title
stringlengths
14
131
paper_url
stringlengths
42
42
authors
listlengths
1
15
type
stringclasses
3 values
primary_area
stringclasses
0 values
abstract
large_stringlengths
332
2.13k
keywords
listlengths
0
21
TL;DR
large_stringlengths
31
250
submission_number
int64
1
2.59k
arxiv_id
stringlengths
10
10
arxiv_id_source
stringclasses
3 values
embedding
listlengths
768
768
Pay Attention to Features, Transfer Learn Faster CNNs
https://openreview.net/forum?id=ryxyCeHtPB
[ "Kafeng Wang", "Xitong Gao", "Yiren Zhao", "Xingjian Li", "Dejing Dou", "Cheng-Zhong Xu" ]
Poster
null
Deep convolutional neural networks are now widely deployed in vision applications, but a limited size of training data can restrict their task performance. Transfer learning offers the chance for CNNs to learn with limited data samples by transferring knowledge from models pretrained on large datasets. Blindly transfer...
[ "transfer learning", "pruning", "faster CNNs" ]
We introduce attentive feature distillation and selection, to fine-tune a large model and produce a faster one.
2,594
null
null
[ 0.00011319012992316857, -0.011161099188029766, -0.0043113455176353455, 0.037016600370407104, 0.04635660722851753, 0.021795248612761497, 0.02420152723789215, 0.0011481110705062747, -0.014288525097072124, -0.05142324045300484, -0.015698255971074104, -0.006257934961467981, -0.05211871489882469,...
Geom-GCN: Geometric Graph Convolutional Networks
https://openreview.net/forum?id=S1e2agrFvS
[ "Hongbin Pei", "Bingzhe Wei", "Kevin Chen-Chuan Chang", "Yu Lei", "Bo Yang" ]
Spotlight
null
Message-passing neural networks (MPNNs) have been successfully applied in a wide variety of applications in the real world. However, two fundamental weaknesses of MPNNs' aggregators limit their ability to represent graph-structured data: losing the structural information of nodes in neighborhoods and lacking the abilit...
[ "Deep Learning", "Graph Convolutional Network", "Network Geometry" ]
For graph neural networks, the aggregation on a graph can benefit from a continuous space underlying the graph.
2,589
2002.05287
title_snapshot
[ -0.022369572892785072, -0.037557296454906464, 0.024246210232377052, 0.03311940282583237, 0.019035547971725464, 0.02356594242155552, 0.012617376632988453, 0.032519273459911346, -0.01620050147175789, -0.056010641157627106, 0.03653423860669136, -0.05004556477069855, -0.09089189022779465, 0.03...
Gradients as Features for Deep Representation Learning
https://openreview.net/forum?id=BkeoaeHKDS
[ "Fangzhou Mu", "Yingyu Liang", "Yin Li" ]
Poster
null
We address the challenging problem of deep representation learning -- the efficient adaption of a pre-trained deep network to different tasks. Specifically, we propose to explore gradient-based features. These features are gradients of the model parameters with respect to a task-specific loss given an input sample. Our...
[ "representation learning", "gradient features", "deep learning" ]
Given a pre-trained model, we explored the per-sample gradients of the model parameters relative to a task-specific loss, and constructed a linear model that combines gradients of model parameters and the activation of the model.
2,585
2004.05529
title_snapshot
[ -0.026539532467722893, -0.013044465333223343, -0.0009224280365742743, 0.03372218832373619, 0.03330638259649277, 0.04379086196422577, 0.01625876873731613, -0.011249647475779057, -0.020413046702742577, -0.02127409353852272, -0.023817572742700577, -0.022434424608945847, -0.05838288739323616, ...
Monotonic Multihead Attention
https://openreview.net/forum?id=Hyg96gBKPS
[ "Xutai Ma", "Juan Miguel Pino", "James Cross", "Liezl Puzon", "Jiatao Gu" ]
Poster
null
Simultaneous machine translation models start generating a target sequence before they have encoded or read the source sequence. Recent approach for this task either apply a fixed policy on transformer, or a learnable monotonic attention on a weaker recurrent neural network based structure. In this paper, we propose a ...
[ "Simultaneous Translation", "Transformer", "Monotonic Attention" ]
Make the transformer streamable with monotonic attention.
2,583
1909.12406
title_snapshot
[ -0.0023034238256514072, 0.009146862663328648, -0.02021813578903675, 0.0506080724298954, -0.01395819615572691, 0.042778484523296356, 0.015966057777404785, 0.02350863441824913, -0.03017354942858219, 0.0014085060684010386, -0.004318230785429478, 0.03649519011378288, -0.0460813008248806, -0.00...
Massively Multilingual Sparse Word Representations
https://openreview.net/forum?id=HyeYTgrFPB
[ "Gábor Berend" ]
Poster
null
In this paper, we introduce Mamus for constructing multilingual sparse word representations. Our algorithm operates by determining a shared set of semantic units which get reutilized across languages, providing it a competitive edge both in terms of speed and evaluation performance. We demonstrate that our proposed alg...
[ "sparse word representations", "multilinguality", "sparse coding" ]
We propose an efficient algorithm for determining multilingually comparable sparse word representations that we release for 27 typologically diverse languages.
2,582
null
null
[ 0.008700262755155563, -0.017689015716314316, -0.006025587674230337, 0.05240762606263161, 0.023483173921704292, 0.06322188675403595, 0.015155725181102753, 0.026479842141270638, -0.032584048807621, -0.018815215677022934, 0.000198150853975676, 0.015142600052058697, -0.08476731181144714, 0.014...
Query-efficient Meta Attack to Deep Neural Networks
https://openreview.net/forum?id=Skxd6gSYDS
[ "Jiawei Du", "Hu Zhang", "Joey Tianyi Zhou", "Yi Yang", "Jiashi Feng" ]
Poster
null
Black-box attack methods aim to infer suitable attack patterns to targeted DNN models by only using output feedback of the models and the corresponding input queries. However, due to lack of prior and inefficiency in leveraging the query and feedback information, existing methods are mostly query-intensive for obtainin...
[ "Adversarial attack", "Meta learning" ]
null
2,580
1906.02398
title_snapshot
[ 0.0019437030423432589, -0.01715739071369171, -0.009719657711684704, 0.06259407848119736, 0.017568597570061684, 0.014480584301054478, 0.03568702191114426, -0.0045343199744820595, -0.032136932015419006, -0.017847184091806412, -0.0022091821301728487, -0.004038456827402115, -0.04541134089231491,...
BREAKING CERTIFIED DEFENSES: SEMANTIC ADVERSARIAL EXAMPLES WITH SPOOFED ROBUSTNESS CERTIFICATES
https://openreview.net/forum?id=HJxdTxHYvB
[ "Amin Ghiasi", "Ali Shafahi", "Tom Goldstein" ]
Poster
null
Defenses against adversarial attacks can be classified into certified and non-certified. Certifiable defenses make networks robust within a certain $\ell_p$-bounded radius, so that it is impossible for the adversary to make adversarial examples in the certificate bound. We present an attack that maintains the impercept...
[]
null
2,579
2003.08937
title_snapshot
[ 0.006214930210262537, -0.03481252118945122, -0.011238540522754192, 0.06243763491511345, 0.025882437825202942, -0.01744859106838703, 0.045598916709423065, -0.018232911825180054, -0.01672721654176712, -0.010071182623505592, -0.002624280983582139, -0.0027098809368908405, -0.05406185984611511, ...
An Exponential Learning Rate Schedule for Deep Learning
https://openreview.net/forum?id=rJg8TeSFDH
[ "Zhiyuan Li", "Sanjeev Arora" ]
Spotlight
null
Intriguing empirical evidence exists that deep learning can work well with exotic schedules for varying the learning rate. This paper suggests that the phenomenon may be due to Batch Normalization or BN(Ioffe & Szegedy, 2015), which is ubiq- uitous and provides benefits in optimization and generalization across all sta...
[ "batch normalization", "weight decay", "learning rate", "deep learning theory" ]
We propose an exponentially growing learning rate schedule for networks with BatchNorm, which surprisingly performs well in practice and is provably equivalent to popular LR schedules like Step Decay.
2,575
1910.07454
title_snapshot
[ -0.01155828032642603, -0.02266721799969673, -0.023412546142935753, 0.038279496133327484, 0.03890354186296463, 0.009610704146325588, 0.03376529738306999, -0.0011586234904825687, -0.008495238609611988, -0.040919095277786255, 0.015875471755862236, -0.00928691029548645, -0.035734713077545166, ...
Enabling Deep Spiking Neural Networks with Hybrid Conversion and Spike Timing Dependent Backpropagation
https://openreview.net/forum?id=B1xSperKvH
[ "Nitin Rathi", "Gopalakrishnan Srinivasan", "Priyadarshini Panda", "Kaushik Roy" ]
Poster
null
Spiking Neural Networks (SNNs) operate with asynchronous discrete events (or spikes) which can potentially lead to higher energy-efficiency in neuromorphic hardware implementations. Many works have shown that an SNN for inference can be formed by copying the weights from a trained Artificial Neural Network (ANN) and se...
[ "spiking neural networks", "ann-snn conversion", "spike-based backpropagation", "imagenet" ]
A hybrid training technique that combines ANN-SNN conversion and spike-based backpropagation to optimize training effort and inference latency.
2,573
2005.01807
title_snapshot
[ -0.015766967087984085, -0.017414234578609467, -0.008585933595895767, 0.038413871079683304, 0.03567144274711609, 0.025627655908465385, 0.015470203012228012, 0.006481636315584183, -0.034704092890024185, -0.021514251828193665, 0.02493009902536869, -0.029962539672851562, -0.04904923960566521, ...
How to 0wn the NAS in Your Spare Time
https://openreview.net/forum?id=S1erpeBFPB
[ "Sanghyun Hong", "Michael Davinroy", "Yiǧitcan Kaya", "Dana Dachman-Soled", "Tudor Dumitraş" ]
Poster
null
New data processing pipelines and novel network architectures increasingly drive the success of deep learning. In consequence, the industry considers top-performing architectures as intellectual property and devotes considerable computational resources to discovering such architectures through neural architecture searc...
[ "Reconstructing Novel Deep Learning Systems" ]
We design an algorithm that reconstructs the key components of a novel deep learning system by exploiting a small amount of information leakage from a cache side-channel attack, Flush+Reload.
