title
stringlengths
8
132
paper_url
stringlengths
41
41
authors
listlengths
1
17
type
stringclasses
2 values
primary_area
stringclasses
0 values
abstract
large_stringlengths
228
1.98k
keywords
listlengths
0
11
TL;DR
large_stringlengths
18
281
submission_number
int64
1
1.17k
arxiv_id
stringlengths
10
10
arxiv_id_source
stringclasses
2 values
embedding
listlengths
768
768
Predicting Floor-Level for 911 Calls with Neural Networks and Smartphone Sensor Data
https://openreview.net/forum?id=ryBnUWb0b
[ "William Falcon", "Henning Schulzrinne" ]
Poster
null
In cities with tall buildings, emergency responders need an accurate floor level location to find 911 callers quickly. We introduce a system to estimate a victim's floor level via their mobile device's sensor data in a two-step process. First, we train a neural network to determine when a smartphone enters or exits a b...
[ "Recurrent Neural Networks", "RNN", "LSTM", "Mobile Device", "Sensors" ]
We used an LSTM to detect when a smartphone walks into a building. Then we predict the device's floor level using data from sensors aboard the smartphone.
682
1710.11122
title_snapshot
[ -0.00882117822766304, -0.015959035605192184, 0.008541809394955635, 0.018125588074326515, 0.07942142337560654, 0.02277195267379284, 0.058060407638549805, 0.006101985927671194, -0.041034895926713943, -0.039659660309553146, -0.010503767989575863, -0.014775238931179047, -0.06964671611785889, -...
Identifying Analogies Across Domains
https://openreview.net/forum?id=BkN_r2lR-
[ "Yedid Hoshen", "Lior Wolf" ]
Poster
null
Identifying analogies across domains without supervision is a key task for artificial intelligence. Recent advances in cross domain image mapping have concentrated on translating images across domains. Although the progress made is impressive, the visual fidelity many times does not suffice for identifying the matching...
[ "unsupervised mapping", "cross domain mapping" ]
Finding correspondences between domains by performing matching/mapping iterations
390
null
null
[ 0.005205980502068996, 0.0014934713253751397, -0.004371128976345062, 0.052265044301748276, 0.025134235620498657, 0.0330098494887352, 0.0026469016447663307, -0.017132431268692017, 0.006303371395915747, -0.019511599093675613, -0.022069700062274933, 0.008829405531287193, -0.08475634455680847, ...
Bi-Directional Block Self-Attention for Fast and Memory-Efficient Sequence Modeling
https://openreview.net/forum?id=H1cWzoxA-
[ "Tao Shen", "Tianyi Zhou", "Guodong Long", "Jing Jiang", "Chengqi Zhang" ]
Poster
null
Recurrent neural networks (RNN), convolutional neural networks (CNN) and self-attention networks (SAN) are commonly used to produce context-aware representations. RNN can capture long-range dependency but is hard to parallelize and not time-efficient. CNN focuses on local dependency but does not perform well on some ta...
[ "deep learning", "attention mechanism", "sequence modeling", "natural language processing", "sentence embedding" ]
A self-attention network for RNN/CNN-free sequence encoding with small memory consumption, highly parallelizable computation and state-of-the-art performance on several NLP tasks
366
1804.00857
title_snapshot
[ -0.021931257098913193, -0.020803553983569145, -0.03071465902030468, 0.0036145253106951714, 0.022578110918402672, 0.033825501799583435, 0.03455242142081261, 0.01983339712023735, -0.0398731492459774, -0.02286730520427227, -0.004843106959015131, -0.013306677341461182, -0.04681742191314697, -0...
WHAI: Weibull Hybrid Autoencoding Inference for Deep Topic Modeling
https://openreview.net/forum?id=S1cZsf-RW
[ "Hao Zhang", "Bo Chen", "Dandan Guo", "Mingyuan Zhou" ]
Poster
null
To train an inference network jointly with a deep generative topic model, making it both scalable to big corpora and fast in out-of-sample prediction, we develop Weibull hybrid autoencoding inference (WHAI) for deep latent Dirichlet allocation, which infers posterior samples via a hybrid of stochastic-gradient MCMC and...
[]
null
916
1803.01328
title_snapshot
[ 0.014240720309317112, -0.006218180991709232, -0.036225367337465286, 0.07505238056182861, 0.03638703003525734, 0.037247829139232635, 0.026777075603604317, -0.019905028864741325, 0.014234423637390137, -0.02247280813753605, -0.017355263233184814, 0.01871182583272457, -0.04965279623866081, 0.0...
Defense-GAN: Protecting Classifiers Against Adversarial Attacks Using Generative Models
https://openreview.net/forum?id=BkJ3ibb0-
[ "Pouya Samangouei", "Maya Kabkab", "Rama Chellappa" ]
Poster
null
In recent years, deep neural network approaches have been widely adopted for machine learning tasks, including classification. However, they were shown to be vulnerable to adversarial perturbations: carefully crafted small perturbations can cause misclassification of legitimate images. We propose Defense-GAN, a new fra...
[]
Defense-GAN uses a Generative Adversarial Network to defend against white-box and black-box attacks in classification models.
714
1805.06605
title_snapshot
[ -0.0026334028225392103, -0.015479927882552147, -0.02545906789600849, 0.06200990825891495, 0.0216456837952137, 0.004062569234520197, 0.021775009110569954, -0.022359322756528854, -0.013557489030063152, -0.031486161053180695, 0.0015997464070096612, 0.006784346420317888, -0.06526253372430801, ...
Unsupervised Learning of Goal Spaces for Intrinsically Motivated Goal Exploration
https://openreview.net/forum?id=S1DWPP1A-
[ "Alexandre Péré", "Sébastien Forestier", "Olivier Sigaud", "Pierre-Yves Oudeyer" ]
Poster
null
Intrinsically motivated goal exploration algorithms enable machines to discover repertoires of policies that produce a diversity of effects in complex environments. These exploration algorithms have been shown to allow real world robots to acquire skills such as tool use in high-dimensional continuous state and action ...
[ "exploration; autonomous goal setting; diversity; unsupervised learning; deep neural network" ]
We propose a novel Intrinsically Motivated Goal Exploration architecture with unsupervised learning of goal space representations, and evaluate how various implementations enable the discovery of a diversity of policies.
132
1803.00781
title_snapshot
[ -0.015695273876190186, -0.020092416554689407, -0.040847018361091614, 0.01732824370265007, 0.04715793952345848, 0.022446060553193092, 0.04299220070242882, -0.004609055817127228, -0.017563732340931892, -0.04291651025414467, -0.024792974814772606, 0.0042251888662576675, -0.04688017815351486, ...
Word translation without parallel data
https://openreview.net/forum?id=H196sainb
[ "Guillaume Lample", "Alexis Conneau", "Marc'Aurelio Ranzato", "Ludovic Denoyer", "Hervé Jégou" ]
Poster
null
State-of-the-art methods for learning cross-lingual word embeddings have relied on bilingual dictionaries or parallel corpora. Recent studies showed that the need for parallel data supervision can be alleviated with character-level information. While these methods showed encouraging results, they are not on par with th...
[ "unsupervised learning", "machine translation", "multilingual embeddings", "parallel dictionary induction", "adversarial training" ]
Aligning languages without the Rosetta Stone: with no parallel data, we construct bilingual dictionaries using adversarial training, cross-domain local scaling, and an accurate proxy criterion for cross-validation.
7
1710.04087
title_snapshot
[ -0.01098568644374609, -0.018647965043783188, -0.009314741007983685, 0.056230511516332626, 0.026987580582499504, 0.03558565676212311, 0.02138718031346798, 0.04158126562833786, 0.0007997015491127968, -0.022743666544556618, -0.03077409788966179, 0.017009513452649117, -0.08357419073581696, -0....
Critical Points of Linear Neural Networks: Analytical Forms and Landscape Properties
https://openreview.net/forum?id=SysEexbRb
[ "Yi Zhou", "Yingbin Liang" ]
Poster
null
Due to the success of deep learning to solving a variety of challenging machine learning tasks, there is a rising interest in understanding loss functions for training neural networks from a theoretical aspect. Particularly, the properties of critical points and the landscape around them are of importance to determine ...
[ "neural networks", "critical points", "analytical form", "landscape" ]
We provide necessary and sufficient analytical forms for the critical points of the square loss functions for various neural networks, and exploit the analytical forms to characterize the landscape properties for the loss functions of these neural networks.
549
1710.11205
title_judge
[ -0.0464487224817276, -0.004708519671112299, 0.017313610762357712, 0.007392146624624729, 0.0258532352745533, 0.05503984913229942, 0.006993961986154318, -0.0008417338249273598, -0.03133567422628403, -0.03646806627511978, -0.01702294312417507, -0.01398836262524128, -0.06295453011989594, 0.026...
Meta-Learning and Universality: Deep Representations and Gradient Descent can Approximate any Learning Algorithm
https://openreview.net/forum?id=HyjC5yWCW
[ "Chelsea Finn", "Sergey Levine" ]
Poster
null
Learning to learn is a powerful paradigm for enabling models to learn from data more effectively and efficiently. A popular approach to meta-learning is to train a recurrent model to read in a training dataset as input and output the parameters of a learned model, or output predictions for new test inputs. Alternativel...
[ "meta-learning", "learning to learn", "universal function approximation" ]
Deep representations combined with gradient descent can approximate any learning algorithm.
513
1710.11622
title_snapshot
[ -0.007225027773529291, -0.0209309384226799, 0.002220800844952464, 0.02033274807035923, 0.039958689361810684, 0.020209547132253647, 0.036387424916028976, 0.012874819338321686, -0.041468068957328796, -0.01136431097984314, -0.023257814347743988, 0.006924136076122522, -0.0613304004073143, 0.00...
Maximum a Posteriori Policy Optimisation
https://openreview.net/forum?id=S1ANxQW0b
[ "Abbas Abdolmaleki", "Jost Tobias Springenberg", "Yuval Tassa", "Remi Munos", "Nicolas Heess", "Martin Riedmiller" ]
Poster
null
We introduce a new algorithm for reinforcement learning called Maximum a-posteriori Policy Optimisation (MPO) based on coordinate ascent on a relative-entropy objective. We show that several existing methods can directly be related to our derivation. We develop two off-policy algorithms and demonstrate that they are co...
[ "Reinforcement Learning", "Variational Inference", "Control" ]
null
1,110
1806.06920
title_snapshot
[ -0.051782529801130295, -0.017168739810585976, 0.036886751651763916, 0.031217413023114204, 0.01084095798432827, 0.03323904052376747, 0.01387869194149971, -0.023587355390191078, -0.025724826380610466, -0.03610273823142052, -0.03131997212767601, 0.029991816729307175, -0.06251335889101028, -0....
The Implicit Bias of Gradient Descent on Separable Data
https://openreview.net/forum?id=r1q7n9gAb
[ "Daniel Soudry", "Elad Hoffer", "Mor Shpigel Nacson", "Nathan Srebro" ]
Poster
null
We show that gradient descent on an unregularized logistic regression problem, for almost all separable datasets, converges to the same direction as the max-margin solution. The result generalizes also to other monotone decreasing loss functions with an infimum at infinity, and we also discuss a multi-class generalizat...
[ "gradient descent", "implicit regularization", "generalization", "margin", "logistic regression", "loss functions", "optimization", "exponential tail", "cross-entropy" ]
The normalized solution of gradient descent on logistic regression (or a similarly decaying loss) slowly converges to the L2 max margin solution on separable data.
358
1710.10345
title_snapshot
[ -0.029198644682765007, -0.010418265126645565, 0.002921631559729576, 0.02093510329723358, 0.04815668612718582, 0.029161883518099785, 0.04068203270435333, 0.014848505146801472, -0.016718054190278053, -0.030011668801307678, -0.019296295940876007, 0.02678500860929489, -0.08203630894422531, -0....
Online Learning Rate Adaptation with Hypergradient Descent
https://openreview.net/forum?id=BkrsAzWAb
[ "Atilim Gunes Baydin", "Robert Cornish", "David Martinez Rubio", "Mark Schmidt", "Frank Wood" ]
Poster
null
We introduce a general method for improving the convergence rate of gradient-based optimizers that is easy to implement and works well in practice. We demonstrate the effectiveness of the method in a range of optimization problems by applying it to stochastic gradient descent, stochastic gradient descent with Nesterov...
[]
null
1,073
1703.04782
title_snapshot
[ -0.01696207746863365, -0.004903807770460844, 0.021591678261756897, 0.0284022968262434, 0.03598853200674057, 0.03612507879734039, 0.026802167296409607, 0.012055512517690659, -0.010776642709970474, -0.013174038380384445, -0.0005348656559363008, 0.019762197509407997, -0.05554678663611412, -0....
