of the matrix attending on a different part of the sentence. We also propose a self-attention mechanism and a special...
[ "Natural language processing", "Deep learning", "Supervised Learning" ]
a new model for extracting an interpretable sentence embedding by introducing self-attention and matrix representation.
319
1703.03130
title_snapshot
[ 0.01602167636156082, -0.03725939616560936, -0.008202764205634594, 0.038542430847883224, 0.034339237958192825, 0.025081314146518707, 0.019787712022662163, 0.00195955578237772, 0.006094896234571934, 0.002268244279548526, -0.016676533967256546, 0.027219535782933235, -0.04083459824323654, 0.00...
Learning to Play in a Day: Faster Deep Reinforcement Learning by Optimality Tightening
https://openreview.net/forum?id=rJ8Je4clg
[ "Frank S.He", "Yang Liu", "Alexander G. Schwing", "Jian Peng" ]
Poster
null
We propose a novel training algorithm for reinforcement learning which combines the strength of deep Q-learning with a constrained optimization approach to tighten optimality and encourage faster reward propagation. Our novel technique makes deep reinforcement learning more practical by drastically reducing the trainin...
[ "Reinforcement Learning", "Optimization", "Games" ]
We propose a novel training algorithm for reinforcement learning which combines the strength of deep Q-learning with a constrained optimization approach to tighten optimality and encourage faster reward propagation.
202
1611.01606
title_snapshot
[ -0.04031243547797203, -0.03980103135108948, -0.011112491600215435, 0.0468362458050251, 0.05352472513914108, 0.02593708224594593, 0.0036701890639960766, -0.013080841861665249, -0.027299927547574043, -0.033009737730026245, -0.028953321278095245, 0.025686929002404213, -0.05634339898824692, 0....
Deep Learning with Dynamic Computation Graphs
https://openreview.net/forum?id=ryrGawqex
[ "Moshe Looks", "Marcello Herreshoff", "DeLesley Hutchins", "Peter Norvig" ]
Poster
null
Neural networks that compute over graph structures are a natural fit for problems in a variety of domains, including natural language (parse trees) and cheminformatics (molecular graphs). However, since the computation graph has a different shape and size for every input, such networks do not directly support batched t...
[ "Deep learning" ]
We make batching effective and easy to use for neural nets where every input may have a different shape (e.g. TreeRNNs).
440
1702.02181
title_snapshot
[ -0.02539326250553131, -0.012405983172357082, -0.02019997499883175, 0.03352118283510208, 0.030948400497436523, 0.036769427359104156, 0.014234877191483974, 0.025525839999318123, -0.019956905394792557, -0.0510161854326725, 0.013549244962632656, -0.02640138939023018, -0.06176812946796417, -0.0...
Deep Variational Information Bottleneck
https://openreview.net/forum?id=HyxQzBceg
[ "Alexander A. Alemi", "Ian Fischer", "Joshua V. Dillon", "Kevin Murphy" ]
Poster
null
We present a variational approximation to the information bottleneck of Tishby et al. (1999). This variational approach allows us to parameterize the information bottleneck model using a neural network and leverage the reparameterization trick for efficient training. We call this method “Deep Variational Information Bo...
[ "Theory", "Computer vision", "Deep learning", "Supervised Learning" ]
Applying the information bottleneck to deep networks using the variational lower bound and reparameterization trick.
241
1612.00410
title_snapshot
[ 0.012671605683863163, -0.018419845029711723, -0.018473811447620392, 0.04509250074625015, 0.04208945110440254, 0.028854969888925552, 0.036668941378593445, 0.0008581697475165129, -0.04001603648066521, -0.04590211063623428, -0.02510952576994896, 0.026683181524276733, -0.05734356492757797, 0.0...
Generalizing Skills with Semi-Supervised Reinforcement Learning
https://openreview.net/forum?id=ryHlUtqge
[ "Chelsea Finn", "Tianhe Yu", "Justin Fu", "Pieter Abbeel", "Sergey Levine" ]
Poster
null
Deep reinforcement learning (RL) can acquire complex behaviors from low-level inputs, such as images. However, real-world applications of such methods require generalizing to the vast variability of the real world. Deep networks are known to achieve remarkable generalization when provided with massive amounts of labele...