2,572
2002.06776
title_judge
[ 0.008430312387645245, -0.04245242103934288, -0.04266255721449852, 0.06005793437361717, 0.0633644089102745, 0.027725882828235626, 0.02424759231507778, 0.009191399440169334, -0.003662731032818556, -0.025997145101428032, 0.041007813066244125, -0.02906830608844757, -0.027478720992803574, 0.014...
The Shape of Data: Intrinsic Distance for Data Distributions
https://openreview.net/forum?id=HyebplHYwB
[ "Anton Tsitsulin", "Marina Munkhoeva", "Davide Mottin", "Panagiotis Karras", "Alex Bronstein", "Ivan Oseledets", "Emmanuel Mueller" ]
Poster
null
The ability to represent and compare machine learning models is crucial in order to quantify subtle model changes, evaluate generative models, and gather insights on neural network architectures. Existing techniques for comparing data distributions focus on global data properties such as mean and covariance; in that se...
[ "Deep Learning", "Generative Models", "Nonlinear Dimensionality Reduction", "Manifold Learning", "Similarity and Distance Learning", "Spectral Methods" ]
We propose a metric for comparing data distributions based on their geometry while not relying on any positional information.
2,564
1905.11141
title_snapshot
[ -0.044358253479003906, -0.024844789877533913, -0.020933926105499268, 0.05139230936765671, 0.02537216804921627, 0.04447779804468155, 0.01675979234278202, -0.014128193259239197, -0.02159108594059944, -0.06334930658340454, -0.015160259790718555, -0.0176684707403183, -0.08439747244119644, 0.02...
Understanding Generalization in Recurrent Neural Networks
https://openreview.net/forum?id=rkgg6xBYDH
[ "Zhuozhuo Tu", "Fengxiang He", "Dacheng Tao" ]
Poster
null
In this work, we develop the theory for analyzing the generalization performance of recurrent neural networks. We first present a new generalization bound for recurrent neural networks based on matrix 1-norm and Fisher-Rao norm. The definition of Fisher-Rao norm relies on a structural lemma about the gradient of RNNs. ...
[ "generalization", "recurrent neural networks", "learning theory" ]
We prove generalization bounds for recurrent neural networks based on matrix 1-norm and Fisher-Rao norm.
2,560
null
null
[ -0.01547178439795971, -0.02201172523200512, 0.011794284917414188, 0.038050130009651184, 0.04002074897289276, 0.045420318841934204, 0.051664359867572784, -0.005862547550350428, -0.050866466015577316, -0.022960657253861427, -0.002510795136913657, 0.013016716577112675, -0.07744668424129486, -...
Conservative Uncertainty Estimation By Fitting Prior Networks
https://openreview.net/forum?id=BJlahxHYDS
[ "Kamil Ciosek", "Vincent Fortuin", "Ryota Tomioka", "Katja Hofmann", "Richard Turner" ]
Poster
null
Obtaining high-quality uncertainty estimates is essential for many applications of deep neural networks. In this paper, we theoretically justify a scheme for estimating uncertainties, based on sampling from a prior distribution. Crucially, the uncertainty estimates are shown to be conservative in the sense that they ne...
[ "uncertainty quantification", "deep learning", "Gaussian process", "epistemic uncertainty", "random network", "prior", "Bayesian inference" ]
We provide theoretical support to uncertainty estimates for deep learning obtained fitting random priors.
2,553
null
null
[ 0.0052934680134058, 0.009241906926035881, -0.03868339955806732, 0.05589950084686279, 0.04794111102819443, 0.047000423073768616, 0.02478157915174961, 0.009016363881528378, -0.02433820255100727, -0.07363453507423401, 0.0022525780368596315, -0.02671818993985653, -0.053733695298433304, -0.0090...
NAS-Bench-1Shot1: Benchmarking and Dissecting One-shot Neural Architecture Search
https://openreview.net/forum?id=SJx9ngStPH
[ "Arber Zela", "Julien Siems", "Frank Hutter" ]
Poster
null
One-shot neural architecture search (NAS) has played a crucial role in making NAS methods computationally feasible in practice. Nevertheless, there is still a lack of understanding on how these weight-sharing algorithms exactly work due to the many factors controlling the dynamics of the process. In order to allow a sc...
[ "Neural Architecture Search", "Deep Learning", "Computer Vision" ]
null
2,547
2001.10422
title_snapshot
[ -0.00954395905137062, -0.016499942168593407, -0.040100790560245514, 0.027700038626790047, 0.039450954645872116, 0.042494919151067734, 0.01470260415226221, -0.015151072293519974, -0.012012539431452751, -0.046322256326675415, 0.025455351918935776, -0.030590031296014786, -0.04912535101175308, ...
Learning to Coordinate Manipulation Skills via Skill Behavior Diversification
https://openreview.net/forum?id=ryxB2lBtvH
[ "Youngwoon Lee", "Jingyun Yang", "Joseph J. Lim" ]
Poster
null
When mastering a complex manipulation task, humans often decompose the task into sub-skills of their body parts, practice the sub-skills independently, and then execute the sub-skills together. Similarly, a robot with multiple end-effectors can perform complex tasks by coordinating sub-skills of each end-effector. To r...
[ "reinforcement learning", "hierarchical reinforcement learning", "modular framework", "skill coordination", "bimanual manipulation" ]
We propose to tackle complex tasks of multiple agents by learning composable primitive skills and coordination of the skills.
2,535
null
null
[ 0.011699582450091839, 0.0010058090556412935, -0.029935747385025024, 0.025632554665207863, 0.0592973567545414, 0.0245909933000803, 0.021520070731639862, 0.002501216484233737, -0.048191968351602554, -0.05446195229887962, -0.012035000137984753, -0.014629213139414787, -0.06624999642372131, -0....
Robust Subspace Recovery Layer for Unsupervised Anomaly Detection
https://openreview.net/forum?id=rylb3eBtwr
[ "Chieh-Hsin Lai", "Dongmian Zou", "Gilad Lerman" ]
Poster
null
We propose a neural network for unsupervised anomaly detection with a novel robust subspace recovery layer (RSR layer). This layer seeks to extract the underlying subspace from a latent representation of the given data and removes outliers that lie away from this subspace. It is used within an autoencoder. The encoder ...
[ "robust subspace recovery", "unsupervised anomaly detection", "outliers", "latent space", "autoencoder" ]
This work proposes an autoencoder with a novel robust subspace recovery layer for unsupervised anomaly detection and demonstrates state-of-the-art results on various datasets.
2,527
1904.00152
title_snapshot
[ -0.0010831691324710846, -0.03482659533619881, -0.010102762840688229, 0.05448133125901222, 0.05023731291294098, -0.013954189606010914, 0.02680939994752407, -0.016963714733719826, -0.02950773388147354, -0.03416449949145317, -0.03830774128437042, -0.012239369563758373, -0.05923187732696533, 0...
Learning Nearly Decomposable Value Functions Via Communication Minimization
https://openreview.net/forum?id=HJx-3grYDB
[ "Tonghan Wang*", "Jianhao Wang*", "Chongyi Zheng", "Chongjie Zhang" ]
Poster
null
Reinforcement learning encounters major challenges in multi-agent settings, such as scalability and non-stationarity. Recently, value function factorization learning emerges as a promising way to address these challenges in collaborative multi-agent systems. However, existing methods have been focusing on learning full...
[ "Multi-agent reinforcement learning", "Nearly decomposable value function", "Minimized communication", "Multi-agent systems" ]
null
2,526
1910.05366
title_snapshot
[ -0.018206680193543434, -0.042511582374572754, -0.008863323368132114, 0.045389022678136826, 0.046605780720710754, 0.0393090583384037, -0.0011562014697119594, -0.018617825582623482, -0.011908876709640026, -0.06455183774232864, -0.009214166551828384, 0.029069535434246063, -0.08842940628528595, ...
Extreme Classification via Adversarial Softmax Approximation
https://openreview.net/forum?id=rJxe3xSYDS
[ "Robert Bamler", "Stephan Mandt" ]
Poster
null
Training a classifier over a large number of classes, known as 'extreme classification', has become a topic of major interest with applications in technology, science, and e-commerce. Traditional softmax regression induces a gradient cost proportional to the number of classes C, which often is prohibitively expensive. ...
[ "Extreme classification", "negative sampling" ]
An efficient, unbiased approximation of the softmax loss function for extreme classification
2,523
2002.06298
title_snapshot
[ -0.01763089746236801, -0.03366141766309738, 0.004991377238184214, 0.043909430503845215, 0.031504396349191666, -0.006637926213443279, 0.0014667230425402522, -0.0354098342359066, -0.017037689685821533, -0.045691583305597305, -0.032531771808862686, 0.001727009192109108, -0.08102785795927048, ...
Information Geometry of Orthogonal Initializations and Training
https://openreview.net/forum?id=rkg1ngrFPr
[ "Piotr Aleksander Sokół", "Il Memming Park" ]
Poster
null
Recently mean field theory has been successfully used to analyze properties of wide, random neural networks. It gave rise to a prescriptive theory for initializing feed-forward neural networks with orthogonal weights, which ensures that both the forward propagated activations and the backpropagated ...
[ "Fisher", "mean-field", "deep learning" ]
nearly isometric DNN initializations imply low parameter space curvature, and a lower condition number, but that's not always great
2,521
1810.03785
title_snapshot
[ -0.032703012228012085, -0.020640041679143906, 0.014749832451343536, 0.029251886531710625, 0.029090367257595062, 0.03427370265126228, 0.01609538309276104, 0.009125054813921452, -0.020776156336069107, -0.04166307672858238, -0.00007704854942858219, 0.0019474453292787075, -0.05048614740371704, ...
Mixed Precision DNNs: All you need is a good parametrization
https://openreview.net/forum?id=Hyx0slrFvH
[ "Stefan Uhlich", "Lukas Mauch", "Fabien Cardinaux", "Kazuki Yoshiyama", "Javier Alonso Garcia", "Stephen Tiedemann", "Thomas Kemp", "Akira Nakamura" ]
Poster
null
Efficient deep neural network (DNN) inference on mobile or embedded devices typically involves quantization of the network parameters and activations. In particular, mixed precision networks achieve better performance than networks with homogeneous bitwidth for the same size constraint. Since choosing the optimal bitwi...