Spherical CNNs
https://openreview.net/forum?id=Hkbd5xZRb
[ "Taco S. Cohen", "Mario Geiger", "Jonas Köhler", "Max Welling" ]
Oral
null
Convolutional Neural Networks (CNNs) have become the method of choice for learning problems involving 2D planar images. However, a number of problems of recent interest have created a demand for models that can analyze spherical images. Examples include omnidirectional vision for drones, robots, and autonomous cars, mo...
[ "deep learning", "equivariance", "convolution", "group convolution", "3D", "vision", "omnidirectional", "shape recognition", "molecular energy regression" ]
We introduce Spherical CNNs, a convolutional network for spherical signals, and apply it to 3D model recognition and molecular energy regression.
615
1801.10130
title_snapshot
[ 0.01357749942690134, 0.019428757950663567, 0.024446815252304077, -0.004423927515745163, 0.022537415847182274, 0.047586578875780106, -0.007188132032752037, 0.012532936409115791, -0.038563407957553864, -0.04990912228822708, -0.0033897929824888706, -0.015593959018588066, -0.07763206213712692, ...
Syntax-Directed Variational Autoencoder for Structured Data
https://openreview.net/forum?id=SyqShMZRb
[ "Hanjun Dai", "Yingtao Tian", "Bo Dai", "Steven Skiena", "Le Song" ]
Poster
null
Deep generative models have been enjoying success in modeling continuous data. However it remains challenging to capture the representations for discrete structures with formal grammars and semantics, e.g., computer programs and molecular structures. How to generate both syntactically and semantically correct data stil...
[ "generative model for structured data", "syntax-directed generation", "molecule and program optimization", "variational autoencoder" ]
A new generative model for discrete structured data. The proposed stochastic lazy attribute converts the offline semantic check into online guidance for stochastic decoding, which effectively addresses the constraints in syntax and semantics, and also achieves superior performance
958
1802.08786
title_snapshot
[ 0.00897517055273056, 0.027800800278782845, -0.04416391998529434, 0.04244391620159149, 0.03899063169956207, 0.04922778904438019, 0.04684646427631378, -0.001921936054714024, 0.01041381899267435, -0.039164308458566666, -0.024177709594368935, 0.015155627392232418, -0.08548662811517715, 0.02104...
Learning to Count Objects in Natural Images for Visual Question Answering
https://openreview.net/forum?id=B12Js_yRb
[ "Yan Zhang", "Jonathon Hare", "Adam Prügel-Bennett" ]
Poster
null
Visual Question Answering (VQA) models have struggled with counting objects in natural images so far. We identify a fundamental problem due to soft attention in these models as a cause. To circumvent this problem, we propose a neural network component that allows robust counting from object proposals. Experiments on a ...
[ "visual question answering", "vqa", "counting" ]
Enabling Visual Question Answering models to count by handling overlapping object proposals.
136
1802.05766
title_snapshot
[ 0.004056731704622507, -0.027360010892152786, 0.005058032460510731, 0.02006514184176922, 0.004001489374786615, 0.04820661246776581, 0.027456413954496384, 0.008611969649791718, -0.06533165276050568, -0.02102339081466198, -0.05023583024740219, 0.050248581916093826, -0.0532064214348793, -0.008...
Smooth Loss Functions for Deep Top-k Classification
https://openreview.net/forum?id=Hk5elxbRW
[ "Leonard Berrada", "Andrew Zisserman", "M. Pawan Kumar" ]
Poster
null
The top-$k$ error is a common measure of performance in machine learning and computer vision. In practice, top-$k$ classification is typically performed with deep neural networks trained with the cross-entropy loss. Theoretical results indeed suggest that cross-entropy is an optimal learning objective for such a task i...
[]
Smooth Loss Function for Top-k Error Minimization
547
1802.07595
title_snapshot
[ -0.025315016508102417, -0.006660567130893469, 0.014455234631896019, 0.031696952879428864, 0.01995968259871006, 0.009783133864402771, 0.017260994762182236, -0.009478062391281128, -0.012896654196083546, -0.03273375332355499, -0.037335120141506195, -0.013760757632553577, -0.04520314931869507, ...
Minimal-Entropy Correlation Alignment for Unsupervised Deep Domain Adaptation
https://openreview.net/forum?id=rJWechg0Z
[ "Pietro Morerio", "Jacopo Cavazza", "Vittorio Murino" ]
Poster
null
In this work, we face the problem of unsupervised domain adaptation with a novel deep learning approach which leverages our finding that entropy minimization is induced by the optimal alignment of second order statistics between source and target domains. We formally demonstrate this hypothesis and, aiming at achieving...
[ "unsupervised domain adaptation", "entropy minimization", "image classification", "deep transfer learning" ]
A new unsupervised deep domain adaptation technique which efficiently unifies correlation alignment and entropy minimization
399
1711.10288
title_snapshot
[ 0.0007122370298020542, -0.01444464921951294, 0.00028114410815760493, 0.033562593162059784, 0.03032728284597397, 0.05045776069164276, 0.03165511041879654, 0.007419539149850607, -0.006735925562679768, -0.04033986106514931, -0.03952386602759361, 0.03921985626220703, -0.07431724667549133, 0.00...
PixelNN: Example-based Image Synthesis
https://openreview.net/forum?id=Syhr6pxCW
[ "Aayush Bansal", "Yaser Sheikh", "Deva Ramanan" ]
Poster
null
We present a simple nearest-neighbor (NN) approach that synthesizes high-frequency photorealistic images from an ``incomplete'' signal such as a low-resolution image, a surface normal map, or edges. Current state-of-the-art deep generative models designed for such conditional image synthesis lack two important things: ...
[ "conditional image synthesis", "nearest neighbors" ]
Pixel-wise nearest neighbors used for generating multiple images from incomplete priors such as a low-res images, surface normals, edges etc.
432
1708.05349
title_snapshot
[ -0.00712076760828495, -0.0360240675508976, -0.015091652050614357, 0.057505007833242416, 0.013987922109663486, 0.04103174805641174, -0.01133669912815094, 0.02255409024655819, -0.012385843321681023, -0.06204657629132271, -0.031258903443813324, -0.016385691240429878, -0.06934119015932083, -0....
Trust-PCL: An Off-Policy Trust Region Method for Continuous Control
https://openreview.net/forum?id=HyrCWeWCb
[ "Ofir Nachum", "Mohammad Norouzi", "Kelvin Xu", "Dale Schuurmans" ]
Poster
null
Trust region methods, such as TRPO, are often used to stabilize policy optimization algorithms in reinforcement learning (RL). While current trust region strategies are effective for continuous control, they typically require a large amount of on-policy interaction with the environment. To address this problem, we prop...
[ "Reinforcement learning" ]
We extend recent insights related to softmax consistency to achieve state-of-the-art results in continuous control.
558
1707.01891
title_snapshot
[ -0.02484126016497612, -0.028649915009737015, -0.01753860153257847, 0.06490597873926163, 0.04874582216143608, 0.008646583184599876, 0.02779524400830269, 0.0031892124097794294, -0.015953613445162773, -0.044979970902204514, -0.021701103076338768, 0.021428152918815613, -0.08706803619861603, -0...
Stochastic Activation Pruning for Robust Adversarial Defense
https://openreview.net/forum?id=H1uR4GZRZ
[ "Guneet S. Dhillon", "Kamyar Azizzadenesheli", "Zachary C. Lipton", "Jeremy D. Bernstein", "Jean Kossaifi", "Aran Khanna", "Animashree Anandkumar" ]
Poster
null
Neural networks are known to be vulnerable to adversarial examples. Carefully chosen perturbations to real images, while imperceptible to humans, induce misclassification and threaten the reliability of deep learning systems in the wild. To guard against adversarial examples, we take inspiration from game theory and ca...
[]
null
828
1803.01442
title_snapshot
[ -0.020094551146030426, -0.029943183064460754, -0.019547348842024803, 0.04132024571299553, 0.007046016864478588, 0.03358945995569229, 0.034780070185661316, -0.010102465748786926, -0.04926028475165367, -0.04219242185354233, -0.019647663459181786, 0.0014560346025973558, -0.04049054905772209, ...
Learning Sparse Latent Representations with the Deep Copula Information Bottleneck
https://openreview.net/forum?id=Hk0wHx-RW
[ "Aleksander Wieczorek*", "Mario Wieser*", "Damian Murezzan", "Volker Roth" ]
Poster
null
Deep latent variable models are powerful tools for representation learning. In this paper, we adopt the deep information bottleneck model, identify its shortcomings and propose a model that circumvents them. To this end, we apply a copula transformation which, by restoring the invariance properties of the information b...
[ "Information Bottleneck", "Deep Information Bottleneck", "Deep Variational Information Bottleneck", "Variational Autoencoder", "Sparsity", "Disentanglement", "Interpretability", "Copula", "Mutual Information" ]
We apply the copula transformation to the Deep Information Bottleneck which leads to restored invariance properties and a disentangled latent space with superior predictive capabilities.
583
1804.06216
title_snapshot
[ 0.005649222061038017, -0.031149383634328842, -0.02547849528491497, 0.04865643382072449, 0.05859638750553131, 0.033733393996953964, 0.014494332484900951, 0.0005820713704451919, -0.015567640773952007, -0.030750665813684464, -0.010746563784778118, 0.021244807168841362, -0.06316360086202621, 0...
Meta-Learning for Semi-Supervised Few-Shot Classification
https://openreview.net/forum?id=HJcSzz-CZ
[ "Mengye Ren", "Eleni Triantafillou", "Sachin Ravi", "Jake Snell", "Kevin Swersky", "Joshua B. Tenenbaum", "Hugo Larochelle", "Richard S. Zemel" ]
Poster
null
In few-shot classification, we are interested in learning algorithms that train a classifier from only a handful of labeled examples. Recent progress in few-shot classification has featured meta-learning, in which a parameterized model for a learning algorithm is defined and trained on episodes representing different c...
[ "Few-shot learning", "semi-supervised learning", "meta-learning" ]
We propose novel extensions of Prototypical Networks that are augmented with the ability to use unlabeled examples when producing prototypes.
788
1803.00676
title_snapshot
[ 0.02406560443341732, -0.05150153115391731, -0.033128611743450165, 0.04200468957424164, 0.02949308231472969, 0.007002879865467548, 0.039650171995162964, -0.010809138417243958, -0.04316452145576477, -0.0070051695220172405, -0.02814771607518196, 0.017343968152999878, -0.06484969705343246, 0.0...
Deep contextualized word representations
https://openreview.net/forum?id=S1p31z-Ab
[ "Matthew E Peters", "Mark Neumann", "Mohit Iyyer", "Matt Gardner", "Christopher Clark", "Kenton Lee", "Luke Zettlemoyer" ]
Poster
null
We introduce a new type of deep contextualized word representation that models both (1) complex characteristics of word use (e.g., syntax and semantics), and (2) how these uses vary across linguistic contexts (i.e., to model polysemy). Our word vectors are learned functions of the internal states of a deep bidirection...
[ "representation learning", "contextualized word embeddings" ]
We introduce a new type of deep contextualized word representation that significantly improves the state of the art for a range of challenging NLP tasks.
759
1802.05365
title_snapshot
[ -0.002492350060492754, -0.007592133712023497, -0.0034281082917004824, 0.04071110114455223, 0.01807609386742115, 0.042209427803754807, 0.01806378737092018, 0.013378721661865711, -0.011608745902776718, -0.0010192649206146598, -0.034244511276483536, 0.01721842773258686, -0.04677005484700203, ...
Stochastic Variational Video Prediction
https://openreview.net/forum?id=rk49Mg-CW
[ "Mohammad Babaeizadeh", "Chelsea Finn", "Dumitru Erhan", "Roy H. Campbell", "Sergey Levine" ]
Poster
null
Predicting the future in real-world settings, particularly from raw sensory observations such as images, is exceptionally challenging. Real-world events can be stochastic and unpredictable, and the high dimensionality and complexity of natural images requires the predictive model to build an intricate understanding of ...
[ "video prediction", "stochastic prediction", "variational inference", "unsupervised learning" ]
Stochastic variational video prediction in real-world settings.
565
1710.11252
title_snapshot
[ 0.02725755237042904, -0.007454450707882643, 0.008266022428870201, 0.045385390520095825, 0.027847809717059135, 0.0335998497903347, 0.02408510260283947, 0.019946977496147156, -0.052143242210149765, -0.05239493027329445, -0.017285877838730812, -0.016798580065369606, -0.05946774408221245, 0.01...
An efficient framework for learning sentence representations
https://openreview.net/forum?id=rJvJXZb0W
[ "Lajanugen Logeswaran", "Honglak Lee" ]
Poster
null
In this work we propose a simple and efficient framework for learning sentence representations from unlabelled data. Drawing inspiration from the distributional hypothesis and recent work on learning sentence representations, we reformulate the problem of predicting the context in which a sentence appears as a classifi...
[ "sentence", "embeddings", "unsupervised", "representations", "learning", "efficient" ]
A framework for learning high-quality sentence representations efficiently.