[ "Reinforcement Learning" ]
We propose an algorithm for generalizing a deep neural network policy using "unlabeled" experience collected in MDPs where rewards are not available.
492
1612.00429
title_snapshot
[ -0.018389366567134857, -0.03733516111969948, -0.027293356135487556, 0.05221705883741379, 0.052726197987794876, 0.009639504365622997, 0.014471217058598995, -0.003667678451165557, -0.02718495763838291, -0.02300272136926651, -0.04541853442788124, 0.02809641882777214, -0.08634914457798004, -0....
Loss-aware Binarization of Deep Networks
https://openreview.net/forum?id=S1oWlN9ll
[ "Lu Hou", "Quanming Yao", "James T. Kwok" ]
Poster
null
Deep neural network models, though very powerful and highly successful, are computationally expensive in terms of space and time. Recently, there have been a number of attempts on binarizing the network weights and activations. This greatly reduces the network size, and replaces the underlying multiplications to additi...
[ "Deep learning", "Applications", "Optimization" ]
203
1611.01600
title_snapshot
[ -0.03190309926867485, -0.038266830146312714, 0.010425474494695663, 0.03725648671388626, 0.010999498888850212, 0.04934482276439667, 0.004841700196266174, 0.002760723466053605, -0.03473719209432602, -0.04557066783308983, -0.015836631879210472, -0.015897681936621666, -0.04233008623123169, -0....
Machine Comprehension Using Match-LSTM and Answer Pointer
https://openreview.net/forum?id=B1-q5Pqxl
[ "Shuohang Wang", "Jing Jiang" ]
Poster
null
Machine comprehension of text is an important problem in natural language processing. A recently released dataset, the Stanford Question Answering Dataset (SQuAD), offers a large number of real questions and their answers created by humans through crowdsourcing. SQuAD provides a challenging testbed for evaluating machi...
[ "Natural language processing", "Deep learning" ]
Using Match-LSTM and Answer Pointer to select a variable length answer from a paragraph
417
1608.07905
title_snapshot
[ 0.011802014894783497, -0.04396814480423927, -0.03551676124334335, 0.06175917014479637, 0.035902947187423706, 0.03545192629098892, 0.026938511058688164, 0.030717652291059494, -0.006993407849222422, -0.007478818297386169, -0.02364988811314106, 0.029785029590129852, -0.04567234590649605, -0.0...
Recurrent Environment Simulators
https://openreview.net/forum?id=B1s6xvqlx
[ "Silvia Chiappa", "Sébastien Racaniere", "Daan Wierstra", "Shakir Mohamed" ]
Poster
null
Models that can simulate how environments change in response to actions can be used by agents to plan and act efficiently. We improve on previous environment simulators from high-dimensional pixel observations by introducing recurrent neural networks that are able to make temporally and spatially coherent predictions f...
[ "Deep learning", "Unsupervised Learning", "Applications" ]
354
1704.02254
title_snapshot
[ 0.007524777203798294, -0.0032056174241006374, -0.008781095035374165, 0.03455227613449097, 0.07006772607564926, 0.041693951934576035, 0.03121146373450756, 0.03728346526622772, -0.04488159716129303, -0.05706241354346275, -0.022444885224103928, -0.018050236627459526, -0.06094970181584358, -0....
Hyperband: Bandit-Based Configuration Evaluation for Hyperparameter Optimization
https://openreview.net/forum?id=ry18Ww5ee
[ "Lisha Li", "Kevin Jamieson", "Giulia DeSalvo", "Afshin Rostamizadeh", "Ameet Talwalkar" ]
Poster
null
Performance of machine learning algorithms depends critically on identifying a good set of hyperparameters. While recent approaches use Bayesian Optimization to adaptively select configurations, we focus on speeding up random search through adaptive resource allocation. We present Hyperband, a novel algorithm for hy...