[ "Deep Neural Network Compression", "Quantization", "Straight through gradients" ]
null
2,519
1905.11452
title_snapshot
[ 0.0008996489341370761, -0.0012780167162418365, -0.022668227553367615, 0.028337029740214348, 0.04426707699894905, 0.07094559818506241, 0.012706327252089977, -0.025584334507584572, -0.02806886099278927, -0.04130600392818451, 0.01810666173696518, -0.034024711698293686, -0.04414718225598335, 0...
PROGRESSIVE LEARNING AND DISENTANGLEMENT OF HIERARCHICAL REPRESENTATIONS
https://openreview.net/forum?id=SJxpsxrYPS
[ "Zhiyuan Li", "Jaideep Vitthal Murkute", "Prashnna Kumar Gyawali", "Linwei Wang" ]
Spotlight
null
Learning rich representation from data is an important task for deep generative models such as variational auto-encoder (VAE). However, by extracting high-level abstractions in the bottom-up inference process, the goal of preserving all factors of variations for top-down generation is compromised. Motivated by the conc...
[ "generative model", "disentanglement", "progressive learning", "VAE" ]
We proposed a progressive learning method to improve learning and disentangling latent representations at different levels of abstraction.
2,518
2002.10549
title_snapshot
[ -0.012383656576275826, 0.0039309123530983925, 0.009723004885017872, 0.048274774104356766, 0.03371530771255493, -0.002287884708493948, 0.06111143156886101, -0.010086232796311378, -0.029098445549607277, -0.030303599312901497, -0.03162353113293648, -0.01928538829088211, -0.06811469048261642, ...
Co-Attentive Equivariant Neural Networks: Focusing Equivariance On Transformations Co-Occurring in Data
https://openreview.net/forum?id=r1g6ogrtDr
[ "David W. Romero", "Mark Hoogendoorn" ]
Poster
null
Equivariance is a nice property to have as it produces much more parameter efficient neural architectures and preserves the structure of the input through the feature mapping. Even though some combinations of transformations might never appear (e.g. an upright face with a horizontal nose), current equivariant architect...
[ "Equivariant Neural Networks", "Attention Mechanisms", "Deep Learning" ]
We utilize attention to restrict equivariant neural networks to the set or co-occurring transformations in data.
2,517
1911.07849
title_snapshot
[ 0.012440305203199387, 0.012910176999866962, 0.02509138360619545, 0.018299778923392296, 0.01132969744503498, 0.04025031626224518, 0.01614641211926937, 0.03608250245451927, -0.036578062921762466, -0.03480779752135277, -0.03528405353426933, -0.010132026858627796, -0.06345845013856888, 0.01052...
Deep Orientation Uncertainty Learning based on a Bingham Loss
https://openreview.net/forum?id=ryloogSKDS
[ "Igor Gilitschenski", "Roshni Sahoo", "Wilko Schwarting", "Alexander Amini", "Sertac Karaman", "Daniela Rus" ]
Poster
null
Reasoning about uncertain orientations is one of the core problems in many perception tasks such as object pose estimation or motion estimation. In these scenarios, poor illumination conditions, sensor limitations, or appearance invariance may result in highly uncertain estimates. In this work, we propose a novel learn...
[ "Orientation Estimation", "Directional Statistics", "Bingham Distribution" ]
A method for learning to predict uncertainties over orientations using the Bingham Distribution
2,513
null
null
[ -0.00003709949305630289, 0.0053016068413853645, -0.006436389405280352, 0.03374303877353668, -0.0015266143018379807, 0.047545675188302994, 0.016999369487166405, 0.012594178318977356, -0.03795301169157028, -0.04586542397737503, -0.05196450278162956, -0.005787396337836981, -0.07231278717517853,...
Reconstructing continuous distributions of 3D protein structure from cryo-EM images
https://openreview.net/forum?id=SJxUjlBtwB
[ "Ellen D. Zhong", "Tristan Bepler", "Joseph H. Davis", "Bonnie Berger" ]
Spotlight
null
Cryo-electron microscopy (cryo-EM) is a powerful technique for determining the structure of proteins and other macromolecular complexes at near-atomic resolution. In single particle cryo-EM, the central problem is to reconstruct the 3D structure of a macromolecule from $10^{4-7}$ noisy and randomly oriented 2D projecti...
[ "generative models", "proteins", "3D reconstruction", "cryo-EM" ]
We propose a deep generative model of volumes for 3D cryo-EM reconstruction from unlabelled 2D images and show that it can learn can learn continuous deformations in protein structure.
2,509
1909.05215
title_snapshot
[ -0.0052947658114135265, 0.02176288701593876, -0.0345643050968647, 0.04002255201339722, 0.028801491484045982, -0.004961878526955843, 0.011310216039419174, 0.006767489481717348, -0.020604578778147697, -0.03483695536851883, -0.0033443698193877935, -0.016390426084399223, -0.07148386538028717, ...
Generalization of Two-layer Neural Networks: An Asymptotic Viewpoint
https://openreview.net/forum?id=H1gBsgBYwH
[ "Jimmy Ba", "Murat Erdogdu", "Taiji Suzuki", "Denny Wu", "Tianzong Zhang" ]
Spotlight
null
This paper investigates the generalization properties of two-layer neural networks in high-dimensions, i.e. when the number of samples $n$, features $d$, and neurons $h$ tend to infinity at the same rate. Specifically, we derive the exact population risk of the unregularized least squares regression problem with two-la...
[ "Neural Networks", "Generalization", "High-dimensional Statistics" ]
Derived population risk of two-layer neural networks in high dimensions and examined presence / absence of "double descent".
2,506
null
null
[ -0.04717554897069931, -0.017629316076636314, 0.011399690993130207, 0.014513789676129818, 0.039451662451028824, 0.0457591637969017, 0.010492337867617607, -0.013606278225779533, -0.028114859014749527, -0.02201603166759014, 0.0027463152073323727, -0.01134936697781086, -0.0534304603934288, 0.0...
Critical initialisation in continuous approximations of binary neural networks
https://openreview.net/forum?id=rylmoxrFDH
[ "George Stamatescu", "Federica Gerace", "Carlo Lucibello", "Ian Fuss", "Langford White" ]
Poster
null
The training of stochastic neural network models with binary ($\pm1$) weights and activations via continuous surrogate networks is investigated. We derive new surrogates using a novel derivation based on writing the stochastic neural network as a Markov chain. This derivation also encompasses existing variants of the s...
[]
signal propagation theory applied to continuous surrogates of binary nets; counter intuitive initialisation; reparameterisation trick not helpful
2,503
1902.00177
title_snapshot
[ -0.030970251187682152, -0.008708013221621513, 0.006258169189095497, 0.0333070233464241, 0.022527486085891724, 0.05643003061413765, 0.013530842959880829, 0.011930149979889393, -0.0354481004178524, -0.02657295949757099, 0.0018474376993253827, 0.006432635709643364, -0.04485275596380234, -0.01...
Sub-policy Adaptation for Hierarchical Reinforcement Learning
https://openreview.net/forum?id=ByeWogStDS
[ "Alexander Li", "Carlos Florensa", "Ignasi Clavera", "Pieter Abbeel" ]
Poster
null
Hierarchical reinforcement learning is a promising approach to tackle long-horizon decision-making problems with sparse rewards. Unfortunately, most methods still decouple the lower-level skill acquisition process and the training of a higher level that controls the skills in a new task. Leaving the skills fixed can le...
[ "Hierarchical Reinforcement Learning", "Transfer", "Skill Discovery" ]
We propose HiPPO, a stable Hierarchical Reinforcement Learning algorithm that can train several levels of the hierarchy simultaneously, giving good performance both in skill discovery and adaptation.
2,497
1906.05862
title_snapshot
[ -0.035516075789928436, -0.022098911926150322, 0.03358466923236847, 0.04375007376074791, 0.04917542263865471, 0.017076995223760605, 0.013139444403350353, -0.02733708545565605, -0.02427707053720951, -0.048324812203645706, -0.005709980148822069, 0.018642138689756393, -0.06545763462781906, -0....
Episodic Reinforcement Learning with Associative Memory
https://openreview.net/forum?id=HkxjqxBYDB
[ "Guangxiang Zhu*", "Zichuan Lin*", "Guangwen Yang", "Chongjie Zhang" ]
Poster
null
Sample efficiency has been one of the major challenges for deep reinforcement learning. Non-parametric episodic control has been proposed to speed up parametric reinforcement learning by rapidly latching on previously successful policies. However, previous work on episodic reinforcement learning neglects the relationsh...
[ "Deep Reinforcement Learning", "Episodic Control", "Episodic Memory", "Associative Memory", "Non-Parametric Method", "Sample Efficiency" ]
null
2,485
null
null
[ -0.041900381445884705, 0.016990356147289276, 0.009977862238883972, 0.03821650519967079, 0.019470402970910072, 0.030289301648736, 0.011490884236991405, 0.024186886847019196, -0.037615131586790085, -0.03563854843378067, -0.014933823607861996, 0.009331309236586094, -0.043740492314100266, -0.0...
Depth-Width Trade-offs for ReLU Networks via Sharkovsky's Theorem
https://openreview.net/forum?id=BJe55gBtvH
[ "Vaggos Chatziafratis", "Sai Ganesh Nagarajan", "Ioannis Panageas", "Xiao Wang" ]
Spotlight
null
Understanding the representational power of Deep Neural Networks (DNNs) and how their structural properties (e.g., depth, width, type of activation unit) affect the functions they can compute, has been an important yet challenging question in deep learning and approximation theory. In a seminal paper, Telgarsky high- l...
[ "Depth-Width trade-offs", "ReLU networks", "chaos theory", "Sharkovsky Theorem", "dynamical systems" ]
In this work, we point to a new connection between DNNs expressivity and Sharkovsky’s Theorem from dynamical systems, that enables us to characterize the depth-width trade-offs of ReLU networks
2,482
1912.04378
title_snapshot
[ -0.028281787410378456, 0.002036206889897585, 0.016836510971188545, 0.020765621215105057, 0.04127824679017067, 0.03821646049618721, 0.046742212027311325, -0.00568894250318408, -0.03695139288902283, -0.04669917747378349, -0.00031482227495871484, -0.005153107922524214, -0.05685187876224518, 0...