661
1803.02893
title_snapshot
[ 0.005022410303354263, -0.004309287294745445, -0.016502920538187027, 0.051733240485191345, 0.017836403101682663, 0.03567757084965706, 0.016155555844306946, 0.023816855624318123, -0.011162315495312214, 0.01080179400742054, -0.006075950339436531, 0.0301668718457222, -0.06178702786564827, 0.00...
On the importance of single directions for generalization
https://openreview.net/forum?id=r1iuQjxCZ
[ "Ari S. Morcos", "David G.T. Barrett", "Neil C. Rabinowitz", "Matthew Botvinick" ]
Poster
null
Despite their ability to memorize large datasets, deep neural networks often achieve good generalization performance. However, the differences between the learned solutions of networks which generalize and those which do not remain unclear. Additionally, the tuning properties of single directions (defined as the activa...
[ "generalization", "analysis", "deep learning", "selectivity" ]
We find that deep networks which generalize poorly are more reliant on single directions than those that generalize well, and evaluate the impact of dropout and batch normalization, as well as class selectivity on single direction reliance.
369
1803.06959
title_snapshot
[ -0.030581574887037277, -0.013058813288807869, 0.00182339025195688, 0.030032994225621223, 0.02681725285947323, 0.023746898397803307, 0.05282488465309143, 0.017989154905080795, -0.03213909640908241, -0.036378152668476105, -0.014833475463092327, -0.0004990965826436877, -0.06921706348657608, -...
Combining Symbolic Expressions and Black-box Function Evaluations in Neural Programs
https://openreview.net/forum?id=Hksj2WWAW
[ "Forough Arabshahi", "Sameer Singh", "Animashree Anandkumar" ]
Poster
null
Neural programming involves training neural networks to learn programs, mathematics, or logic from data. Previous works have failed to achieve good generalization performance, especially on problems and programs with high complexity or on large domains. This is because they mostly rely either on black-box function eval...
[ "symbolic reasoning", "mathematical equations", "recursive neural networks", "neural programing" ]
null
720
1801.04342
title_snapshot
[ -0.03495708480477333, -0.002581464359536767, -0.04043480381369591, 0.05179893970489502, 0.02201739512383938, 0.06556295603513718, 0.030470792204141617, 0.0036662169732153416, -0.034461647272109985, -0.014308029785752296, -0.011975246481597424, 0.03621629253029823, -0.04791298508644104, -0....
Deep Complex Networks
https://openreview.net/forum?id=H1T2hmZAb
[ "Chiheb Trabelsi", "Olexa Bilaniuk", "Ying Zhang", "Dmitriy Serdyuk", "Sandeep Subramanian", "Joao Felipe Santos", "Soroush Mehri", "Negar Rostamzadeh", "Yoshua Bengio", "Christopher J Pal" ]
Poster
null
At present, the vast majority of building blocks, techniques, and architectures for deep learning are based on real-valued operations and representations. However, recent work on recurrent neural networks and older fundamental theoretical analysis suggests that complex numbers could have a richer representational capac...
[ "deep learning", "complex-valued neural networks" ]
null
1,172
1705.09792
title_snapshot
[ -0.0030894598457962275, -0.02214437536895275, 0.004397695418447256, 0.0494903102517128, 0.05146406218409538, 0.03823599964380264, 0.015831081196665764, -0.000846593058668077, -0.014335870742797852, -0.04810179024934769, -0.005645222030580044, 0.006050902884453535, -0.06357617676258087, 0.0...
Compressing Word Embeddings via Deep Compositional Code Learning
https://openreview.net/forum?id=BJRZzFlRb
[ "Raphael Shu", "Hideki Nakayama" ]
Poster
null
Natural language processing (NLP) models often require a massive number of parameters for word embeddings, resulting in a large storage or memory footprint. Deploying neural NLP models to mobile devices requires compressing the word embeddings without any significant sacrifices in performance. For this purpose, we prop...
[ "natural language processing", "word embedding", "compression", "deep learning" ]
Compressing the word embeddings over 94% without hurting the performance.
331
1711.01068
title_snapshot
[ -0.018872613087296486, -0.039474617689847946, -0.021794414147734642, 0.04099154844880104, 0.043415654450654984, 0.050442736595869064, 0.006396484095603228, 0.018313612788915634, 0.00020654960826504976, -0.04271675646305084, -0.03422738239169121, -0.001581320771947503, -0.06360693275928497, ...
Viterbi-based Pruning for Sparse Matrix with Fixed and High Index Compression Ratio
https://openreview.net/forum?id=S1D8MPxA-
[ "Dongsoo Lee", "Daehyun Ahn", "Taesu Kim", "Pierce I. Chuang", "Jae-Joon Kim" ]
Poster
null
Weight pruning has proven to be an effective method in reducing the model size and computation cost while not sacrificing the model accuracy. Conventional sparse matrix formats, however, involve irregular index structures with large storage requirement and sequential reconstruction process, resulting in inefficient use...
[ "pruning", "sparse matrix", "memory footprint", "model size", "model compression" ]
We present a new pruning method and sparse matrix format to enable high index compression ratio and parallel index decoding process.
288
null
null
[ 0.003795238444581628, -0.04646560922265053, 0.000461057061329484, 0.025816364213824272, 0.052983831614255905, 0.07266286015510559, 0.015593347139656544, 0.01771414838731289, -0.045537982136011124, -0.07246560603380203, 0.004835622850805521, -0.014813641086220741, -0.038924530148506165, 0.0...
Interpretable Counting for Visual Question Answering
https://openreview.net/forum?id=S1J2ZyZ0Z
[ "Alexander Trott", "Caiming Xiong", "Richard Socher" ]
Poster
null
Questions that require counting a variety of objects in images remain a major challenge in visual question answering (VQA). The most common approaches to VQA involve either classifying answers based on fixed length representations of both the image and question or summing fractional counts estimated from each section o...
[ "Counting", "VQA", "Object detection" ]
We perform counting for visual question answering; our model produces interpretable outputs by counting directly from detected objects.
475
1712.08697
title_snapshot
[ 0.015164060518145561, -0.023379411548376083, -0.00405474379658699, 0.023855557665228844, 0.005244908854365349, 0.04723960906267166, 0.01103917881846428, -0.001373618608340621, -0.05225948616862297, -0.00917062908411026, -0.06008492037653923, 0.03805438056588173, -0.06156713888049126, -0.00...
Semantically Decomposing the Latent Spaces of Generative Adversarial Networks
https://openreview.net/forum?id=S1nQvfgA-
[ "Chris Donahue", "Zachary C. Lipton", "Akshay Balsubramani", "Julian McAuley" ]
Poster
null
We propose a new algorithm for training generative adversarial networks to jointly learn latent codes for both identities (e.g. individual humans) and observations (e.g. specific photographs). In practice, this means that by fixing the identity portion of latent codes, we can generate diverse images of the same subject...
[ "disentangled representations", "generative adversarial networks", "generative modeling", "image synthesis" ]
SD-GANs disentangle latent codes according to known commonalities in a dataset (e.g. photographs depicting the same person).
224
1705.07904
title_snapshot
[ 0.0026821899227797985, -0.018761729821562767, -0.042135242372751236, 0.06207934394478798, 0.02376149781048298, 0.003895650850608945, 0.012477791868150234, -0.001949892146512866, -0.009347699582576752, -0.04650503769516945, -0.02844548597931862, -0.010354069992899895, -0.07150561362504959, ...
Adaptive Quantization of Neural Networks
https://openreview.net/forum?id=SyOK1Sg0W
[ "Soroosh Khoram", "Jing Li" ]
Poster
null
Despite the state-of-the-art accuracy of Deep Neural Networks (DNN) in various classification problems, their deployment onto resource constrained edge computing devices remains challenging due to their large size and complexity. Several recent studies have reported remarkable results in reducing this complexity throug...
[ "Deep Neural Networks", "Model Quantization", "Model Compression" ]
An adaptive method for fixed-point quantization of neural networks based on theoretical analysis rather than heuristics.
246
null
null
[ -0.0355028472840786, -0.029171353206038475, -0.018033718690276146, 0.029838277027010918, 0.051573481410741806, 0.07041444629430771, -0.004476693458855152, -0.021509360522031784, -0.04292864352464676, -0.04775772988796234, -0.022337425500154495, -0.012945815920829773, -0.050742775201797485, ...
Rethinking the Smaller-Norm-Less-Informative Assumption in Channel Pruning of Convolution Layers
https://openreview.net/forum?id=HJ94fqApW
[ "Jianbo Ye", "Xin Lu", "Zhe Lin", "James Z. Wang" ]
Poster
null
Model pruning has become a useful technique that improves the computational efficiency of deep learning, making it possible to deploy solutions in resource-limited scenarios. A widely-used practice in relevant work assumes that a smaller-norm parameter or feature plays a less informative role at the inference time. In ...
[ "model pruning", "batch normalization", "convolutional neural network", "ISTA" ]
A CNN model pruning method using ISTA and rescaling trick to enforce sparsity of scaling parameters in batch normalization.
99
1802.00124
title_snapshot
[ 0.0007465849630534649, -0.03039354272186756, -0.004461041651666164, 0.028757765889167786, 0.02783745341002941, 0.024494854733347893, 0.013596981763839722, -0.017980895936489105, -0.015942886471748352, -0.0527159608900547, -0.028543639928102493, 0.00858865026384592, -0.05215969681739807, -0...
Intrinsic Motivation and Automatic Curricula via Asymmetric Self-Play
https://openreview.net/forum?id=SkT5Yg-RZ
[ "Sainbayar Sukhbaatar", "Zeming Lin", "Ilya Kostrikov", "Gabriel Synnaeve", "Arthur Szlam", "Rob Fergus" ]
Poster
null
We describe a simple scheme that allows an agent to learn about its environment in an unsupervised manner. Our scheme pits two versions of the same agent, Alice and Bob, against one another. Alice proposes a task for Bob to complete; and then Bob attempts to complete the task. In this work we will focus on two kinds o...
[ "self-play", "automatic curriculum", "intrinsic motivation", "unsupervised learning", "reinforcement learning" ]
Unsupervised learning for reinforcement learning using an automatic curriculum of self-play
612
1703.05407
title_snapshot
[ -0.01772954873740673, -0.03230220824480057, -0.020948419347405434, 0.040022749453783035, 0.04840755835175514, -0.008566321805119514, 0.03512771055102348, -0.005836171098053455, -0.018271593376994133, -0.022886723279953003, -0.03837807849049568, 0.043464288115501404, -0.04198629781603813, -...
Large scale distributed neural network training through online distillation
https://openreview.net/forum?id=rkr1UDeC-
[ "Rohan Anil", "Gabriel Pereyra", "Alexandre Passos", "Robert Ormandi", "George E. Dahl", "Geoffrey E. Hinton" ]
Poster
null
Techniques such as ensembling and distillation promise model quality improvements when paired with almost any base model. However, due to increased test-time cost (for ensembles) and increased complexity of the training pipeline (for distillation), these techniques are challenging to use in industrial settings. In this...
[ "distillation", "distributed training", "neural networks", "deep learning" ]
We perform large scale experiments to show that a simple online variant of distillation can help us scale distributed neural network training to more machines.
291
1804.03235
title_snapshot
[ -0.02267969772219658, -0.017090603709220886, -0.028401795774698257, 0.05898481234908104, 0.036675382405519485, 0.020085901021957397, 0.024395855143666267, 0.02600053884088993, -0.030906017869710922, -0.020641885697841644, -0.009137820452451706, -0.02918318659067154, -0.07227224111557007, 0...
Universal Agent for Disentangling Environments and Tasks
https://openreview.net/forum?id=B1mvVm-C-
[ "Jiayuan Mao", "Honghua Dong", "Joseph J. Lim" ]
Poster
null
Recent state-of-the-art reinforcement learning algorithms are trained under the goal of excelling in one specific task. Hence, both environment and task specific knowledge are entangled into one framework. However, there are often scenarios where the environment (e.g. the physical world) is fixed while only the target ...
[ "reinforcement learning", "transfer learning" ]
We propose a DRL framework that disentangles task and environment specific knowledge.
1,160
null
null
[ 0.011101999320089817, -0.01891806162893772, 0.010362384840846062, 0.040559861809015274, 0.055916909128427505, 0.014014714397490025, 0.03756364807486534, 0.0196603424847126, -0.012237143702805042, -0.041425615549087524, -0.032481350004673004, 0.03241492435336113, -0.08486702293157578, -0.01...
FusionNet: Fusing via Fully-aware Attention with Application to Machine Comprehension
https://openreview.net/forum?id=BJIgi_eCZ
[ "Hsin-Yuan Huang", "Chenguang Zhu", "Yelong Shen", "Weizhu Chen" ]
Poster
null
This paper introduces a new neural structure called FusionNet, which extends existing attention approaches from three perspectives. First, it puts forward a novel concept of "History of Word" to characterize attention information from the lowest word-level embedding up to the highest semantic-level representation. Seco...