[]
359
null
null
[ -0.010861887596547604, -0.021594811230897903, -0.008909585885703564, 0.060098737478256226, 0.0336710624396801, 0.03860492631793022, 0.0385778471827507, -0.03842237964272499, 0.012987364083528519, -0.06350459158420563, -0.022754600271582603, -0.00807582214474678, -0.0546417273581028, -0.019...
Hierarchical Multiscale Recurrent Neural Networks
https://openreview.net/forum?id=S1di0sfgl
[ "Junyoung Chung", "Sungjin Ahn", "Yoshua Bengio" ]
Poster
null
Learning both hierarchical and temporal representation has been among the long- standing challenges of recurrent neural networks. Multiscale recurrent neural networks have been considered as a promising approach to resolve this issue, yet there has been a lack of empirical evidence showing that this type of models can ...
[ "Natural language processing", "Deep learning" ]
Propose a recurrent neural network architecture that can discover the underlying hierarchical structure in the temporal data.
12
1609.01704
title_snapshot
[ -0.03345983475446701, -0.0023861995432525873, 0.0209383275359869, 0.03662479668855667, 0.03718952462077141, 0.0481891930103302, 0.02139793150126934, 0.013581681065261364, -0.040631916373968124, -0.04178617522120476, 0.018880462273955345, -0.013269678689539433, -0.03389935567975044, 0.01145...
Tree-structured decoding with doubly-recurrent neural networks
https://openreview.net/forum?id=HkYhZDqxg
[ "David Alvarez-Melis", "Tommi S. Jaakkola" ]
Poster
null
We propose a neural network architecture for generating tree-structured objects from encoded representations. The core of the method is a doubly-recurrent neural network that models separately the width and depth recurrences across the tree, and combines them inside each cell to generate an output. The topology of the ...
[ "Natural language processing", "Supervised Learning", "Structured prediction" ]
A new architecture for generating tree-structured objects from encoded representations, which models separately the width and depth recurrences across the tree and predicts both content and topology.
362
null
null
[ -0.010280972346663475, 0.0025410649832338095, 0.0035515516065061092, 0.024770652875304222, 0.0405135303735733, 0.05909527465701103, 0.035274822264909744, 0.01197847444564104, -0.030865555629134178, -0.011364512145519257, 0.0012248177081346512, -0.0167123693972826, -0.052517183125019073, -0...
Training Compressed Fully-Connected Networks with a Density-Diversity Penalty
https://openreview.net/forum?id=Hku9NK5lx
[ "Shengjie Wang", "Haoran Cai", "Jeff Bilmes", "William Noble" ]
Poster
null
Deep models have achieved great success on a variety of challenging tasks. How- ever, the models that achieve great performance often have an enormous number of parameters, leading to correspondingly great demands on both computational and memory resources, especially for fully-connected layers. In this work, we propos...
[ "Deep learning" ]
We propose a new ''density-diversity penalty'' to fully-connected layers to get significantly high sparsity and low diversity trained matrices, while keeping the performance the same.
478
null
null
[ -0.00033162988256663084, -0.03994838520884514, -0.029062708839774132, 0.05490712821483612, 0.04454955458641052, 0.047413915395736694, 0.0053318641148507595, -0.011290067806839943, -0.036445263773202896, -0.06417788565158844, -0.0046294210478663445, -0.014622359536588192, -0.05678834766149521...
Diet Networks: Thin Parameters for Fat Genomics
https://openreview.net/forum?id=Sk-oDY9ge
[ "Adriana Romero", "Pierre Luc Carrier", "Akram Erraqabi", "Tristan Sylvain", "Alex Auvolat", "Etienne Dejoie", "Marc-André Legault", "Marie-Pierre Dubé", "Julie G. Hussin", "Yoshua Bengio" ]
Poster
null
Learning tasks such as those involving genomic data often poses a serious challenge: the number of input features can be orders of magnitude larger than the number of training examples, making it difficult to avoid overfitting, even when using the known regularization techniques. We focus here on tasks in which the inp...
[ "Deep learning", "Unsupervised Learning", "Supervised Learning", "Applications" ]
Drastically reducing the number of parameters, when the number of input features is orders of magnitude larger than the number of training examples, such as in genomics.