Energy-based models for atomic-resolution protein conformations
https://openreview.net/forum?id=S1e_9xrFvS
[ "Yilun Du", "Joshua Meier", "Jerry Ma", "Rob Fergus", "Alexander Rives" ]
Spotlight
null
We propose an energy-based model (EBM) of protein conformations that operates at atomic scale. The model is trained solely on crystallized protein data. By contrast, existing approaches for scoring conformations use energy functions that incorporate knowledge of physical principles and features that are the complex pro...
[ "energy-based model", "transformer", "energy function", "protein conformation" ]
Energy-based models trained on crystallized protein structures predict native side chain configurations and automatically discover molecular energy features.
2,477
2004.13167
title_snapshot
[ -0.021617885679006577, 0.009439326822757721, -0.024643193930387497, 0.021303776651620865, 0.0531868040561676, -0.02209070324897766, 0.016565805301070213, 0.01229593064635992, 0.00268104812130332, -0.024853425100445747, 0.03392970561981201, 0.01875574328005314, -0.10244761407375336, 0.01373...
Federated Learning with Matched Averaging
https://openreview.net/forum?id=BkluqlSFDS
[ "Hongyi Wang", "Mikhail Yurochkin", "Yuekai Sun", "Dimitris Papailiopoulos", "Yasaman Khazaeni" ]
Talk
null
Federated learning allows edge devices to collaboratively learn a shared model while keeping the training data on device, decoupling the ability to do model training from the need to store the data in the cloud. We propose Federated matched averaging (FedMA) algorithm designed for federated learning of modern neural ne...
[ "federated learning" ]
Communication efficient federated learning with layer-wise matching
2,476
2002.06440
title_snapshot
[ 0.01881209947168827, -0.06244654953479767, -0.0003850204520858824, 0.04746605455875397, 0.02545606717467308, 0.02079133875668049, 0.03236455097794533, 0.012061639688909054, -0.022837167605757713, -0.025772323831915855, 0.022771280258893967, -0.037779271602630615, -0.06312770396471024, 0.01...
Pitfalls of In-Domain Uncertainty Estimation and Ensembling in Deep Learning
https://openreview.net/forum?id=BJxI5gHKDr
[ "Arsenii Ashukha", "Alexander Lyzhov", "Dmitry Molchanov", "Dmitry Vetrov" ]
Poster
null
Uncertainty estimation and ensembling methods go hand-in-hand. Uncertainty estimation is one of the main benchmarks for assessment of ensembling performance. At the same time, deep learning ensembles have provided state-of-the-art results in uncertainty estimation. In this work, we focus on in-domain uncertainty for im...
[ "uncertainty", "in-domain uncertainty", "deep ensembles", "ensemble learning", "deep learning" ]
We highlight the problems with common metrics of in-domain uncertainty and perform a broad study of modern ensembling techniques.
2,473
2002.06470
title_snapshot
[ 0.005996251478791237, -0.002217493951320648, -0.049056828022003174, 0.046337906271219254, 0.049784157425165176, 0.014830386266112328, 0.030329225584864616, -0.0396784208714962, -0.012027042917907238, -0.05470754951238632, -0.04006904736161232, 0.01589374989271164, -0.06730615347623825, -0....
DiffTaichi: Differentiable Programming for Physical Simulation
https://openreview.net/forum?id=B1eB5xSFvr
[ "Yuanming Hu", "Luke Anderson", "Tzu-Mao Li", "Qi Sun", "Nathan Carr", "Jonathan Ragan-Kelley", "Fredo Durand" ]
Poster
null
We present DiffTaichi, a new differentiable programming language tailored for building high-performance differentiable physical simulators. Based on an imperative programming language, DiffTaichi generates gradients of simulation steps using source code transformations that preserve arithmetic intensity and parallelism...
[ "Differentiable programming", "robotics", "optimal control", "physical simulation", "machine learning system" ]
We study the problem of learning and optimizing through physical simulations via differentiable programming, using our proposed DiffSim programming language and compiler.
2,469
1910.00935
title_snapshot
[ -0.03269863501191139, -0.01484800223261118, -0.010338298045098782, 0.02756395936012268, 0.05368778482079506, 0.023830760270357132, 0.0032068490982055664, 0.013951670378446579, -0.0324980728328228, -0.03646470978856087, 0.011332660913467407, -0.026735305786132812, -0.07754488289356232, 0.03...
A Mutual Information Maximization Perspective of Language Representation Learning
https://openreview.net/forum?id=Syx79eBKwr
[ "Lingpeng Kong", "Cyprien de Masson d'Autume", "Lei Yu", "Wang Ling", "Zihang Dai", "Dani Yogatama" ]
Spotlight
null
We show state-of-the-art word representation learning methods maximize an objective function that is a lower bound on the mutual information between different parts of a word sequence (i.e., a sentence). Our formulation provides an alternative perspective that unifies classical word embedding models (e.g., Skip-gram) a...
[]
null
2,466
1910.08350
title_snapshot
[ -0.025835053995251656, 0.0004671154310926795, -0.009754755534231663, 0.04874251037836075, 0.02554253116250038, 0.044032108038663864, 0.036267202347517014, 0.010769744403660297, -0.012273060157895088, -0.011235135607421398, -0.021951600909233093, 0.02919294498860836, -0.05687737464904785, -...
Domain Adaptive Multibranch Networks
https://openreview.net/forum?id=rJxycxHKDS
[ "Róger Bermúdez-Chacón", "Mathieu Salzmann", "Pascal Fua" ]
Poster
null
We tackle unsupervised domain adaptation by accounting for the fact that different domains may need to be processed differently to arrive to a common feature representation effective for recognition. To this end, we introduce a deep learning framework where each domain undergoes a different sequence of operations, allo...
[ "Domain Adaptation", "Computer Vision" ]
A Multiflow Network is a dynamic architecture for domain adaptation that learns potentially different computational graphs per domain, so as to map them to a common representation where inference can be performed in a domain-agnostic fashion.
2,456
null
null
[ 0.0037385865580290556, -0.03308303281664848, 0.011793827638030052, 0.020643962547183037, 0.05599643662571907, 0.02878125198185444, 0.010142363607883453, 0.010567977093160152, -0.015267659910023212, -0.03176146373152733, -0.02188514545559883, 0.020771384239196777, -0.05662093684077263, 0.00...
Unbiased Contrastive Divergence Algorithm for Training Energy-Based Latent Variable Models
https://openreview.net/forum?id=r1eyceSYPr
[ "Yixuan Qiu", "Lingsong Zhang", "Xiao Wang" ]
Spotlight
null
The contrastive divergence algorithm is a popular approach to training energy-based latent variable models, which has been widely used in many machine learning models such as the restricted Boltzmann machines and deep belief nets. Despite its empirical success, the contrastive divergence algorithm is also known to have...
[ "energy model", "restricted Boltzmann machine", "contrastive divergence", "unbiased Markov chain Monte Carlo", "distribution coupling" ]
We have developed a new training algorithm for energy-based latent variable models that completely removes the bias of contrastive divergence.
2,455
null
null
[ -0.005467674694955349, -0.03193674609065056, -0.009990605525672436, 0.03448081761598587, 0.03763573616743088, -0.012211481109261513, 0.03350655734539032, 0.01170183066278696, -0.03259382024407387, -0.021135777235031128, 0.025793468579649925, 0.02336178719997406, -0.07053062319755554, 0.006...
Differentiable Reasoning over a Virtual Knowledge Base
https://openreview.net/forum?id=SJxstlHFPH
[ "Bhuwan Dhingra", "Manzil Zaheer", "Vidhisha Balachandran", "Graham Neubig", "Ruslan Salakhutdinov", "William W. Cohen" ]
Talk
null
We consider the task of answering complex multi-hop questions using a corpus as a virtual knowledge base (KB). In particular, we describe a neural module, DrKIT, that traverses textual data like a KB, softly following paths of relations between mentions of entities in the corpus. At each step the module uses a combinat...
[ "Question Answering", "Multi-Hop QA", "Deep Learning", "Knowledge Bases", "Information Extraction", "Data Structures for QA" ]
Differentiable multi-hop access to a textual knowledge base of indexed contextual representations
2,447
2002.10640
title_snapshot
[ 0.006869269069284201, -0.012582600116729736, 0.01520349271595478, 0.08769062906503677, 0.047475479543209076, 0.011167464777827263, 0.017269650474190712, -0.021289028227329254, -0.027080604806542397, 0.005951759405434132, -0.0330631323158741, 0.0376429483294487, -0.04649342969059944, -0.001...
Making Sense of Reinforcement Learning and Probabilistic Inference
https://openreview.net/forum?id=S1xitgHtvS
[ "Brendan O'Donoghue", "Ian Osband", "Catalin Ionescu" ]
Spotlight
null
Reinforcement learning (RL) combines a control problem with statistical estimation: The system dynamics are not known to the agent, but can be learned through experience. A recent line of research casts ‘RL as inference’ and suggests a particular framework to generalize the RL problem as probabilistic inference. Our pa...
[ "Reinforcement learning", "Bayesian inference", "Exploration" ]
Popular algorithms that cast "RL as Inference" ignore the role of uncertainty and exploration. We highlight the importance of these issues and present a coherent framework for RL and inference that handles them gracefully.
2,446
2001.00805
title_snapshot
[ -0.027635483071208, -0.01873939484357834, -0.022083718329668045, 0.04053926095366478, 0.05490310490131378, 0.011819543316960335, 0.02175106853246689, -0.003205150132998824, -0.03969612717628479, -0.011499281041324139, -0.01902202144265175, 0.037555668503046036, -0.07095324993133545, -0.038...
Prediction Poisoning: Towards Defenses Against DNN Model Stealing Attacks
https://openreview.net/forum?id=SyevYxHtDB
[ "Tribhuvanesh Orekondy", "Bernt Schiele", "Mario Fritz" ]
Poster
null
High-performance Deep Neural Networks (DNNs) are increasingly deployed in many real-world applications e.g., cloud prediction APIs. Recent advances in model functionality stealing attacks via black-box access (i.e., inputs in, predictions out) threaten the business model of such applications, which require a lot of tim...