[ "Attention Mechanism", "Machine Comprehension", "Natural Language Processing", "Deep Learning" ]
We propose a light-weight enhancement for attention and a neural architecture, FusionNet, to achieve SotA on SQuAD and adversarial SQuAD.
317
1711.07341
title_snapshot
[ -0.01667144149541855, -0.04381360113620758, 0.005522029474377632, 0.05329841747879982, 0.009681801311671734, 0.027347303926944733, 0.008060705848038197, 0.024100156500935555, -0.001489542075432837, -0.018628405407071114, -0.007759619504213333, 0.0025165146216750145, -0.05539911985397339, -...
Global Optimality Conditions for Deep Neural Networks
https://openreview.net/forum?id=BJk7Gf-CZ
[ "Chulhee Yun", "Suvrit Sra", "Ali Jadbabaie" ]
Poster
null
We study the error landscape of deep linear and nonlinear neural networks with the squared error loss. Minimizing the loss of a deep linear neural network is a nonconvex problem, and despite recent progress, our understanding of this loss surface is still incomplete. For deep linear networks, we present necessary and s...
[ "deep linear neural networks", "global optimality", "deep learning" ]
We provide efficiently checkable necessary and sufficient conditions for global optimality in deep linear neural networks, with some initial extensions to nonlinear settings.
784
1707.02444
title_snapshot
[ -0.029508046805858612, 0.002724555553868413, 0.011691978201270103, 0.05214463919401169, 0.03617069870233536, 0.053442616015672684, 0.02038470469415188, 0.005020730197429657, -0.022278888151049614, -0.030224181711673737, -0.010652292519807816, -0.005618609953671694, -0.06422695517539978, 0....
Guide Actor-Critic for Continuous Control
https://openreview.net/forum?id=BJk59JZ0b
[ "Voot Tangkaratt", "Abbas Abdolmaleki", "Masashi Sugiyama" ]
Poster
null
Actor-critic methods solve reinforcement learning problems by updating a parameterized policy known as an actor in a direction that increases an estimate of the expected return known as a critic. However, existing actor-critic methods only use values or gradients of the critic to update the policy parameter. In this pa...
[ "Reinforcement learning", "actor-critic", "continuous control" ]
This paper proposes a novel actor-critic method that uses Hessians of a critic to update an actor.
511
1705.07606
title_snapshot
[ -0.038978517055511475, -0.03749898821115494, -0.020993974059820175, 0.026021437719464302, 0.030495576560497284, 0.03697992116212845, 0.008435642346739769, -0.0021405534353107214, -0.04213438183069229, -0.04751569405198097, -0.013217371888458729, 0.047173287719488144, -0.0869838148355484, -...
Kronecker-factored Curvature Approximations for Recurrent Neural Networks
https://openreview.net/forum?id=HyMTkQZAb
[ "James Martens", "Jimmy Ba", "Matt Johnson" ]
Poster
null
Kronecker-factor Approximate Curvature (Martens & Grosse, 2015) (K-FAC) is a 2nd-order optimization method which has been shown to give state-of-the-art performance on large-scale neural network optimization tasks (Ba et al., 2017). It is based on an approximation to the Fisher information matrix (FIM) that makes assu...
[ "optimization", "K-FAC", "natural gradient", "recurrent neural networks" ]
We extend the K-FAC method to RNNs by developing a new family of Fisher approximations.
1,098
null
null
[ -0.0010086182737722993, -0.04531831666827202, 0.010975612327456474, 0.032633405178785324, 0.0392073392868042, 0.06172020733356476, 0.016467155888676643, 0.019534973427653313, -0.03543778881430626, -0.04563894122838974, 0.021497827023267746, -0.018133725970983505, -0.04790552705526352, -0.0...
Learning a neural response metric for retinal prosthesis
https://openreview.net/forum?id=HJhIM0xAW
[ "Nishal P Shah", "Sasidhar Madugula", "EJ Chichilnisky", "Yoram Singer", "Jonathon Shlens" ]
Poster
null
Retinal prostheses for treating incurable blindness are designed to electrically stimulate surviving retinal neurons, causing them to send artificial visual signals to the brain. However, electrical stimulation generally cannot precisely reproduce normal patterns of neural activity in the retina. Therefore, an electr...
[ "Metric learning", "Computational Neuroscience", "Retina", "Neural Prosthesis" ]
Using triplets to learn a metric for comparing neural responses and improve the performance of a prosthesis.
445
null
null
[ 0.009565506130456924, 0.004435622598975897, -0.006322883069515228, -0.008380278013646603, 0.05585797503590584, 0.05867834389209747, 0.009385193698108196, 0.01693865843117237, -0.06157223507761955, -0.029022332280874252, -0.013203468173742294, 0.016154035925865173, -0.06633543223142624, -0....
Certified Defenses against Adversarial Examples
https://openreview.net/forum?id=Bys4ob-Rb
[ "Aditi Raghunathan", "Jacob Steinhardt", "Percy Liang" ]
Poster
null
While neural networks have achieved high accuracy on standard image classification benchmarks, their accuracy drops to nearly zero in the presence of small adversarial perturbations to test inputs. Defenses based on regularization and adversarial training have been proposed, but often followed by new, stronger attacks ...
[ "adversarial examples", "certificate of robustness", "convex relaxations" ]
We demonstrate a certifiable, trainable, and scalable method for defending against adversarial examples.
711
1801.09344
title_snapshot
[ -0.0020361500792205334, -0.034484196454286575, -0.01566980965435505, 0.06057806685566902, 0.03629155457019806, 0.010528276674449444, 0.03397147357463837, -0.0324573740363121, -0.0464291013777256, -0.03264039754867554, -0.002076741773635149, 0.0008958104299381375, -0.05557669699192047, 0.00...
Activation Maximization Generative Adversarial Nets
https://openreview.net/forum?id=HyyP33gAZ
[ "Zhiming Zhou", "Han Cai", "Shu Rong", "Yuxuan Song", "Kan Ren", "Weinan Zhang", "Jun Wang", "Yong Yu" ]
Poster
null
Class labels have been empirically shown useful in improving the sample quality of generative adversarial nets (GANs). In this paper, we mathematically study the properties of the current variants of GANs that make use of class label information. With class aware gradient and cross-entropy decomposition, we reveal how ...
[ "Generative Adversarial Nets", "GANs", "Evaluation Metrics", "Generative Model", "Deep Learning", "Adversarial Learning", "Inception Score", "AM Score" ]
Understand how class labels help GAN training. Propose a new evaluation metric for generative models.
403
1703.02000
title_snapshot
[ 0.0033749700523912907, -0.026106152683496475, -0.011101406067609787, 0.033515386283397675, -0.0012349217431619763, -0.0020413293968886137, 0.019558874890208244, -0.03201732039451599, 0.005646591540426016, -0.06320351362228394, -0.019683830440044403, 0.006638558581471443, -0.05777071788907051...
Evidence Aggregation for Answer Re-Ranking in Open-Domain Question Answering
https://openreview.net/forum?id=rJl3yM-Ab
[ "Shuohang Wang", "Mo Yu", "Jing Jiang", "Wei Zhang", "Xiaoxiao Guo", "Shiyu Chang", "Zhiguo Wang", "Tim Klinger", "Gerald Tesauro", "Murray Campbell" ]
Poster
null
Very recently, it comes to be a popular approach for answering open-domain questions by first searching question-related passages, then applying reading comprehension models to extract answers. Existing works usually extract answers from single passages independently, thus not fully make use of the multiple searched pa...
[ "Question Answering", "Deep Learning" ]
We propose a method that can make use of the multiple passages information for open-domain QA.
758
1711.05116
title_snapshot
[ -0.022495990619063377, -0.03402644395828247, 0.014950140379369259, 0.056444644927978516, 0.0504971519112587, -0.0026274663396179676, 0.03448154777288437, -0.011646264232695103, -0.01154640968888998, -0.000869851850438863, -0.019487090408802032, 0.03288860619068146, -0.08295290917158127, -0...
Few-Shot Learning with Graph Neural Networks
https://openreview.net/forum?id=BJj6qGbRW
[ "Victor Garcia Satorras", "Joan Bruna Estrach" ]
Poster
null
We propose to study the problem of few-shot learning with the prism of inference on a partially observed graphical model, constructed from a collection of input images whose label can be either observed or not. By assimilating generic message-passing inference algorithms with their neural-network counterparts, we defin...
[]
null
910
1711.04043
title_snapshot
[ -0.004311266355216503, -0.017335671931505203, -0.020935887470841408, 0.06784810870885849, 0.020423198118805885, 0.025245409458875656, 0.043846942484378815, 0.02422882616519928, -0.02815852500498295, -0.016909349709749222, 0.009642045944929123, 0.019832398742437363, -0.06975105404853821, -0...
Action-dependent Control Variates for Policy Optimization via Stein Identity
https://openreview.net/forum?id=H1mCp-ZRZ
[ "Hao Liu*", "Yihao Feng*", "Yi Mao", "Dengyong Zhou", "Jian Peng", "Qiang Liu" ]
Poster
null
Policy gradient methods have achieved remarkable successes in solving challenging reinforcement learning problems. However, it still often suffers from the large variance issue on policy gradient estimation, which leads to poor sample efficiency during training. In this work, we propose a control variate method to effe...
[ "reinforcement learning", "control variates", "sample efficiency", "variance reduction" ]
null
736
1710.11198
title_judge
[ -0.021133890375494957, -0.013595902360975742, -0.010843878611922264, 0.021351881325244904, 0.03705941513180733, 0.0413215346634388, 0.0421752966940403, -0.038653869181871414, -0.019938642159104347, -0.03405395895242691, 0.00825792457908392, 0.003587043844163418, -0.07338108122348785, -0.00...
Decision Boundary Analysis of Adversarial Examples
https://openreview.net/forum?id=BkpiPMbA-
[ "Warren He", "Bo Li", "Dawn Song" ]
Poster
null
Deep neural networks (DNNs) are vulnerable to adversarial examples, which are carefully crafted instances aiming to cause prediction errors for DNNs. Recent research on adversarial examples has examined local neighborhoods in the input space of DNN models. However, previous work has limited what regions to consider, fo...
[ "adversarial machine learning", "supervised representation learning", "decision regions", "decision boundaries" ]
Looking at decision boundaries around an input gives you more information than a fixed small neighborhood
863
null
null
[ -0.05266328528523445, -0.025491168722510338, -0.015072574838995934, 0.048505619168281555, 0.021035604178905487, 0.003221428021788597, 0.02802581898868084, -0.022392915561795235, -0.012742199003696442, -0.04350631684064865, -0.025745343416929245, -0.005611917935311794, -0.07342180609703064, ...
Memory-based Parameter Adaptation
https://openreview.net/forum?id=rkfOvGbCW
[ "Pablo Sprechmann", "Siddhant M. Jayakumar", "Jack W. Rae", "Alexander Pritzel", "Adria Puigdomenech Badia", "Benigno Uria", "Oriol Vinyals", "Demis Hassabis", "Razvan Pascanu", "Charles Blundell" ]
Poster
null
Deep neural networks have excelled on a wide range of problems, from vision to language and game playing. Neural networks very gradually incorporate information into weights as they process data, requiring very low learning rates. If the training distribution shifts, the network is slow to adapt, and when it does adap...
[]
null
857
1802.10542
title_snapshot
[ -0.04748104512691498, 0.01719006337225437, -0.016123928129673004, 0.013039793819189072, 0.030052093788981438, 0.05249514803290367, -0.004140317440032959, 0.007411973550915718, -0.023601463064551353, -0.03799949586391449, -0.012000911869108677, 0.011566507630050182, -0.06300050765275955, -0...
FearNet: Brain-Inspired Model for Incremental Learning
https://openreview.net/forum?id=SJ1Xmf-Rb
[ "Ronald Kemker", "Christopher Kanan" ]
Poster
null
Incremental class learning involves sequentially learning classes in bursts of examples from the same class. This violates the assumptions that underlie methods for training standard deep neural networks, and will cause them to suffer from catastrophic forgetting. Arguably, the best method for incremental class learni...
[ "Incremental Learning", "Lifelong Learning", "Supervised Learning", "Catastrophic Forgetting", "Brain-Inspired", "Neural Networks" ]
FearNet is a memory efficient neural-network, inspired by memory formation in the mammalian brain, that is capable of incremental class learning without catastrophic forgetting.
802
1711.10563
title_snapshot
[ -0.018933815881609917, -0.004749429877847433, -0.005006520077586174, 0.03486213460564613, 0.03679903969168663, 0.006232856772840023, 0.03440031036734581, 0.019342981278896332, -0.04962413012981415, -0.03276553750038147, -0.007841398939490318, -0.004896161612123251, -0.032974109053611755, -...