505
1611.09340
title_snapshot
[ -0.0039051491767168045, -0.014734776690602303, -0.031065665185451508, 0.025912711396813393, 0.051790751516819, 0.04942699521780014, 0.014264675788581371, -0.012624196708202362, -0.018934741616249084, -0.03235342353582382, 0.01650504767894745, 0.007008821703493595, -0.06484464555978775, 0.0...
What does it take to generate natural textures?
https://openreview.net/forum?id=BJhZeLsxx
[ "Ivan Ustyuzhaninov *", "Wieland Brendel *", "Leon Gatys", "Matthias Bethge" ]
Poster
null
Natural image generation is currently one of the most actively explored fields in Deep Learning. Many approaches, e.g. for state-of-the-art artistic style transfer or natural texture synthesis, rely on the statistics of hierarchical representations in supervisedly trained deep neural networks. It is, however, unclear w...
[ "Deep learning", "Unsupervised Learning" ]
Natural textures of high perceptual quality can be generated from networks with only a single layer, no pooling and random filters.
555
null
null
[ 0.011916925199329853, -0.02239481918513775, 0.01466256845742464, 0.043083399534225464, 0.050258807837963104, 0.024269109591841698, 0.028896324336528778, 0.04384677857160568, -0.03411988168954849, -0.07971561700105667, -0.05810563638806343, -0.01251816377043724, -0.04437180235981941, 0.0216...
Learning Features of Music From Scratch
https://openreview.net/forum?id=rkFBJv9gg
[ "John Thickstun", "Zaid Harchaoui", "Sham Kakade" ]
Poster
null
This paper introduces a new large-scale music dataset, MusicNet, to serve as a source of supervision and evaluation of machine learning methods for music research. MusicNet consists of hundreds of freely-licensed classical music recordings by 10 composers, written for 11 instruments, together with instrument/note a...
[ "Applications" ]
We introduce a new large-scale music dataset, define a multi-label classification task, and benchmark machine learning architectures on this task.
346
1611.09827
title_snapshot
[ -0.010939391329884529, -0.027036244049668312, -0.01241407822817564, 0.035138703882694244, 0.048025984317064285, -0.013059590011835098, 0.020772747695446014, -0.005981008522212505, -0.0321909636259079, -0.04827389121055603, 0.0013867986854165792, 0.021062511950731277, -0.05952278897166252, ...
Automatic Rule Extraction from Long Short Term Memory Networks
https://openreview.net/forum?id=SJvYgH9xe
[ "W. James Murdoch", "Arthur Szlam" ]
Poster
null
Although deep learning models have proven effective at solving problems in natural language processing, the mechanism by which they come to their conclusions is often unclear. As a result, these models are generally treated as black boxes, yielding no insight of the underlying learned patterns. In this paper we cons...
[ "Natural language processing", "Deep learning", "Applications" ]
We introduce a word importance score for LSTMs, and show that we can use it to replicate an LSTM's performance using a simple, rules-based classifier.
235
1702.02540
title_snapshot
[ -0.016752874478697777, -0.020972033962607384, -0.014971298165619373, 0.025806212797760963, 0.03885975107550621, 0.026169590651988983, 0.011563075706362724, 0.017298046499490738, -0.027010811492800713, -0.0020028676372021437, -0.06513400375843048, 0.022070201113820076, -0.06051294505596161, ...
Reasoning with Memory Augmented Neural Networks for Language Comprehension
https://openreview.net/forum?id=Hk8TGSKlg
[ "Tsendsuren Munkhdalai", "Hong Yu" ]
Poster
null
Hypothesis testing is an important cognitive process that supports human reasoning. In this paper, we introduce a computational hypothesis testing approach based on memory augmented neural networks. Our approach involves a hypothesis testing loop that reconsiders and progressively refines a previously formed hypothesis...
[ "Natural language processing", "Deep learning" ]
95
1610.06454
title_snapshot
[ -0.03460819274187088, 0.022590190172195435, -0.019790014252066612, 0.06647896766662598, 0.059838324785232544, 0.022282829508185387, 0.04766581580042839, 0.018349235877394676, -0.04131925478577614, -0.014874937944114208, -0.013341651298105717, 0.050378620624542236, -0.042712222784757614, -0...