[ "model functionality stealing", "adversarial machine learning" ]
We propose the first approach that can resist DNN model stealing/extraction attacks
2,438
1906.10908
title_snapshot
[ -0.00994215440005064, -0.012400061823427677, -0.02571609988808632, 0.06032774969935417, 0.0478362999856472, 0.00904540903866291, 0.024797771126031876, -0.019490709528326988, -0.007335429545491934, 0.0009896599221974611, 0.00972446333616972, -0.007059091702103615, -0.0495399609208107, 0.005...
SCALOR: Generative World Models with Scalable Object Representations
https://openreview.net/forum?id=SJxrKgStDH
[ "Jindong Jiang*", "Sepehr Janghorbani*", "Gerard De Melo", "Sungjin Ahn" ]
Poster
null
Scalability in terms of object density in a scene is a primary challenge in unsupervised sequential object-oriented representation learning. Most of the previous models have been shown to work only on scenes with a few objects. In this paper, we propose SCALOR, a probabilistic generative world model for learning SCALab...
[]
null
2,433
1910.02384
title_snapshot
[ 0.0057129282504320145, -0.018290305510163307, 0.0012656747130677104, 0.04454788565635681, 0.0010520886862650514, 0.015985636040568352, 0.0029753423295915127, 0.03990556299686432, -0.03689933940768242, -0.018727019429206848, -0.029145603999495506, -0.0023372352588921785, -0.06505957990884781,...
Neural tangent kernels, transportation mappings, and universal approximation
https://openreview.net/forum?id=HklQYxBKwS
[ "Ziwei Ji", "Matus Telgarsky", "Ruicheng Xian" ]
Poster
null
This paper establishes rates of universal approximation for the shallow neural tangent kernel (NTK): network weights are only allowed microscopic changes from random initialization, which entails that activations are mostly unchanged, and the network is nearly equivalent to its linearization. Concretely, the paper has ...
[ "Neural Tangent Kernel", "universal approximation", "Barron", "transport mapping" ]
The NTK linearization is a universal approximator, even when looking arbitrarily close to initialization
2,429
1910.06956
title_snapshot
[ -0.05498168244957924, -0.024566585198044777, 0.02331809513270855, 0.04926765710115433, 0.017552634701132774, 0.05531885474920273, -0.0004444836522452533, 0.01862158440053463, -0.03238007798790932, -0.04496565833687782, -0.008122721686959267, 0.007964875549077988, -0.051744911819696426, -0....
Learning to Move with Affordance Maps
https://openreview.net/forum?id=BJgMFxrYPB
[ "William Qi", "Ravi Teja Mullapudi", "Saurabh Gupta", "Deva Ramanan" ]
Poster
null
The ability to autonomously explore and navigate a physical space is a fundamental requirement for virtually any mobile autonomous agent, from household robotic vacuums to autonomous vehicles. Traditional SLAM-based approaches for exploration and navigation largely focus on leveraging scene geometry, but fail to model ...
[ "navigation", "exploration" ]
We address the task of autonomous exploration and navigation using spatial affordance maps that can be learned in a self-supervised manner, these outperform classic geometric baselines while being more sample efficient than contemporary RL algorithms
2,427
2001.02364
title_snapshot
[ 0.001712146564386785, -0.0026030566077679396, 0.02120278775691986, 0.0316346250474453, 0.045218661427497864, 0.006719236262142658, 0.030525390058755875, 0.011625689454376698, -0.03697647154331207, -0.038617003709077835, -0.0414150133728981, -0.008094696328043938, -0.06096963584423065, -0.0...
Deep Learning For Symbolic Mathematics
https://openreview.net/forum?id=S1eZYeHFDS
[ "Guillaume Lample", "François Charton" ]
Spotlight
null
Neural networks have a reputation for being better at solving statistical or approximate problems than at performing calculations or working with symbolic data. In this paper, we show that they can be surprisingly good at more elaborated tasks in mathematics, such as symbolic integration and solving differential equati...
[ "symbolic", "math", "deep learning", "transformers" ]
We train a neural network to compute function integrals, and to solve complex differential equations.
2,425
1912.01412
title_snapshot
[ -0.03599587082862854, -0.017385834828019142, -0.03136635571718216, 0.04971213638782501, 0.04999842122197151, 0.042897049337625504, 0.015930069610476494, 0.00024042936274781823, -0.008886735886335373, -0.03419942408800125, 0.004995300434529781, 0.01533572282642126, -0.037525519728660583, 0....
Differentiable learning of numerical rules in knowledge graphs
https://openreview.net/forum?id=rJleKgrKwS
[ "Po-Wei Wang", "Daria Stepanova", "Csaba Domokos", "J. Zico Kolter" ]
Poster
null
Rules over a knowledge graph (KG) capture interpretable patterns in data and can be used for KG cleaning and completion. Inspired by the TensorLog differentiable logic framework, which compiles rule inference into a sequence of differentiable operations, recently a method called Neural LP has been proposed for learning...
[ "knowledge graphs", "rule learning", "differentiable neural logic" ]
We present an efficient approach to integrating numerical comparisons into differentiable rule learning in knowledge graphs
2,423
null
null
[ -0.02435135655105114, -0.010365741327404976, 0.009809802286326885, 0.012419824488461018, 0.06711280345916748, 0.014109925366938114, 0.013873424381017685, -0.031821198761463165, -0.052354611456394196, -0.017329376190900803, -0.0020922713447362185, 0.03396615758538246, -0.06478672474622726, ...
Consistency Regularization for Generative Adversarial Networks
https://openreview.net/forum?id=S1lxKlSKPH
[ "Han Zhang", "Zizhao Zhang", "Augustus Odena", "Honglak Lee" ]
Poster
null
Generative Adversarial Networks (GANs) are known to be difficult to train, despite considerable research effort. Several regularization techniques for stabilizing training have been proposed, but they introduce non-trivial computational overheads and interact poorly with existing techniques like spectral normalization....
[ "Generative Adversarial Networks", "Consistency Regularization", "GAN" ]
null
2,422
1910.12027
title_snapshot
[ 0.011856704950332642, -0.05221400409936905, -0.03644661605358124, 0.055393096059560776, 0.0324927493929863, 0.02211281657218933, 0.012072510085999966, -0.01652451977133751, -0.028806015849113464, -0.06946908682584763, -0.016281956806778908, 0.005112634040415287, -0.08225269615650177, 0.014...
On Generalization Error Bounds of Noisy Gradient Methods for Non-Convex Learning
https://openreview.net/forum?id=SkxxtgHKPS
[ "Jian Li", "Xuanyuan Luo", "Mingda Qiao" ]
Poster
null
Generalization error (also known as the out-of-sample error) measures how well the hypothesis learned from training data generalizes to previously unseen data. Proving tight generalization error bounds is a central question in statistical learning theory. In this paper, we obtain generalization error bounds ...
[ "learning theory", "generalization", "nonconvex learning", "stochastic gradient descent", "Langevin dynamics" ]
We give some generalization error bounds of noisy gradient methods such as SGLD, Langevin dynamics, noisy momentum and so forth.
2,421
1902.00621
title_snapshot
[ -0.011719826608896255, 0.0065669110044837, -0.010550648905336857, 0.024321887642145157, 0.033554598689079285, 0.04033695161342621, 0.05487503111362457, -0.02305172011256218, -0.032490503042936325, -0.041249264031648636, -0.013768344186246395, 0.009457016363739967, -0.09247995913028717, -0....
SUMO: Unbiased Estimation of Log Marginal Probability for Latent Variable Models
https://openreview.net/forum?id=SylkYeHtwr
[ "Yucen Luo", "Alex Beatson", "Mohammad Norouzi", "Jun Zhu", "David Duvenaud", "Ryan P. Adams", "Ricky T. Q. Chen" ]
Spotlight
null
Standard variational lower bounds used to train latent variable models produce biased estimates of most quantities of interest. We introduce an unbiased estimator of the log marginal likelihood and its gradients for latent variable models based on randomized truncation of infinite series. If parameterized by an encoder...
[]
We create an unbiased estimator for the log probability of latent variable models, extending such models to a larger scope of applications.
2,419
2004.00353
title_snapshot
[ 0.02169424667954445, -0.02568458393216133, -0.03543538227677345, 0.040489744395017624, 0.027938442304730415, 0.016880685463547707, 0.04235800355672836, 0.009594976902008057, -0.039860740303993225, -0.014112768694758415, 0.005775701254606247, 0.014741498976945877, -0.055344924330711365, -0....
Scale-Equivariant Steerable Networks
https://openreview.net/forum?id=HJgpugrKPS
[ "Ivan Sosnovik", "Michał Szmaja", "Arnold Smeulders" ]
Poster
null
The effectiveness of Convolutional Neural Networks (CNNs) has been substantially attributed to their built-in property of translation equivariance. However, CNNs do not have embedded mechanisms to handle other types of transformations. In this work, we pay attention to scale changes, which regularly appear in various t...
[ "Scale Equivariance", "Steerable Filters" ]
null
2,415
1910.11093
title_snapshot
[ 0.025846650823950768, -0.01521432213485241, 0.041828855872154236, -0.0013402113690972328, 0.02156033366918564, 0.03094656392931938, 0.01670300029218197, 0.03952118381857872, -0.04063102975487709, -0.05208038166165352, -0.005235427524894476, -0.006025671027600765, -0.05158543586730957, 0.01...
DeepSphere: a graph-based spherical CNN
https://openreview.net/forum?id=B1e3OlStPB
[ "Michaël Defferrard", "Martino Milani", "Frédérick Gusset", "Nathanaël Perraudin" ]
Spotlight
null
Designing a convolution for a spherical neural network requires a delicate tradeoff between efficiency and rotation equivariance. DeepSphere, a method based on a graph representation of the discretized sphere, strikes a controllable balance between these two desiderata. This contribution is twofold. First, we study bot...
[ "spherical cnns", "graph neural networks", "geometric deep learning" ]
A graph-based spherical CNN that strikes an interesting balance of trade-offs for a wide variety of applications.
2,413
2012.15000
title_snapshot
[ 0.03704635053873062, -0.0043638646602630615, 0.014425777830183506, 0.017412371933460236, 0.009086446836590767, 0.03806256875395775, 0.010971336625516415, 0.015323670580983162, -0.029904697090387344, -0.06992395967245102, -0.005944503005594015, -0.029734183102846146, -0.08701735734939575, 0...