Neural Map: Structured Memory for Deep Reinforcement Learning
https://openreview.net/forum?id=Bk9zbyZCZ
[ "Emilio Parisotto", "Ruslan Salakhutdinov" ]
Poster
null
A critical component to enabling intelligent reasoning in partially observable environments is memory. Despite this importance, Deep Reinforcement Learning (DRL) agents have so far used relatively simple memory architectures, with the main methods to overcome partial observability being either a temporal convolution ov...
[ "deep reinforcement learning", "deep learning", "memory" ]
null
471
1702.08360
title_snapshot
[ -0.025288568809628487, -0.023729797452688217, -0.025138046592473984, 0.05993164703249931, 0.04155655950307846, 0.032087840139865875, -0.00018911666120402515, 0.006452439818531275, -0.027169980108737946, -0.033508285880088806, -0.024490732699632645, 0.0064714159816503525, -0.04483290761709213...
Certifying Some Distributional Robustness with Principled Adversarial Training
https://openreview.net/forum?id=Hk6kPgZA-
[ "Aman Sinha", "Hongseok Namkoong", "John Duchi" ]
Oral
null
Neural networks are vulnerable to adversarial examples and researchers have proposed many heuristic attack and defense mechanisms. We address this problem through the principled lens of distributionally robust optimization, which guarantees performance under adversarial input perturbations. By considering a Lagrangian...
[ "adversarial training", "distributionally robust optimization", "deep learning", "optimization", "learning theory" ]
We provide a fast, principled adversarial training procedure with computational and statistical performance guarantees.
596
1710.10571
title_snapshot
[ -0.007804980501532555, -0.011412586085498333, -0.01589023880660534, 0.056908655911684036, 0.034938711673021317, 0.016292069107294083, 0.02281440794467926, -0.023962289094924927, -0.008868278935551643, -0.01467803493142128, 0.005895864684134722, -0.004753234796226025, -0.05383477732539177, ...
Latent Constraints: Learning to Generate Conditionally from Unconditional Generative Models
https://openreview.net/forum?id=Sy8XvGb0-
[ "Jesse Engel", "Matthew Hoffman", "Adam Roberts" ]
Poster
null
Deep generative neural networks have proven effective at both conditional and unconditional modeling of complex data distributions. Conditional generation enables interactive control, but creating new controls often requires expensive retraining. In this paper, we develop a method to condition generation without retrai...
[ "VAE", "GAN", "generative networks", "conditional generation", "latent-variable models" ]
A new approach to conditional generation by constraining the latent space of an unconditional generative model.
853
1711.05772
title_snapshot
[ -0.021834298968315125, -0.02183639071881771, -0.026114070788025856, 0.0725179985165596, 0.04189824312925339, 0.02838592603802681, 0.006994300056248903, -0.027795124799013138, -0.02500101737678051, -0.02709205262362957, -0.039780762046575546, 0.02701840177178383, -0.06243189051747322, 0.015...
Matrix capsules with EM routing
https://openreview.net/forum?id=HJWLfGWRb
[ "Geoffrey E Hinton", "Sara Sabour", "Nicholas Frosst" ]
Poster
null
A capsule is a group of neurons whose outputs represent different properties of the same entity. Each layer in a capsule network contains many capsules. We describe a version of capsules in which each capsule has a logistic unit to represent the presence of an entity and a 4x4 matrix which could learn to represent the ...
[ "Computer Vision", "Deep Learning", "Dynamic routing" ]
Capsule networks with learned pose matrices and EM routing improves state of the art classification on smallNORB, improves generalizability to new view points, and white box adversarial robustness.
789
null
null
[ 0.0042094942182302475, -0.030543681234121323, -0.008416504599153996, 0.04848172143101692, 0.018243761733174324, 0.004220666829496622, 0.002525515854358673, -0.009987916797399521, -0.019732899963855743, -0.04660964757204056, -0.016048341989517212, -0.006280454341322184, -0.0755813717842102, ...
A Hierarchical Model for Device Placement
https://openreview.net/forum?id=Hkc-TeZ0W
[ "Azalia Mirhoseini", "Anna Goldie", "Hieu Pham", "Benoit Steiner", "Quoc V. Le", "Jeff Dean" ]
Poster
null
We introduce a hierarchical model for efficient placement of computational graphs onto hardware devices, especially in heterogeneous environments with a mixture of CPUs, GPUs, and other computational devices. Our method learns to assign graph operations to groups and to allocate those groups to available devices. The g...
[ "deep learning", "device placement", "policy gradient optimization" ]
We introduce a hierarchical model for efficient, end-to-end placement of computational graphs onto hardware devices.
629
null
null
[ -0.015545811504125595, -0.0008216627757064998, 0.00008697382145328447, 0.04437988996505737, 0.029917383566498756, 0.045523520559072495, 0.004597014281898737, 0.01451349537819624, -0.004095443524420261, -0.03742491826415062, 0.018886057659983635, -0.01239217072725296, -0.0552561990916729, 0...
Apprentice: Using Knowledge Distillation Techniques To Improve Low-Precision Network Accuracy
https://openreview.net/forum?id=B1ae1lZRb
[ "Asit Mishra", "Debbie Marr" ]
Poster
null
Deep learning networks have achieved state-of-the-art accuracies on computer vision workloads like image classification and object detection. The performant systems, however, typically involve big models with numerous parameters. Once trained, a challenging aspect for such top performing models is deployment on resourc...
[ "Ternary", "4-bits", "low precision", "knowledge distillation", "knowledge transfer", "model compression" ]
We show that knowledge transfer techniques can improve the accuracy of low precision networks and set new state-of-the-art accuracy for ternary and 4-bits precision.
540
1711.05852
title_snapshot
[ 0.023730045184493065, -0.013568347319960594, -0.024069769307971, 0.061475154012441635, 0.06071656197309494, 0.01875177212059498, 0.021425802260637283, -0.015669628977775574, -0.014031996950507164, -0.018867196515202522, -0.01966298185288906, -0.010254313237965107, -0.05281146988272667, 0.0...
Residual Connections Encourage Iterative Inference
https://openreview.net/forum?id=SJa9iHgAZ
[ "Stanisław Jastrzebski", "Devansh Arpit", "Nicolas Ballas", "Vikas Verma", "Tong Che", "Yoshua Bengio" ]
Poster
null
Residual networks (Resnets) have become a prominent architecture in deep learning. However, a comprehensive understanding of Resnets is still a topic of ongoing research. A recent view argues that Resnets perform iterative refinement of features. We attempt to further expose properties of this aspect. To this end, we s...
[ "residual network", "iterative inference", "deep learning" ]
Residual connections really perform iterative inference
262
1710.04773
title_snapshot
[ 0.00934735219925642, -0.015264085493981838, 0.010781055316329002, 0.06181473657488823, 0.04419275000691414, 0.05097836256027222, 0.023949552327394485, -0.010355810634791851, -0.013762914575636387, -0.030069928616285324, 0.009841566905379295, 0.011673104017972946, -0.045160822570323944, -0....
Twin Networks: Matching the Future for Sequence Generation
https://openreview.net/forum?id=BydLzGb0Z
[ "Dmitriy Serdyuk", "Nan Rosemary Ke", "Alessandro Sordoni", "Adam Trischler", "Chris Pal", "Yoshua Bengio" ]
Poster
null
We propose a simple technique for encouraging generative RNNs to plan ahead. We train a ``backward'' recurrent network to generate a given sequence in reverse order, and we encourage states of the forward model to predict cotemporal states of the backward model. The backward network is used only during training, and pl...
[ "generative rnns", "long term dependencies", "speech recognition", "image captioning" ]
The paper introduces a method of training generative recurrent networks that helps to plan ahead. We run a second RNN in a reverse direction and make a soft constraint between cotemporal forward and backward states.
790
1708.06742
title_snapshot
[ -0.008271950297057629, -0.02015741914510727, -0.022272001951932907, 0.04198600724339485, 0.04696467146277428, 0.056018125265836716, 0.03369242325425148, 0.024967415258288383, -0.030413279309868813, -0.026412025094032288, 0.012687873095273972, 0.02485404908657074, -0.04787275567650795, -0.0...
Towards Image Understanding from Deep Compression Without Decoding
https://openreview.net/forum?id=HkXWCMbRW
[ "Robert Torfason", "Fabian Mentzer", "Eirikur Agustsson", "Michael Tschannen", "Radu Timofte", "Luc Van Gool" ]
Poster
null
Motivated by recent work on deep neural network (DNN)-based image compression methods showing potential improvements in image quality, savings in storage, and bandwidth reduction, we propose to perform image understanding tasks such as classification and segmentation directly on the compressed representations produced ...
[]
null
1,036
1803.06131
title_snapshot
[ -0.008439581841230392, -0.02636866644024849, -0.026697903871536255, 0.03664586693048477, 0.04994327947497368, 0.057813744992017746, -0.024289004504680634, -0.002532634884119034, -0.030207160860300064, -0.05670509859919548, -0.030142689123749733, -0.0173626821488142, -0.04016025736927986, 0...
Mixed Precision Training
https://openreview.net/forum?id=r1gs9JgRZ
[ "Paulius Micikevicius", "Sharan Narang", "Jonah Alben", "Gregory Diamos", "Erich Elsen", "David Garcia", "Boris Ginsburg", "Michael Houston", "Oleksii Kuchaiev", "Ganesh Venkatesh", "Hao Wu" ]
Poster
null
Increasing the size of a neural network typically improves accuracy but also increases the memory and compute requirements for training the model. We introduce methodology for training deep neural networks using half-precision floating point numbers, without losing model accuracy or having to modify hyper-parameters. T...
[ "Half precision", "float16", "Convolutional neural networks", "Recurrent neural networks" ]
null
203
1710.03740
title_snapshot
[ 0.002818535780534148, -0.006538063287734985, -0.042982615530490875, 0.02797006629407406, 0.032614294439554214, 0.03543594852089882, -0.000653549563139677, -0.018743110820651054, -0.054213471710681915, -0.03245893865823746, -0.00011824666580650955, -0.010700547136366367, -0.05621076002717018,...
Relational Neural Expectation Maximization: Unsupervised Discovery of Objects and their Interactions
https://openreview.net/forum?id=ryH20GbRW
[ "Sjoerd van Steenkiste", "Michael Chang", "Klaus Greff", "Jürgen Schmidhuber" ]
Poster
null
Common-sense physical reasoning is an essential ingredient for any intelligent agent operating in the real-world. For example, it can be used to simulate the environment, or to infer the state of parts of the world that are currently unobserved. In order to match real-world conditions this causal knowledge must be lear...
[ "Common-sense Physical Reasoning", "Intuitive Physics", "Representation Learning", "Model building" ]
We introduce a novel approach to common-sense physical reasoning that learns to discover objects and model their physical interactions from raw visual images in a purely unsupervised fashion
1,056
1802.10353
title_snapshot
[ 0.006894792430102825, 0.015383390709757805, -0.0016873141285032034, 0.04353103041648865, 0.03248664364218712, 0.01591399312019348, 0.004867549054324627, 0.017836397513747215, -0.03688729554414749, -0.016836808994412422, -0.031012635678052902, 0.042794983834028244, -0.070090651512146, -0.03...
Learning Deep Mean Field Games for Modeling Large Population Behavior
https://openreview.net/forum?id=HktK4BeCZ
[ "Jiachen Yang", "Xiaojing Ye", "Rakshit Trivedi", "Huan Xu", "Hongyuan Zha" ]
Oral
null
We consider the problem of representing collective behavior of large populations and predicting the evolution of a population distribution over a discrete state space. A discrete time mean field game (MFG) is motivated as an interpretable model founded on game theory for understanding the aggregate effect of individual...
[ "mean field games", "reinforcement learning", "Markov decision processes", "inverse reinforcement learning", "deep learning", "inverse optimal control", "computational social science", "population modeling" ]
Inference of a mean field game (MFG) model of large population behavior via a synthesis of MFG and Markov decision processes.
254
1711.03156
title_snapshot
[ -0.045910805463790894, -0.016468465328216553, 0.005566626787185669, 0.030990133062005043, 0.05092582106590271, 0.01845450885593891, 0.015265163034200668, 0.038887351751327515, -0.024704303592443466, -0.029537823051214218, 0.0061803036369383335, 0.014897901564836502, -0.06813157349824905, -...
Learning Intrinsic Sparse Structures within Long Short-Term Memory
https://openreview.net/forum?id=rk6cfpRjZ
[ "Wei Wen", "Yuxiong He", "Samyam Rajbhandari", "Minjia Zhang", "Wenhan Wang", "Fang Liu", "Bin Hu", "Yiran Chen", "Hai Li" ]
Poster
null
Model compression is significant for the wide adoption of Recurrent Neural Networks (RNNs) in both user devices possessing limited resources and business clusters requiring quick responses to large-scale service requests. This work aims to learn structurally-sparse Long Short-Term Memory (LSTM) by reducing the sizes of...