Geometry of Polysemy
https://openreview.net/forum?id=HJpfMIFll
[ "Jiaqi Mu", "Suma Bhat", "Pramod Viswanath" ]
Poster
null
Vector representations of words have heralded a transformational approach to classical problems in NLP; the most popular example is word2vec. However, a single vector does not suffice to model the polysemous nature of many (frequent) words, i.e., words with multiple meanings. In this paper, we propose a three-fold a...
[ "Natural language processing" ]
98
1610.07569
title_snapshot
[ -0.013726551085710526, -0.02799125574529171, -0.002009999705478549, 0.035616543143987656, 0.01512302365154028, 0.05626911669969559, 0.027105553075671196, 0.0030383295379579067, 0.00869667250663042, -0.004596192389726639, -0.004447837360203266, 0.007323301397264004, -0.05856925994157791, 0....
Unsupervised Cross-Domain Image Generation
https://openreview.net/forum?id=Sk2Im59ex
[ "Yaniv Taigman", "Adam Polyak", "Lior Wolf" ]
Poster
null
We study the problem of transferring a sample in one domain to an analog sample in another domain. Given two related domains, S and T, we would like to learn a generative function G that maps an input sample from S to the domain T, such that the output of a given representation function f, which accepts inputs in eithe...
[ "Computer vision", "Deep learning", "Unsupervised Learning", "Transfer Learning" ]
533
1611.02200
title_snapshot
[ 0.0027352592442184687, -0.016513757407665253, -0.012354533188045025, 0.05540888383984566, 0.03443897143006325, 0.0022833982948213816, 0.009960069321095943, -0.0024727184791117907, 0.005871904082596302, -0.05190707743167877, -0.009840317070484161, 0.0065848794765770435, -0.06646236777305603, ...
Learning Invariant Feature Spaces to Transfer Skills with Reinforcement Learning
https://openreview.net/forum?id=Hyq4yhile
[ "Abhishek Gupta", "Coline Devin", "YuXuan Liu", "Pieter Abbeel", "Sergey Levine" ]
Poster
null
People can learn a wide range of tasks from their own experience, but can also learn from observing other creatures. This can accelerate acquisition of new skills even when the observed agent differs substantially from the learning agent in terms of morphology. In this paper, we examine how reinforcement learning algor...
[ "Deep learning", "Reinforcement Learning", "Transfer Learning" ]
Learning a common feature space between robots with different morphology or actuation to transfer skills.
571
1703.02949
title_snapshot
[ -0.009625365026295185, -0.009586190804839134, -0.04507870227098465, 0.040116824209690094, 0.05472080409526825, 0.02345135249197483, 0.024731390178203583, -0.00019845330098178238, -0.031862515956163406, -0.04080897942185402, -0.03840954601764679, 0.01150060910731554, -0.07521210610866547, -...
Entropy-SGD: Biasing Gradient Descent Into Wide Valleys
https://openreview.net/forum?id=B1YfAfcgl
[ "Pratik Chaudhari", "Anna Choromanska", "Stefano Soatto", "Yann LeCun", "Carlo Baldassi", "Christian Borgs", "Jennifer Chayes", "Levent Sagun", "Riccardo Zecchina" ]
Poster
null
This paper proposes a new optimization algorithm called Entropy-SGD for training deep neural networks that is motivated by the local geometry of the energy landscape. Local extrema with low generalization error have a large proportion of almost-zero eigenvalues in the Hessian with very few positive or negative eigenval...
[ "Deep learning", "Optimization" ]
This paper focuses on developing new optimization tools for deep learning that are tailored to exploit the local geometric properties of the objective function.
185
1611.01838
title_snapshot
[ -0.02007174678146839, 0.001382163492962718, -0.006594466976821423, 0.051369573920965195, 0.03183421492576599, 0.05349496752023697, 0.0359203964471817, -0.013064870610833168, -0.012838218361139297, -0.050959646701812744, 0.0036977098789066076, -0.007676893845200539, -0.05995739623904228, -0...