Classification-Based Anomaly Detection for General Data
https://openreview.net/forum?id=H1lK_lBtvS
[ "Liron Bergman", "Yedid Hoshen" ]
Poster
null
Anomaly detection, finding patterns that substantially deviate from those seen previously, is one of the fundamental problems of artificial intelligence. Recently, classification-based methods were shown to achieve superior results on this task. In this work, we present a unifying view and propose an open-set method, G...
[ "anomaly detection" ]
Anomaly detection method that uses: openset techniques for better generalization, random-transformation classification for non-image data.
2,406
2005.02359
title_snapshot
[ -0.02048158459365368, -0.02426646091043949, 0.012227533385157585, 0.03314957022666931, 0.05886992812156677, 0.0030515557155013084, 0.031228993088006973, -0.015865962952375412, -0.023539066314697266, -0.02171258069574833, -0.009172932244837284, 0.025423288345336914, -0.10497121512889862, -0...
Unrestricted Adversarial Examples via Semantic Manipulation
https://openreview.net/forum?id=Sye_OgHFwH
[ "Anand Bhattad", "Min Jin Chong", "Kaizhao Liang", "Bo Li", "D. A. Forsyth" ]
Poster
null
Machine learning models, especially deep neural networks (DNNs), have been shown to be vulnerable against adversarial examples which are carefully crafted samples with a small magnitude of the perturbation. Such adversarial perturbations are usually restricted by bounding their $\mathcal{L}_p$ norm such that they are ...
[ "Adversarial Examples", "Semantic Manipulation", "Image Colorization", "Texture Transfer" ]
We introduce unrestricted perturbations that manipulate semantically meaningful image-based visual descriptors - color and texture - in order to generate effective and photorealistic adversarial examples.
2,403
1904.06347
title_snapshot
[ 0.0008561889408156276, -0.037557680159807205, -0.027556823566555977, 0.055347997695207596, 0.0368708036839962, 0.005775713827461004, 0.03647011145949364, -0.012923164293169975, -0.02912455052137375, -0.023640742525458336, -0.05965270474553108, 0.001014423556625843, -0.06812970340251923, -0...
Discriminative Particle Filter Reinforcement Learning for Complex Partial observations
https://openreview.net/forum?id=HJl8_eHYvS
[ "Xiao Ma", "Peter Karkus", "David Hsu", "Wee Sun Lee", "Nan Ye" ]
Poster
null
Deep reinforcement learning is successful in decision making for sophisticated games, such as Atari, Go, etc. However, real-world decision making often requires reasoning with partial information extracted from complex visual observations. This paper presents Discriminative Particle Filter Reinforcement Learning (DPF...
[ "Reinforcement Learning", "Partial Observability", "Differentiable Particle Filtering" ]
We introduce DPFRL, a framework for reinforcement learning under partial and complex observations with an importance-weighted particle filter
2,399
2002.09884
title_snapshot
[ -0.026118772104382515, -0.0036596048157662153, 0.01092566829174757, 0.042842842638492584, 0.04289514198899269, 0.03584368899464607, 0.006208949722349644, 0.011139586567878723, -0.04127582907676697, -0.04530860483646393, -0.003451855154708028, 0.0031343260779976845, -0.08434943109750748, -0...
Learning to Group: A Bottom-Up Framework for 3D Part Discovery in Unseen Categories
https://openreview.net/forum?id=rkl8dlHYvB
[ "Tiange Luo", "Kaichun Mo", "Zhiao Huang", "Jiarui Xu", "Siyu Hu", "Liwei Wang", "Hao Su" ]
Poster
null
We address the problem of learning to discover 3D parts for objects in unseen categories. Being able to learn the geometry prior of parts and transfer this prior to unseen categories pose fundamental challenges on data-driven shape segmentation approaches. Formulated as a contextual bandit problem, we propose a learnin...
[ "Shape Segmentation", "Zero-Shot Learning", "Learning Representations" ]
A zero-shot segmentation framework for 3D shapes. Model the segmentation as a decision-making process, we propose an iterative method to dynamically extend the receptive field for achieving universal shape segmentation.
2,398
2002.06478
title_snapshot
[ 0.0035100996028631926, 0.0010226453887298703, 0.00866272859275341, 0.047016680240631104, 0.0201236791908741, 0.05061037838459015, 0.005114361643791199, -0.0008261032635346055, -0.02925659343600273, -0.044207390397787094, -0.056762926280498505, -0.004628332331776619, -0.040018003433942795, ...
State Alignment-based Imitation Learning
https://openreview.net/forum?id=rylrdxHFDr
[ "Fangchen Liu", "Zhan Ling", "Tongzhou Mu", "Hao Su" ]
Poster
null
Consider an imitation learning problem that the imitator and the expert have different dynamics models. Most of existing imitation learning methods fail because they focus on the imitation of actions. We propose a novel state alignment-based imitation learning method to train the imitator by following the state sequenc...
[ "Imitation learning", "Reinforcement Learning" ]
null
2,397
1911.10947
title_snapshot
[ -0.009006356820464134, -0.030427971854805946, -0.030977046117186546, 0.04273902624845505, 0.03589313477277756, 0.04256528243422508, 0.027530638501048088, 0.014468088746070862, -0.029899615794420242, -0.03590826317667961, 0.0038075274787843227, -0.0005976415704935789, -0.08028960973024368, ...
Neural Arithmetic Units
https://openreview.net/forum?id=H1gNOeHKPS
[ "Andreas Madsen", "Alexander Rosenberg Johansen" ]
Spotlight
null
Neural networks can approximate complex functions, but they struggle to perform exact arithmetic operations over real numbers. The lack of inductive bias for arithmetic operations leaves neural networks without the underlying logic necessary to extrapolate on tasks such as addition, subtraction, and multiplication. We ...
[]
null
2,395
2001.05016
title_snapshot
[ -0.006866893265396357, 0.009311330504715443, 0.0052125887013971806, 0.0308715607970953, 0.03934742137789726, 0.06822492927312851, 0.024715382605791092, 0.005066128447651863, -0.04953622445464134, -0.022811448201537132, 0.027359433472156525, -0.016793368384242058, -0.053503233939409256, 0.0...
Lipschitz constant estimation of Neural Networks via sparse polynomial optimization
https://openreview.net/forum?id=rJe4_xSFDB
[ "Fabian Latorre", "Paul Rolland", "Volkan Cevher" ]
Poster
null
We introduce LiPopt, a polynomial optimization framework for computing increasingly tighter upper bound on the Lipschitz constant of neural networks. The underlying optimization problems boil down to either linear (LP) or semidefinite (SDP) programming. We show how to use the sparse connectivity of a network, to signif...
[ "robust networks", "Lipschitz constant", "polynomial optimization" ]
LP-based upper bounds on the Lipschitz constant of Neural Networks
2,394
2004.08688
title_snapshot
[ -0.05242716148495674, -0.01233688835054636, -0.0027760069351643324, 0.056163009256124496, 0.0281708762049675, 0.04611651226878166, 0.023287013173103333, 0.006778014823794365, -0.020580533891916275, -0.02960793860256672, -0.003162241308018565, 0.018434546887874603, -0.06326261162757874, -0....
Effect of Activation Functions on the Training of Overparametrized Neural Nets
https://openreview.net/forum?id=rkgfdeBYvH
[ "Abhishek Panigrahi", "Abhishek Shetty", "Navin Goyal" ]
Poster
null
It is well-known that overparametrized neural networks trained using gradient based methods quickly achieve small training error with appropriate hyperparameter settings. Recent papers have proved this statement theoretically for highly overparametrized networks under reasonable assumptions. These results either assume...
[ "activation functions", "deep learning theory", "neural networks" ]
We provide theoretical results about the effect of activation function on the training of highly overparametrized 2-layer neural networks
2,390
1908.05660
title_snapshot
[ -0.05927490442991257, -0.019711047410964966, 0.017948955297470093, 0.03298806771636009, 0.02755184844136238, 0.0391271747648716, 0.032301049679517746, -0.0014842933742329478, -0.011228621937334538, -0.041843149811029434, 0.005301546771079302, 0.011502431705594063, -0.06067182868719101, 0.0...
Provable Filter Pruning for Efficient Neural Networks
https://openreview.net/forum?id=BJxkOlSYDH
[ "Lucas Liebenwein", "Cenk Baykal", "Harry Lang", "Dan Feldman", "Daniela Rus" ]
Poster
null
We present a provable, sampling-based approach for generating compact Convolutional Neural Networks (CNNs) by identifying and removing redundant filters from an over-parameterized network. Our algorithm uses a small batch of input data points to assign a saliency score to each filter and constructs an importance sampli...
[ "theory", "compression", "filter pruning", "neural networks" ]
A sampling-based filter pruning approach for convolutional neural networks exhibiting provable guarantees on the size and performance of the pruned network.
2,382
1911.07412
title_snapshot
[ -0.008940022438764572, -0.030334843322634697, 0.008924050256609917, 0.045563820749521255, 0.044746071100234985, 0.04287350922822952, -0.007770130410790443, -0.013369476422667503, -0.04138961434364319, -0.051654133945703506, -0.01742408238351345, -0.01705504208803177, -0.06811773031949997, ...
End to End Trainable Active Contours via Differentiable Rendering
https://openreview.net/forum?id=rkxawlHKDr
[ "Shir Gur", "Tal Shaharabany", "Lior Wolf" ]
Poster
null
We present an image segmentation method that iteratively evolves a polygon. At each iteration, the vertices of the polygon are displaced based on the local value of a 2D shift map that is inferred from the input image via an encoder-decoder architecture. The main training loss that is used is the difference between the...
[]
null
2,378
1912.00367
title_snapshot
[ 0.0014738261234015226, -0.01686011627316475, 0.02016640268266201, 0.030857669189572334, 0.016397159546613693, 0.04622117802500725, 0.013822225853800774, 0.01504242978990078, -0.04934673383831978, -0.09136322885751724, -0.03336208313703537, -0.028552355244755745, -0.027266662567853928, 0.01...
Compositional Language Continual Learning
https://openreview.net/forum?id=rklnDgHtDS
[ "Yuanpeng Li", "Liang Zhao", "Kenneth Church", "Mohamed Elhoseiny" ]
Poster
null
Motivated by the human's ability to continually learn and gain knowledge over time, several research efforts have been pushing the limits of machines to constantly learn while alleviating catastrophic forgetting. Most of the existing methods have been focusing on continual learning of label prediction tasks, which have...