[ "Sparsity", "Model Compression", "Acceleration", "LSTMs", "Recurrent Neural Networks", "Structural Learning" ]
null
2
1709.05027
title_snapshot
[ -0.023803263902664185, -0.039260078221559525, -0.011184191331267357, 0.03988277539610863, 0.03170493617653847, 0.054781462997198105, 0.03321865200996399, 0.010099278762936592, -0.06140948459506035, -0.0331338495016098, 0.009834397584199905, -0.01946643553674221, -0.0628988966345787, 0.0125...
Empirical Risk Landscape Analysis for Understanding Deep Neural Networks
https://openreview.net/forum?id=B1QgVti6Z
[ "Pan Zhou", "Jiashi Feng" ]
Poster
null
This work aims to provide comprehensive landscape analysis of empirical risk in deep neural networks (DNNs), including the convergence behavior of its gradient, its stationary points and the empirical risk itself to their corresponding population counterparts, which reveals how various network parameters determine the...
[ "Deep Learning Analysis", "Deep Learning Theory", "Empirical Risk", "Landscape Analysis", "Nonconvex Optimization" ]
null
51
null
null
[ -0.03560647740960121, -0.007699762471020222, -0.007461911533027887, 0.03685097396373749, 0.03836003690958023, 0.04958746209740639, 0.0016800850862637162, 0.014403504319489002, -0.01592453569173813, -0.03240462392568588, -0.008181077428162098, -0.027331333607435226, -0.05963212624192238, -0...
A DIRT-T Approach to Unsupervised Domain Adaptation
https://openreview.net/forum?id=H1q-TM-AW
[ "Rui Shu", "Hung Bui", "Hirokazu Narui", "Stefano Ermon" ]
Poster
null
Domain adaptation refers to the problem of leveraging labeled data in a source domain to learn an accurate model in a target domain where labels are scarce or unavailable. A recent approach for finding a common representation of the two domains is via domain adversarial training (Ganin & Lempitsky, 2015), which attempt...
[ "domain adaptation", "unsupervised learning", "semi-supervised learning" ]
SOTA on unsupervised domain adaptation by leveraging the cluster assumption.
989
1802.08735
title_snapshot
[ 0.003767917165532708, -0.02034858800470829, 0.00909724272787571, 0.05492107570171356, 0.0458969846367836, 0.004604430869221687, 0.0436326339840889, -0.0011603633174672723, 0.018970152363181114, -0.02887457050383091, -0.04024596884846687, 0.0475640669465065, -0.07449107617139816, 0.00498254...
Zero-Shot Visual Imitation
https://openreview.net/forum?id=BkisuzWRW
[ "Deepak Pathak", "Parsa Mahmoudieh", "Guanghao Luo", "Pulkit Agrawal", "Dian Chen", "Yide Shentu", "Evan Shelhamer", "Jitendra Malik", "Alexei A. Efros", "Trevor Darrell" ]
Oral
null
The current dominant paradigm for imitation learning relies on strong supervision of expert actions to learn both 'what' and 'how' to imitate. We pursue an alternative paradigm wherein an agent first explores the world without any expert supervision and then distills its experience into a goal-conditioned skill policy ...
[ "imitation", "zero-shot", "self-supervised", "robotics", "skills", "navigation", "manipulation", "vizdoom", "reinforcement" ]
Agents can learn to imitate solely visual demonstrations (without actions) at test time after learning from their own experience without any form of supervision at training time.
875
1804.08606
title_snapshot
[ 0.020898472517728806, -0.03280359506607056, -0.027245109900832176, 0.03952234238386154, 0.04440053552389145, 0.02872454561293125, 0.035602908581495285, 0.016009967774152756, -0.04927825182676315, -0.032892193645238876, -0.03811908885836601, 0.0012382923159748316, -0.07999849319458008, -0.0...
Parameter Space Noise for Exploration
https://openreview.net/forum?id=ByBAl2eAZ
[ "Matthias Plappert", "Rein Houthooft", "Prafulla Dhariwal", "Szymon Sidor", "Richard Y. Chen", "Xi Chen", "Tamim Asfour", "Pieter Abbeel", "Marcin Andrychowicz" ]
Poster
null
Deep reinforcement learning (RL) methods generally engage in exploratory behavior through noise injection in the action space. An alternative is to add noise directly to the agent's parameters, which can lead to more consistent exploration and a richer set of behaviors. Methods such as evolutionary strategies use param...
[ "reinforcement learning", "exploration", "parameter noise" ]
Parameter space noise allows reinforcement learning algorithms to explore by perturbing parameters instead of actions, often leading to significantly improved exploration performance.
379
1706.01905
title_snapshot
[ -0.0313863568007946, 0.010729112662374973, -0.028476882725954056, 0.04802047461271286, 0.03464715555310249, 0.05219421535730362, 0.035437654703855515, -0.010140317492187023, -0.024090271443128586, -0.06876775622367859, 0.010780610144138336, -0.009313510730862617, -0.0749584287405014, -0.02...
Compositional Attention Networks for Machine Reasoning
https://openreview.net/forum?id=S1Euwz-Rb
[ "Drew A. Hudson", "Christopher D. Manning" ]
Poster
null
We present Compositional Attention Networks, a novel fully differentiable neural network architecture, designed to facilitate explicit and expressive reasoning. While many types of neural networks are effective at learning and generalizing from massive quantities of data, this model moves away from monolithic black-box...
[ "Deep Learning", "Reasoning", "Memory", "Attention", "VQA", "CLEVR", "Recurrent Neural Networks", "Module Networks", "Compositionality" ]
We present a novel architecture, based on dynamic memory, attention and composition for the task of machine reasoning.
858
1803.03067
title_snapshot
[ 0.005047094076871872, 0.005789743270725012, 0.006343800574541092, 0.041581131517887115, 0.02171895280480385, 0.02846437320113182, 0.006657086778432131, 0.028081834316253662, -0.04856232553720474, -0.001023442018777132, -0.022021329030394554, 0.002402907470241189, -0.054260142147541046, 0.0...
Generalizing Across Domains via Cross-Gradient Training
https://openreview.net/forum?id=r1Dx7fbCW
[ "Shiv Shankar*", "Vihari Piratla*", "Soumen Chakrabarti", "Siddhartha Chaudhuri", "Preethi Jyothi", "Sunita Sarawagi" ]
Poster
null
We present CROSSGRAD , a method to use multi-domain training data to learn a classifier that generalizes to new domains. CROSSGRAD does not need an adaptation phase via labeled or unlabeled data, or domain features in the new domain. Most existing domain adaptation methods attempt to erase domain signals using techniqu...
[ "domain generalization", "domain adaptation", "adversarial learning", "adversarial examples" ]
Domain guided augmentation of data provides a robust and stable method of domain generalization
800
1804.10745
title_snapshot
[ -0.005917400121688843, -0.0011607954511418939, 0.023470312356948853, 0.048164255917072296, 0.0373663567006588, -0.002448130864650011, 0.028944285586476326, -0.028325259685516357, -0.007706543896347284, -0.024109888821840286, -0.007700618822127581, 0.02272988110780716, -0.08080699294805527, ...
Hierarchical Subtask Discovery with Non-Negative Matrix Factorization
https://openreview.net/forum?id=ry80wMW0W
[ "Adam C. Earle", "Andrew M. Saxe", "Benjamin Rosman" ]
Poster
null
Hierarchical reinforcement learning methods offer a powerful means of planning flexible behavior in complicated domains. However, learning an appropriate hierarchical decomposition of a domain into subtasks remains a substantial challenge. We present a novel algorithm for subtask discovery, based on the recently introd...
[ "Reinforcement Learning", "Hierarchy", "Subtask Discovery", "Linear Markov Decision Process" ]
We present a novel algorithm for hierarchical subtask discovery which leverages the multitask linear Markov decision process framework.
865
1708.00463
title_snapshot
[ -0.04751906916499138, -0.01611645147204399, 0.004061299376189709, 0.034991443157196045, 0.048812732100486755, 0.011167025193572044, 0.017163071781396866, -0.02531246654689312, -0.009829452261328697, -0.05373312160372734, -0.011779285967350006, 0.014878415502607822, -0.07242091000080109, -0...
Variational image compression with a scale hyperprior
https://openreview.net/forum?id=rkcQFMZRb
[ "Johannes Ballé", "David Minnen", "Saurabh Singh", "Sung Jin Hwang", "Nick Johnston" ]
Poster
null
We describe an end-to-end trainable model for image compression based on variational autoencoders. The model incorporates a hyperprior to effectively capture spatial dependencies in the latent representation. This hyperprior relates to side information, a concept universal to virtually all modern image codecs, but larg...
[]
null
885
1802.01436
title_snapshot
[ -0.010882779955863953, -0.03469809144735336, 0.001274710870347917, 0.03576952964067459, 0.035876352339982986, 0.06505818665027618, 0.011817521415650845, -0.024199027568101883, -0.04148614779114723, -0.045538529753685, -0.014577449299395084, -0.015285876579582691, -0.036490317434072495, 0.0...
Deep Active Learning for Named Entity Recognition
https://openreview.net/forum?id=ry018WZAZ
[ "Yanyao Shen", "Hyokun Yun", "Zachary C. Lipton", "Yakov Kronrod", "Animashree Anandkumar" ]
Poster
null
Deep learning has yielded state-of-the-art performance on many natural language processing tasks including named entity recognition (NER). However, this typically requires large amounts of labeled data. In this work, we demonstrate that the amount of labeled training data can be drastically reduced when deep learning i...
[ "active learning", "deep learning", "named entity recognition" ]
We introduce a lightweight architecture for named entity recognition and carry out incremental active learning, which is able to match state-of-the-art performance with just 25% of the original training data.
677
1707.05928
title_snapshot
[ -0.02741106040775776, -0.021768946200609207, -0.003287140280008316, 0.036268968135118484, 0.019911987707018852, 0.021090656518936157, -0.0010790728265419602, 0.006055411417037249, -0.003223604755476117, -0.02377711609005928, -0.04704652354121208, 0.003925845492631197, -0.05371290072798729, ...
A Compressed Sensing View of Unsupervised Text Embeddings, Bag-of-n-Grams, and LSTMs
https://openreview.net/forum?id=B1e5ef-C-
[ "Sanjeev Arora", "Mikhail Khodak", "Nikunj Saunshi", "Kiran Vodrahalli" ]
Poster
null
Low-dimensional vector embeddings, computed using LSTMs or simpler techniques, are a popular approach for capturing the “meaning” of text and a form of unsupervised learning useful for downstream tasks. However, their power is not theoretically understood. The current paper derives formal understanding by looking at th...
[ "theory", "LSTM", "unsupervised learning", "word embeddings", "compressed sensing", "sparse recovery", "document representation", "text classification" ]
We use the theory of compressed sensing to prove that LSTMs can do at least as well on linear text classification as Bag-of-n-Grams.
772
null
null
[ 0.003950120881199837, -0.039584532380104065, -0.014368887059390545, 0.03207596391439438, 0.03917819261550903, 0.01331399753689766, 0.044861193746328354, 0.012915513478219509, -0.028907308354973793, -0.05136602371931076, -0.0068093230947852135, -0.0012908446369692683, -0.061218928545713425, ...
Many Paths to Equilibrium: GANs Do Not Need to Decrease a Divergence At Every Step
https://openreview.net/forum?id=ByQpn1ZA-
[ "William Fedus*", "Mihaela Rosca*", "Balaji Lakshminarayanan", "Andrew M. Dai", "Shakir Mohamed", "Ian Goodfellow" ]
Poster
null
Generative adversarial networks (GANs) are a family of generative models that do not minimize a single training criterion. Unlike other generative models, the data distribution is learned via a game between a generator (the generative model) and a discriminator (a teacher providing training signal) that each minimize t...
[ "Deep learning", "GAN" ]
We find evidence that divergence minimization may not be an accurate characterization of GAN training.
521
1710.08446
title_snapshot
[ -0.03025548905134201, -0.025308338925242424, -0.009984683245420456, 0.0382014662027359, 0.011597112752497196, -0.007914785295724869, 0.0035062991082668304, 0.024840962141752243, -0.013943699188530445, -0.06349798291921616, -0.00585362920537591, 0.01866491325199604, -0.07068251073360443, 0....
Multi-Mention Learning for Reading Comprehension with Neural Cascades
https://openreview.net/forum?id=HyRnez-RW
[ "Swabha Swayamdipta", "Ankur P. Parikh", "Tom Kwiatkowski" ]
Poster
null
Reading comprehension is a challenging task, especially when executed across longer or across multiple evidence documents, where the answer is likely to reoccur. Existing neural architectures typically do not scale to the entire evidence, and hence, resort to selecting a single passage in the document (either via trunc...
[ "reading comprehension", "multi-loss", "question answering", "scalable", "TriviaQA", "feed-forward", "latent variable", "attention" ]
We propose neural cascades, a simple and trivially parallelizable approach to reading comprehension, consisting only of feed-forward nets and attention that achieves state-of-the-art performance on the TriviaQA dataset.