[ "Compositionality", "Continual Learning", "Lifelong Learning", "Sequence to Sequence Modeling" ]
null
2,375
null
null
[ 0.0029247100464999676, -0.011357695795595646, -0.028161557391285896, 0.02676524966955185, 0.03927372395992279, 0.03715267777442932, 0.01747393235564232, 0.04430852830410004, -0.022195950150489807, -0.00874403677880764, -0.025760861113667488, 0.027449006214737892, -0.054628029465675354, 0.0...
Adversarial Lipschitz Regularization
https://openreview.net/forum?id=Bke_DertPB
[ "Dávid Terjék" ]
Poster
null
Generative adversarial networks (GANs) are one of the most popular approaches when it comes to training generative models, among which variants of Wasserstein GANs are considered superior to the standard GAN formulation in terms of learning stability and sample quality. However, Wasserstein GANs require the critic to b...
[ "generative adversarial networks", "wasserstein generative adversarial networks", "lipschitz regularization", "adversarial training" ]
alternative to gradient penalty
2,365
1907.05681
title_snapshot
[ -0.027360759675502777, -0.041337255388498306, 0.00826859287917614, 0.03814614564180374, 0.02716347947716713, 0.03139800950884819, 0.024310216307640076, 0.014428147114813328, -0.0119623439386487, -0.07166673988103867, -0.019520118832588196, 0.0016574631445109844, -0.053490735590457916, -0.0...
Adversarial Training and Provable Defenses: Bridging the Gap
https://openreview.net/forum?id=SJxSDxrKDr
[ "Mislav Balunovic", "Martin Vechev" ]
Talk
null
We present COLT, a new method to train neural networks based on a novel combination of adversarial training and provable defenses. The key idea is to model neural network training as a procedure which includes both, the verifier and the adversary. In every iteration, the verifier aims to certify the network using conve...
[ "adversarial examples", "adversarial training", "provable defense", "convex relaxations", "deep learning" ]
We propose a novel combination of adversarial training and provable defenses which produces a model with state-of-the-art accuracy and certified robustness on CIFAR-10.
2,360
null
null
[ -0.04000990465283394, -0.015954148024320602, -0.027636846527457237, 0.054148752242326736, 0.02874915674328804, 0.011621621437370777, 0.030916206538677216, -0.013626507483422756, -0.02942700870335102, -0.02992604300379753, -0.006472666747868061, 0.029801445081830025, -0.052165303379297256, ...
A Function Space View of Bounded Norm Infinite Width ReLU Nets: The Multivariate Case
https://openreview.net/forum?id=H1lNPxHKDH
[ "Greg Ongie", "Rebecca Willett", "Daniel Soudry", "Nathan Srebro" ]
Poster
null
We give a tight characterization of the (vectorized Euclidean) norm of weights required to realize a function $f:\mathbb{R}\rightarrow \mathbb{R}^d$ as a single hidden-layer ReLU network with an unbounded number of units (infinite width), extending the univariate characterization of Savarese et al. (2019) to the multiv...
[ "inductive bias", "regularization", "infinite-width networks", "ReLU networks" ]
We characterize the space of functions realizable as a ReLU network with an unbounded number of units (infinite width), but where the Euclidean norm of the weights is bounded.
2,356
1910.01635
title_snapshot
[ -0.008349212817847729, -0.03219309076666832, 0.004115558695048094, -0.004121709149330854, 0.035002004355192184, 0.04549576714634895, 0.03777264058589935, -0.02174500748515129, -0.03024875372648239, -0.044030848890542984, -0.008844265714287758, 0.0068741957657039165, -0.08518412709236145, -...
Lite Transformer with Long-Short Range Attention
https://openreview.net/forum?id=ByeMPlHKPH
[ "Zhanghao Wu*", "Zhijian Liu*", "Ji Lin", "Yujun Lin", "Song Han" ]
Poster
null
Transformer has become ubiquitous in natural language processing (e.g., machine translation, question answering); however, it requires enormous amount of computations to achieve high performance, which makes it not suitable for mobile applications that are tightly constrained by the hardware resources and battery. In t...
[ "efficient model", "transformer" ]
null
2,353
2004.11886
title_snapshot
[ -0.028984306380152702, -0.01960490085184574, 0.01719421148300171, 0.011760411784052849, 0.0306156724691391, 0.02860991097986698, 0.012822646647691727, 0.012861023657023907, -0.008792716078460217, -0.0037926416844129562, -0.047829825431108475, 0.024180060252547264, -0.07438965886831284, 0.0...
Mutual Information Gradient Estimation for Representation Learning
https://openreview.net/forum?id=ByxaUgrFvH
[ "Liangjian Wen", "Yiji Zhou", "Lirong He", "Mingyuan Zhou", "Zenglin Xu" ]
Poster
null
Mutual Information (MI) plays an important role in representation learning. However, MI is unfortunately intractable in continuous and high-dimensional settings. Recent advances establish tractable and scalable MI estimators to discover useful representation. However, most of the existing methods are not capable of pro...
[ "Mutual Information", "Score Estimation", "Representation Learning", "Information Bottleneck" ]
null
2,341
2005.01123
title_snapshot
[ -0.01358655747026205, 0.0048339879140257835, -0.011732696555554867, 0.027174536138772964, 0.01920871064066887, 0.039847686886787415, 0.03303905576467514, -0.0026767223607748747, -0.025735951960086823, -0.03042932040989399, 0.0031669435556977987, 0.009382561780512333, -0.06213916465640068, ...
Regularizing activations in neural networks via distribution matching with the Wasserstein metric
https://openreview.net/forum?id=rygwLgrYPB
[ "Taejong Joo", "Donggu Kang", "Byunghoon Kim" ]
Poster
null
Regularization and normalization have become indispensable components in training deep neural networks, resulting in faster training and improved generalization performance. We propose the projected error function regularization loss (PER) that encourages activations to follow the standard normal distribution. PER rand...
[ "regularization", "Wasserstein metric", "deep learning" ]
null
2,326
2002.05366
title_snapshot
[ -0.021677633747458458, -0.023620612919330597, -0.01081193145364523, 0.039344921708106995, 0.029468417167663574, 0.046145860105752945, 0.025096112862229347, -0.010713383555412292, -0.017575107514858246, -0.033687081187963486, -0.007532804738730192, -0.0035019749775528908, -0.0528007335960865,...
Gradient Descent Maximizes the Margin of Homogeneous Neural Networks
https://openreview.net/forum?id=SJeLIgBKPS
[ "Kaifeng Lyu", "Jian Li" ]
Talk
null
In this paper, we study the implicit regularization of the gradient descent algorithm in homogeneous neural networks, including fully-connected and convolutional neural networks with ReLU or LeakyReLU activations. In particular, we study the gradient descent or gradient flow (i.e., gradient descent with infinitesimal s...
[ "margin", "homogeneous", "gradient descent" ]
We study the implicit bias of gradient descent and prove under a minimal set of assumptions that the parameter direction of homogeneous models converges to KKT points of a natural margin maximization problem.
2,324
1906.05890
title_snapshot
[ -0.018174072727560997, -0.017652137205004692, 0.007623977959156036, 0.04114421084523201, 0.03421838954091072, 0.0214744433760643, 0.04059289023280144, 0.0022160825319588184, -0.03706931322813034, -0.03945869952440262, 0.01739572547376156, -0.01904209703207016, -0.05966031551361084, 0.00301...
Transferring Optimality Across Data Distributions via Homotopy Methods
https://openreview.net/forum?id=S1gEIerYwH
[ "Matilde Gargiani", "Andrea Zanelli", "Quoc Tran Dinh", "Moritz Diehl", "Frank Hutter" ]
Poster
null
Homotopy methods, also known as continuation methods, are a powerful mathematical tool to efficiently solve various problems in numerical analysis, including complex non-convex optimization problems where no or only little prior knowledge regarding the localization of the solutions is available. In this work, we propo...
[ "deep learning", "numerical optimization", "transfer learning" ]
We propose a new homotopy-based method to transfer "optimality knowledge" across different data distributions in order to speed up training of deep models.
2,320
null
null
[ -0.029468394815921783, -0.027963872998952866, 0.007510406896471977, 0.026391053572297096, 0.044048164039850235, 0.06566827744245529, 0.018325483426451683, -0.0384533517062664, -0.020956387743353844, -0.06645497679710388, -0.011135953478515148, -0.01936328038573265, -0.06675764918327332, 0....
Latent Normalizing Flows for Many-to-Many Cross-Domain Mappings
https://openreview.net/forum?id=SJxE8erKDH
[ "Shweta Mahajan", "Iryna Gurevych", "Stefan Roth" ]
Poster
null
Learned joint representations of images and text form the backbone of several important cross-domain tasks such as image captioning. Prior work mostly maps both domains into a common latent representation in a purely supervised fashion. This is rather restrictive, however, as the two domains follow distinct generative ...
[]
null
2,319
2002.06661
title_snapshot
[ 0.004683136474341154, -0.043141525238752365, -0.00962720438838005, 0.04759657010436058, 0.017103487625718117, 0.012510355561971664, 0.018161578103899956, -0.008585023693740368, 0.005327960476279259, -0.027040787041187286, -0.030217887833714485, -0.007768742274492979, -0.042193491011857986, ...
Dynamic Model Pruning with Feedback
https://openreview.net/forum?id=SJem8lSFwB
[ "Tao Lin", "Sebastian U. Stich", "Luis Barba", "Daniil Dmitriev", "Martin Jaggi" ]
Poster
null
Deep neural networks often have millions of parameters. This can hinder their deployment to low-end devices, not only due to high memory requirements but also because of increased latency at inference. We propose a novel model compression method that generates a sparse trained model without additional overhead: by allo...
[ "network pruning", "dynamic reparameterization", "model compression" ]
null
2,317
2006.07253
title_snapshot
[ -0.008799592033028603, -0.05710805952548981, -0.023132042959332466, 0.041110485792160034, 0.038574300706386566, 0.06579005718231201, 0.001626987592317164, -0.0003734930360224098, -0.05416344106197357, -0.06179632991552353, 0.010645542293787003, -0.0060553159564733505, -0.05319688841700554, ...