774
1711.00894
title_snapshot
[ -0.00962012354284525, -0.010871524922549725, -0.0042268275283277035, 0.037999529391527176, 0.038537122309207916, 0.003581506200134754, 0.014453443698585033, 0.02525196596980095, -0.021913280710577965, -0.005274366587400436, 0.01499966625124216, 0.04100198671221733, -0.035854801535606384, -...
Skip RNN: Learning to Skip State Updates in Recurrent Neural Networks
https://openreview.net/forum?id=HkwVAXyCW
[ "Víctor Campos", "Brendan Jou", "Xavier Giró-i-Nieto", "Jordi Torres", "Shih-Fu Chang" ]
Poster
null
Recurrent Neural Networks (RNNs) continue to show outstanding performance in sequence modeling tasks. However, training RNNs on long sequences often face challenges like slow inference, vanishing gradients and difficulty in capturing long term dependencies. In backpropagation through time settings, these issues are ti...
[ "recurrent neural networks", "dynamic learning", "conditional computation" ]
A modification for existing RNN architectures which allows them to skip state updates while preserving the performance of the original architectures.
121
1708.06834
title_snapshot
[ -0.023903843015432358, -0.03379189223051071, -0.03744148463010788, 0.05042993277311325, 0.06655604392290115, 0.04324808344244957, 0.03384914994239807, 0.030534422025084496, -0.04487430304288864, -0.04005701467394829, 0.018234675750136375, 0.01792498491704464, -0.06273780018091202, -0.00596...
Dynamic Neural Program Embeddings for Program Repair
https://openreview.net/forum?id=BJuWrGW0Z
[ "Ke Wang", "Rishabh Singh", "Zhendong Su" ]
Poster
null
Neural program embeddings have shown much promise recently for a variety of program analysis tasks, including program synthesis, program repair, code completion, and fault localization. However, most existing program embeddings are based on syntactic features of programs, such as token sequences or abstract syntax tree...
[ "Program Embedding", "Program Semantics", "Dynamic Traces" ]
A new way of learning semantic program embedding
832
null
null
[ -0.0062029920518398285, -0.029269134625792503, -0.03685355931520462, 0.04462466016411781, 0.051115166395902634, 0.043954066932201385, 0.03973567485809326, 0.006409587804228067, -0.012264820747077465, -0.037470776587724686, -0.019919799640774727, 0.02561819739639759, -0.05064866319298744, -...
Can recurrent neural networks warp time?
https://openreview.net/forum?id=SJcKhk-Ab
[ "Corentin Tallec", "Yann Ollivier" ]
Poster
null
Successful recurrent models such as long short-term memories (LSTMs) and gated recurrent units (GRUs) use \emph{ad hoc} gating mechanisms. Empirically these models have been found to improve the learning of medium to long term temporal dependencies and to help with vanishing gradient issues. We prove that learnable ...
[ "RNN" ]
Proves that gating mechanisms provide invariance to time transformations. Introduces and tests a new initialization for LSTMs from this insight.
519
1804.11188
title_snapshot
[ -0.01052321307361126, -0.008533060550689697, -0.008297155611217022, 0.035981908440589905, 0.029277049005031586, 0.021576115861535072, 0.04165799915790558, 0.023892894387245178, -0.03950163349509239, -0.012658541090786457, 0.0053758383728563786, -0.015806809067726135, -0.0697450339794159, 0...
Consequentialist conditional cooperation in social dilemmas with imperfect information
https://openreview.net/forum?id=BkabRiQpb
[ "Alexander Peysakhovich", "Adam Lerer" ]
Poster
null
Social dilemmas, where mutual cooperation can lead to high payoffs but participants face incentives to cheat, are ubiquitous in multi-agent interaction. We wish to construct agents that cooperate with pure cooperators, avoid exploitation by pure defectors, and incentivize cooperation from the rest. However, often the a...
[ "deep reinforcement learning", "cooperation", "social dilemma", "multi-agent systems" ]
We show how to use deep RL to construct agents that can solve social dilemmas beyond matrix games.
10
1710.06975
title_snapshot
[ -0.016097547486424446, -0.025575701147317886, -0.028160132467746735, 0.06059948727488518, 0.03406580165028572, 0.022435378283262253, 0.020692268386483192, 0.0009409486665390432, -0.02182122692465782, -0.032705482095479965, -0.03608406335115433, 0.013891905546188354, -0.06412049382925034, -...
Generating Natural Adversarial Examples
https://openreview.net/forum?id=H1BLjgZCb
[ "Zhengli Zhao", "Dheeru Dua", "Sameer Singh" ]
Poster
null
Due to their complex nature, it is hard to characterize the ways in which machine learning models can misbehave or be exploited when deployed. Recent work on adversarial examples, i.e. inputs with minor perturbations that result in substantially different model predictions, is helpful in evaluating the robustness of th...
[ "adversarial examples", "generative adversarial networks", "interpretability", "image classification", "textual entailment", "machine translation" ]
We propose a framework to generate “natural” adversaries against black-box classifiers for both visual and textual domains, by doing the search for adversaries in the latent semantic space.
623
1710.11342
title_snapshot
[ 0.0014956099912524223, -0.04005393758416176, -0.02822619490325451, 0.07055895775556564, 0.024509049952030182, 0.019917527213692665, 0.03713512420654297, 0.006931832060217857, -0.030682221055030823, -0.024937007576227188, -0.03589572757482529, 0.013584356755018234, -0.07818930596113205, 0.0...
Large Scale Optimal Transport and Mapping Estimation
https://openreview.net/forum?id=B1zlp1bRW
[ "Vivien Seguy", "Bharath Bhushan Damodaran", "Remi Flamary", "Nicolas Courty", "Antoine Rolet", "Mathieu Blondel" ]
Poster
null
This paper presents a novel two-step approach for the fundamental problem of learning an optimal map from one distribution to another. First, we learn an optimal transport (OT) plan, which can be thought as a one-to-many map between the two distributions. To that end, we propose a stochastic dual approach of regularize...
[ "optimal transport", "Wasserstein", "domain adaptation", "generative models", "Monge map", "optimal mapping" ]
Learning optimal mapping with deepNN between distributions along with theoretical guarantees.
523
1711.02283
title_snapshot
[ -0.0284099280834198, -0.025192847475409508, 0.023416150361299515, 0.02952823042869568, 0.045806314796209335, 0.03231147676706314, 0.014066989533603191, 0.005738952197134495, -0.021791093051433563, -0.07268889993429184, 0.014349124394357204, -0.022638937458395958, -0.061588600277900696, 0.0...
Stabilizing Adversarial Nets with Prediction Methods
https://openreview.net/forum?id=Skj8Kag0Z
[ "Abhay Yadav", "Sohil Shah", "Zheng Xu", "David Jacobs", "Tom Goldstein" ]
Poster
null
Adversarial neural networks solve many important problems in data science, but are notoriously difficult to train. These difficulties come from the fact that optimal weights for adversarial nets correspond to saddle points, and not minimizers, of the loss function. The alternating stochastic gradient methods typically ...
[ "adversarial networks", "optimization" ]
We present a simple modification to the alternating SGD method, called a prediction step, that improves the stability of adversarial networks.
422
1705.07364
title_snapshot
[ -0.018134664744138718, -0.03310157358646393, -0.03429484739899635, 0.039744868874549866, 0.024247540161013603, 0.021305667236447334, 0.030527910217642784, -0.030331967398524284, -0.0441623255610466, -0.050612110644578934, -0.01560080237686634, -0.014384807087481022, -0.07907551527023315, -...
SEARNN: Training RNNs with global-local losses
https://openreview.net/forum?id=HkUR_y-RZ
[ "Rémi Leblond", "Jean-Baptiste Alayrac", "Anton Osokin", "Simon Lacoste-Julien" ]
Poster
null
We propose SEARNN, a novel training algorithm for recurrent neural networks (RNNs) inspired by the "learning to search" (L2S) approach to structured prediction. RNNs have been widely successful in structured prediction applications such as machine translation or parsing, and are commonly trained using maximum likelihoo...
[ "Structured prediction", "RNNs" ]
We introduce SeaRNN, a novel algorithm for RNN training, inspired by the learning to search approach to structured prediction, in order to avoid the limitations of MLE training.
496
1706.04499
title_snapshot
[ -0.005389773286879063, -0.020387306809425354, -0.01211018767207861, 0.04646391421556473, 0.037856943905353546, 0.04764872044324875, 0.026561936363577843, 0.013771715573966503, -0.026113169267773628, -0.013699604198336601, 0.0027765247505158186, 0.02024977095425129, -0.043887246400117874, -...
Simulated+Unsupervised Learning With Adaptive Data Generation and Bidirectional Mappings
https://openreview.net/forum?id=SkHDoG-Cb
[ "Kangwook Lee", "Hoon Kim", "Changho Suh" ]
Poster
null
Collecting a large dataset with high quality annotations is expensive and time-consuming. Recently, Shrivastava et al. (2017) propose Simulated+Unsupervised (S+U) learning: It first learns a mapping from synthetic data to real data, translates a large amount of labeled synthetic data to the ones that resemble real data...
[]
null
930
null
null
[ -0.022293943911790848, -0.04886902868747711, 0.0035488333087414503, 0.012972346507012844, 0.04526929184794426, 0.025042608380317688, 0.029018228873610497, 0.014887232333421707, -0.013196557760238647, -0.04003308713436127, -0.03574481979012489, 0.013867473229765892, -0.06304985284805298, 0....
Beyond Shared Hierarchies: Deep Multitask Learning through Soft Layer Ordering
https://openreview.net/forum?id=BkXmYfbAZ
[ "Elliot Meyerson", "Risto Miikkulainen" ]
Poster
null
Existing deep multitask learning (MTL) approaches align layers shared between tasks in a parallel ordering. Such an organization significantly constricts the types of shared structure that can be learned. The necessity of parallel ordering for deep MTL is first tested by comparing it with permuted ordering of shared la...
[ "multitask learning", "deep learning", "modularity" ]
Relaxing the constraint of shared hierarchies enables more effective deep multitask learning.
884
1711.00108
title_snapshot
[ -0.02152295783162117, -0.025728626176714897, -0.008271360769867897, 0.030436113476753235, 0.03811059519648552, 0.010677087120711803, 0.016536373645067215, 0.03232228755950928, -0.0228439848870039, -0.0494995042681694, -0.00741738174110651, 0.017928315326571465, -0.07671420276165009, 0.0038...
Towards better understanding of gradient-based attribution methods for Deep Neural Networks
https://openreview.net/forum?id=Sy21R9JAW
[ "Marco Ancona", "Enea Ceolini", "Cengiz Öztireli", "Markus Gross" ]
Poster
null
Understanding the flow of information in Deep Neural Networks (DNNs) is a challenging problem that has gain increasing attention over the last few years. While several methods have been proposed to explain network predictions, there have been only a few attempts to compare them from a theoretical perspective. What is m...
[ "Deep Neural Networks", "Attribution methods", "Theory of deep learning" ]
Four existing backpropagation-based attribution methods are fundamentally similar. How to assess it?
151
1711.06104
title_snapshot
[ -0.006794000510126352, -0.015702569857239723, -0.012490397319197655, 0.04112550988793373, 0.03210880607366562, 0.05262797325849533, 0.006872903555631638, -0.00848926417529583, -0.010152164846658707, -0.053909752517938614, -0.015263029374182224, -0.007930207066237926, -0.060246389359235764, ...
Unsupervised Machine Translation Using Monolingual Corpora Only
https://openreview.net/forum?id=rkYTTf-AZ
[ "Guillaume Lample", "Alexis Conneau", "Ludovic Denoyer", "Marc'Aurelio Ranzato" ]
Poster
null
Machine translation has recently achieved impressive performance thanks to recent advances in deep learning and the availability of large-scale parallel corpora. There have been numerous attempts to extend these successes to low-resource language pairs, yet requiring tens of thousands of parallel sentences. In this wor...
[ "unsupervised", "machine translation", "adversarial" ]
We propose a new unsupervised machine translation model that can learn without using parallel corpora; experimental results show impressive performance on multiple corpora and pairs of languages.
1,019
1711.00043
title_snapshot
[ 0.0042992630042135715, -0.037707455456256866, -0.03153203800320625, 0.04285392165184021, 0.0332019180059433, 0.033310603350400925, 0.03360942378640175, 0.035877905786037445, 0.0016716765239834785, -0.013603519648313522, -0.00990345235913992, 0.044644176959991455, -0.058033909648656845, 0.0...
NerveNet: Learning Structured Policy with Graph Neural Networks
https://openreview.net/forum?id=S1sqHMZCb
[ "Tingwu Wang", "Renjie Liao", "Jimmy Ba", "Sanja Fidler" ]
Poster
null
We address the problem of learning structured policies for continuous control. In traditional reinforcement learning, policies of agents are learned by MLPs which take the concatenation of all observations from the environment as input for predicting actions. In this work, we propose NerveNet to explicitly model the st...
[ "reinforcement learning", "transfer learning", "graph neural network" ]
using graph neural network to model structural information of the agents to improve policy and transferability
840
null
null
[ -0.031814880669116974, -0.04787299409508705, -0.01135520450770855, 0.06065880134701729, 0.04276696592569351, 0.02300621196627617, 0.010437287390232086, -0.010339912958443165, -0.030156804248690605, -0.03607737645506859, 0.013379179872572422, 0.008350878022611141, -0.07603050768375397, -0.0...
Leave no Trace: Learning to Reset for Safe and Autonomous Reinforcement Learning
https://openreview.net/forum?id=S1vuO-bCW
[ "Benjamin Eysenbach", "Shixiang Gu", "Julian Ibarz", "Sergey Levine" ]
Poster
null
Deep reinforcement learning algorithms can learn complex behavioral skills, but real-world application of these methods requires a considerable amount of experience to be collected by the agent. In practical settings, such as robotics, this involves repeatedly attempting a task, resetting the environment between each a...
[ "manual reset", "continual learning", "reinforcement learning", "safety" ]
We propose an autonomous method for safe and efficient reinforcement learning that simultaneously learns a forward and backward policy, with the backward policy resetting the environment for a subsequent attempt.
698
1711.06782
title_snapshot
[ -0.02483801729977131, -0.01870957389473915, -0.012697427533566952, 0.03937038034200668, 0.06739314645528793, -0.006908354349434376, 0.02704700268805027, -0.022467827424407005, -0.036443520337343216, -0.05623408034443855, -0.01688099093735218, 0.020118586719036102, -0.049315378069877625, -0...
Neural Language Modeling by Jointly Learning Syntax and Lexicon
https://openreview.net/forum?id=rkgOLb-0W
[ "Yikang Shen", "Zhouhan Lin", "Chin-wei Huang", "Aaron Courville" ]
Poster
null
We propose a neural language model capable of unsupervised syntactic structure induction. The model leverages the structure information to form better semantic representations and better language modeling. Standard recurrent neural networks are limited by their structure and fail to efficiently use syntactic informatio...
[ "Language model", "unsupervised parsing" ]
In this paper, We propose a novel neural language model, called the Parsing-Reading-Predict Networks (PRPN), that can simultaneously induce the syntactic structure from unannotated sentences and leverage the inferred structure to learn a better language model.
679
1711.02013
title_snapshot
[ -0.027808409184217453, 0.008215625770390034, -0.012935223989188671, 0.024062465876340866, 0.03110850602388382, 0.056330204010009766, 0.02225366048514843, 0.01317279227077961, -0.02007007598876953, -0.01070780586451292, 0.009610683657228947, 0.015847237780690193, -0.04182937741279602, 0.002...
Adaptive Dropout with Rademacher Complexity Regularization
https://openreview.net/forum?id=S1uxsye0Z
[ "Ke Zhai", "Huan Wang" ]
Poster
null
We propose a novel framework to adaptively adjust the dropout rates for the deep neural network based on a Rademacher complexity bound. The state-of-the-art deep learning algorithms impose dropout strategy to prevent feature co-adaptation. However, choosing the dropout rates remains an art of heuristics or relies on em...
[ "model complexity", "regularization", "deep learning", "model generalization", "adaptive dropout" ]
We propose a novel framework to adaptively adjust the dropout rates for the deep neural network based on a Rademacher complexity bound.
204
null
null
[ -0.033227868378162384, -0.016539307311177254, 0.0030210022814571857, 0.0373014397919178, 0.03708600625395775, 0.023905102163553238, 0.024615507572889328, 0.00019309182243887335, -0.01986396685242653, -0.03247940540313721, -0.03866755962371826, 0.0048600067384541035, -0.04217035695910454, 0...
Implicit Causal Models for Genome-wide Association Studies
https://openreview.net/forum?id=SyELrEeAb
[ "Dustin Tran", "David M. Blei" ]
Poster
null
Progress in probabilistic generative models has accelerated, developing richer models with neural architectures, implicit densities, and with scalable algorithms for their Bayesian inference. However, there has been limited progress in models that capture causal relationships, for example, how individual genetic factor...
[]
Implicit models applied to causality and genetics
236
1710.10742
title_snapshot
[ 0.009961416013538837, -0.013727167621254921, -0.02794981747865677, 0.030475430190563202, 0.025245551019906998, 0.0315653532743454, 0.05600700154900551, 0.024175088852643967, -0.002033255761489272, -0.030439060181379318, -0.00224944856017828, -0.002999379998072982, -0.06945473700761795, 0.0...
Training GANs with Optimism
https://openreview.net/forum?id=SJJySbbAZ
[ "Constantinos Daskalakis", "Andrew Ilyas", "Vasilis Syrgkanis", "Haoyang Zeng" ]
Poster
null
We address the issue of limit cycling behavior in training Generative Adversarial Networks and propose the use of Optimistic Mirror Decent (OMD) for training Wasserstein GANs. Recent theoretical results have shown that optimistic mirror decent (OMD) can enjoy faster regret rates in the context of zero-sum games. WGANs ...
[ "GANs", "Optimistic Mirror Decent", "Cycling", "Last Iterate Convergence", "Optimistic Adam" ]
We propose the use of optimistic mirror decent to address cycling problems in the training of GANs. We also introduce the Optimistic Adam algorithm
669
1711.00141
title_snapshot
[ -0.03719209507107735, -0.03980572149157524, 0.00878847111016512, 0.0436563715338707, 0.009365049190819263, 0.023234717547893524, 0.014225772581994534, 0.037758950144052505, -0.0031132043804973364, -0.05298280715942383, 0.01137553621083498, 0.005126538220793009, -0.08264444023370743, -0.018...
Latent Space Oddity: on the Curvature of Deep Generative Models
https://openreview.net/forum?id=SJzRZ-WCZ
[ "Georgios Arvanitidis", "Lars Kai Hansen", "Søren Hauberg" ]
Poster
null
Deep generative models provide a systematic way to learn nonlinear data distributions through a set of latent variables and a nonlinear "generator" function that maps latent points into the input space. The nonlinearity of the generator implies that the latent space gives a distorted view of the input space. Under mild...
[ "Generative models", "Riemannian Geometry", "Latent Space" ]
null
657
1710.11379
title_snapshot
[ 0.00016783870523795485, -0.011893272399902344, -0.017223872244358063, 0.06505440175533295, 0.029646916314959526, 0.035289742052555084, 0.013609768822789192, 0.02384060062468052, -0.011750830337405205, -0.059132326394319534, -0.019561901688575745, -0.024478472769260406, -0.0469035878777504, ...
Learning Awareness Models
https://openreview.net/forum?id=r1HhRfWRZ
[ "Brandon Amos", "Laurent Dinh", "Serkan Cabi", "Thomas Rothörl", "Sergio Gómez Colmenarejo", "Alistair Muldal", "Tom Erez", "Yuval Tassa", "Nando de Freitas", "Misha Denil" ]
Poster
null
We consider the setting of an agent with a fixed body interacting with an unknown and uncertain external world. We show that models trained to predict proprioceptive information about the agent's body come to represent objects in the external world. In spite of being trained with only internally available signals, thes...
[ "Awareness", "Prediction", "Seq2seq", "Robots" ]
We train predictive models on proprioceptive information and show they represent properties of external objects.
1,065
1804.06318
title_snapshot
[ -0.008488968014717102, 0.0040456741116940975, -0.010371919721364975, 0.01525131519883871, 0.021604962646961212, 0.02263791859149933, 0.019474074244499207, 0.03207540512084961, -0.06403021514415741, -0.03002336248755455, -0.032031185925006866, 0.015288494527339935, -0.07975646108388901, -0....
Monotonic Chunkwise Attention
https://openreview.net/forum?id=Hko85plCW
[ "Chung-Cheng Chiu*", "Colin Raffel*" ]
Poster
null
Sequence-to-sequence models with soft attention have been successfully applied to a wide variety of problems, but their decoding process incurs a quadratic time and space cost and is inapplicable to real-time sequence transduction. To address these issues, we propose Monotonic Chunkwise Attention (MoChA), which adaptiv...
[ "attention", "sequence-to-sequence", "speech recognition", "document summarization" ]
An online and linear-time attention mechanism that performs soft attention over adaptively-located chunks of the input sequence.
425
1712.05382
title_snapshot
[ -0.036388132721185684, -0.023820875212550163, -0.033271148800849915, 0.03911888599395752, 0.005930671934038401, 0.041998784989118576, 0.04039309546351433, 0.05071166530251503, -0.019237106665968895, -0.007028611842542887, -0.026210570707917213, 0.02581816539168358, -0.05894673243165016, -0...
Emergent Communication in a Multi-Modal, Multi-Step Referential Game
https://openreview.net/forum?id=rJGZq6g0-
[ "Katrina Evtimova", "Andrew Drozdov", "Douwe Kiela", "Kyunghyun Cho" ]
Poster
null
Inspired by previous work on emergent communication in referential games, we propose a novel multi-modal, multi-step referential game, where the sender and receiver have access to distinct modalities of an object, and their information exchange is bidirectional and of arbitrary duration. The multi-modal multi-step set...
[ "emergent communication", "multi-agent systems", "multi-modal" ]
null
423
1705.10369
title_snapshot
[ -0.04946418106555939, -0.020475195720791817, -0.004161825869232416, 0.022907234728336334, 0.02594763971865177, 0.003486666362732649, 0.031239522621035576, 0.03770267963409424, -0.019294895231723785, -0.03436419367790222, -0.035459913313388824, 0.04918042570352554, -0.07341652363538742, -0....
Attacking Binarized Neural Networks
https://openreview.net/forum?id=HkTEFfZRb
[ "Angus Galloway", "Graham W. Taylor", "Medhat Moussa" ]
Poster
null
Neural networks with low-precision weights and activations offer compelling efficiency advantages over their full-precision equivalents. The two most frequently discussed benefits of quantization are reduced memory consumption, and a faster forward pass when implemented with efficient bitwise operations. We propose a t...
[ "adversarial examples", "adversarial attacks", "binary", "binarized neural networks" ]
We conduct adversarial attacks against binarized neural networks and show that we reduce the impact of the strongest attacks, while maintaining comparable accuracy in a black-box setting
887
1711.00449
title_snapshot
[ -0.0067796604707837105, 0.00015548785449936986, -0.022735847160220146, 0.061744723469018936, 0.01812700554728508, 0.023150864988565445, 0.021415648981928825, -0.007993994280695915, -0.02797827124595642, -0.02374536357820034, 0.014920737594366074, -0.01786622777581215, -0.04807354137301445, ...
The Kanerva Machine: A Generative Distributed Memory
https://openreview.net/forum?id=S1HlA-ZAZ
[ "Yan Wu", "Greg Wayne", "Alex Graves", "Timothy Lillicrap" ]
Poster
null
We present an end-to-end trained memory system that quickly adapts to new data and generates samples like them. Inspired by Kanerva's sparse distributed memory, it has a robust distributed reading and writing mechanism. The memory is analytically tractable, which enables optimal on-line compression via a Bayesian upda...
[ "memory", "generative model", "inference", "neural network", "hierarchical model" ]
A generative memory model that combines slow-learning neural networks and a fast-adapting linear Gaussian model as memory.
738
1804.01756
title_snapshot
[ -0.026674514636397362, 0.007016276009380817, -0.026059182360768318, 0.037363432347774506, 0.031156102195382118, 0.016104137524962425, 0.007193251978605986, 0.029058312997221947, -0.027810612693428993, -0.040678489953279495, 0.009239043109118938, 0.0011397839989513159, -0.06979843974113464, ...
Learning how to explain neural networks: PatternNet and PatternAttribution
https://openreview.net/forum?id=Hkn7CBaTW
[ "Pieter-Jan Kindermans", "Kristof T. Schütt", "Maximilian Alber", "Klaus-Robert Müller", "Dumitru Erhan", "Been Kim", "Sven Dähne" ]
Poster
null
DeConvNet, Guided BackProp, LRP, were invented to better understand deep neural networks. We show that these methods do not produce the theoretically correct explanation for a linear model. Yet they are used on multi-layer networks with millions of parameters. This is a cause for concern since linear models are simple ...
[ "machine learning", "interpretability", "deep learning" ]
Without learning, it is impossible to explain a machine learning model's decisions.
72
1705.05598
title_snapshot
[ -0.0023759803734719753, -0.014026708900928497, -0.032706037163734436, 0.02841593138873577, 0.04907620698213577, 0.027963116765022278, 0.007664922159165144, -0.017526770010590553, -0.020493721589446068, -0.050170764327049255, -0.009507541544735432, 0.035037752240896225, -0.05612095072865486, ...