On the interaction between supervision and self-play in emergent communication
https://openreview.net/forum?id=rJxGLlBtwH
[ "Ryan Lowe*", "Abhinav Gupta*", "Jakob Foerster", "Douwe Kiela", "Joelle Pineau" ]
Poster
null
A promising approach for teaching artificial agents to use natural language involves using human-in-the-loop training. However, recent work suggests that current machine learning methods are too data inefficient to be trained in this way from scratch. In this paper, we investigate the relationship between two categorie...
[ "multi-agent communication", "self-play", "emergent languages" ]
null
2,315
2002.01093
title_snapshot
[ -0.027204686775803566, -0.03248212859034538, -0.02105526067316532, 0.028869641944766045, 0.049239255487918854, 0.0052221715450286865, 0.03561456501483917, 0.01026181224733591, -0.01539959479123354, -0.03168412670493126, -0.0334874652326107, 0.047711171209812164, -0.07531796395778656, -0.00...
A Meta-Transfer Objective for Learning to Disentangle Causal Mechanisms
https://openreview.net/forum?id=ryxWIgBFPS
[ "Yoshua Bengio", "Tristan Deleu", "Nasim Rahaman", "Nan Rosemary Ke", "Sebastien Lachapelle", "Olexa Bilaniuk", "Anirudh Goyal", "Christopher Pal" ]
Poster
null
We propose to use a meta-learning objective that maximizes the speed of transfer on a modified distribution to learn how to modularize acquired knowledge. In particular, we focus on how to factor a joint distribution into appropriate conditionals, consistent with the causal directions. We explain when this can work, us...
[ "meta-learning", "transfer learning", "structure learning", "modularity", "causality" ]
This paper proposes a meta-learning objective based on speed of adaptation to transfer distributions to discover a modular decomposition and causal variables.
2,311
1901.10912
title_snapshot
[ -0.009237394668161869, -0.006659194361418486, 0.0008277701563201845, 0.031670182943344116, 0.04382278025150299, 0.0360909067094326, 0.026388021185994148, -0.0035090046003460884, -0.01804673671722412, -0.031896889209747314, -0.0034810525830835104, 0.03002208285033703, -0.06029970198869705, ...
Expected Information Maximization: Using the I-Projection for Mixture Density Estimation
https://openreview.net/forum?id=ByglLlHFDS
[ "Philipp Becker", "Oleg Arenz", "Gerhard Neumann" ]
Poster
null
Modelling highly multi-modal data is a challenging problem in machine learning. Most algorithms are based on maximizing the likelihood, which corresponds to the M(oment)-projection of the data distribution to the model distribution. The M-projection forces the model to average over modes it cannot represent. In contras...
[ "density estimation", "information projection", "mixture models", "generative learning", "multimodal modeling" ]
A novel, non-adversarial, approach to learn latent variable models in general and mixture models in particular by computing the I-Projection solely based on samples.
2,310
2001.08682
title_snapshot
[ -0.007089992985129356, -0.010239197872579098, 0.024778353050351143, 0.05007199943065643, 0.03580065071582794, 0.03196007013320923, 0.01965508610010147, 0.006708941888064146, -0.021400053054094315, -0.03394654765725136, -0.029774796217679977, 0.021928558126091957, -0.06403812021017075, -0.0...
Truth or backpropaganda? An empirical investigation of deep learning theory
https://openreview.net/forum?id=HyxyIgHFvr
[ "Micah Goldblum", "Jonas Geiping", "Avi Schwarzschild", "Michael Moeller", "Tom Goldstein" ]
Spotlight
null
We empirically evaluate common assumptions about neural networks that are widely held by practitioners and theorists alike. In this work, we: (1) prove the widespread existence of suboptimal local minima in the loss landscape of neural networks, and we use our theory to find examples; (2) show that small-norm paramete...
[ "Deep learning", "generalization", "loss landscape", "robustness" ]
We call into question commonly held beliefs regarding the loss landscape, optimization, network width, and rank.
2,307
1910.00359
title_snapshot
[ -0.04818292707204819, -0.03288428857922554, -0.013494456186890602, 0.053447604179382324, 0.05101761221885681, 0.026462946087121964, 0.020343950018286705, 0.0128999724984169, -0.03557491675019264, -0.04397495090961456, 0.01320564653724432, 0.03402434289455414, -0.037331193685531616, -0.0092...
Deep Audio Priors Emerge From Harmonic Convolutional Networks
https://openreview.net/forum?id=rygjHxrYDB
[ "Zhoutong Zhang", "Yunyun Wang", "Chuang Gan", "Jiajun Wu", "Joshua B. Tenenbaum", "Antonio Torralba", "William T. Freeman" ]
Poster
null
Convolutional neural networks (CNNs) excel in image recognition and generation. Among many efforts to explain their effectiveness, experiments show that CNNs carry strong inductive biases that capture natural image priors. Do deep networks also have inductive biases for audio signals? In this paper, we empirically show...
[ "Audio", "Deep Prior" ]
A new operation called Harmonic Convolution makes deep network model audio priors without training.
2,298
null
null
[ -0.0008142178412526846, -0.018577300012111664, -0.0027290452271699905, 0.055472344160079956, 0.024229636415839195, 0.028051920235157013, 0.0399491973221302, 0.015758411958813667, -0.04152596369385719, -0.06629818677902222, 0.02908731997013092, 0.010415051132440567, -0.044434111565351486, 0...
Drawing Early-Bird Tickets: Toward More Efficient Training of Deep Networks
https://openreview.net/forum?id=BJxsrgStvr
[ "Haoran You", "Chaojian Li", "Pengfei Xu", "Yonggan Fu", "Yue Wang", "Xiaohan Chen", "Richard G. Baraniuk", "Zhangyang Wang", "Yingyan Lin" ]
Spotlight
null
(Frankle & Carbin, 2019) shows that there exist winning tickets (small but critical subnetworks) for dense, randomly initialized networks, that can be trained alone to achieve comparable accuracies to the latter in a similar number of iterations. However, the identification of these winning tickets still requires the c...
[]
null
2,297
1909.11957
title_judge
[ -0.04273761063814163, -0.05142515152692795, -0.0129108726978302, 0.03958936780691147, 0.04512636736035347, 0.015974629670381546, 0.011004630476236343, 0.009121083654463291, -0.017260130494832993, -0.040441930294036865, -0.0021538957953453064, -0.017894230782985687, -0.056475840508937836, -...
Improving Generalization in Meta Reinforcement Learning using Learned Objectives
https://openreview.net/forum?id=S1evHerYPr
[ "Louis Kirsch", "Sjoerd van Steenkiste", "Juergen Schmidhuber" ]
Spotlight
null
Biological evolution has distilled the experiences of many learners into the general learning algorithms of humans. Our novel meta reinforcement learning algorithm MetaGenRL is inspired by this process. MetaGenRL distills the experiences of many complex agents to meta-learn a low-complexity neural objective function th...
[ "meta reinforcement learning", "meta learning", "reinforcement learning" ]
We introduce MetaGenRL, a novel meta reinforcement learning algorithm. Unlike prior work, MetaGenRL can generalize to new environments that are entirely different from those used for meta-training.
2,289
1910.04098
title_snapshot
[ -0.00753948325291276, -0.037244439125061035, 0.010063759982585907, 0.022444220259785652, 0.03989984840154648, 0.005126968491822481, 0.026760602369904518, 0.0019122464582324028, -0.04443446919322014, -0.010847073048353195, -0.02102285996079445, 0.04235957935452461, -0.08761138468980789, -0....
A closer look at the approximation capabilities of neural networks
https://openreview.net/forum?id=rkevSgrtPr
[ "Kai Fong Ernest Chong" ]
Poster
null
The universal approximation theorem, in one of its most general versions, says that if we consider only continuous activation functions σ, then a standard feedforward neural network with one hidden layer is able to approximate any continuous multivariate function f to any given approximation threshold ε, if and only if...
[ "deep learning", "approximation", "universal approximation theorem" ]
A quantitative refinement of the universal approximation theorem via an algebraic approach.
2,288
2002.06505
title_snapshot
[ -0.03527483344078064, -0.006387050729244947, 0.010859787464141846, 0.027607059106230736, 0.029955187812447548, 0.04821731522679329, 0.023949453607201576, 0.00042223691707476974, -0.029658865183591843, -0.021098310127854347, -0.009881124831736088, 0.00906634982675314, -0.06256973743438721, ...
Kaleidoscope: An Efficient, Learnable Representation For All Structured Linear Maps
https://openreview.net/forum?id=BkgrBgSYDS
[ "Tri Dao", "Nimit Sohoni", "Albert Gu", "Matthew Eichhorn", "Amit Blonder", "Megan Leszczynski", "Atri Rudra", "Christopher Ré" ]
Spotlight
null
Modern neural network architectures use structured linear transformations, such as low-rank matrices, sparse matrices, permutations, and the Fourier transform, to improve inference speed and reduce memory usage compared to general linear maps. However, choosing which of the myriad structured transformations to use (and...
[ "structured matrices", "efficient ML", "algorithms", "butterfly matrices", "arithmetic circuits" ]
We propose a differentiable family of "kaleidoscope matrices," prove that all structured matrices can be represented in this form, and use them to replace hand-crafted linear maps in deep learning models.
2,283
2012.14966
title_snapshot
[ -0.021922828629612923, -0.009492326527833939, -0.02968750335276127, 0.029226094484329224, 0.018255606293678284, 0.03437506780028343, 0.01721915230154991, 0.0031417873688042164, 0.00993113499134779, -0.019467461854219437, -0.00628685113042593, -0.01692587323486805, -0.07563610374927521, 0.0...
Residual Energy-Based Models for Text Generation
https://openreview.net/forum?id=B1l4SgHKDH
[ "Yuntian Deng", "Anton Bakhtin", "Myle Ott", "Arthur Szlam", "Marc'Aurelio Ranzato" ]
Poster
null
Text generation is ubiquitous in many NLP tasks, from summarization, to dialogue and machine translation. The dominant parametric approach is based on locally normalized models which predict one word at a time. While these work remarkably well, they are plagued by exposure bias due to the greedy nature of the generatio...
[ "energy-based models", "text generation" ]
We show that Energy-Based models when trained on the residual of an auto-regressive language model can be used effectively and efficiently to generate text.
2,281
2004.11714
title_snapshot
[ -0.00507770199328661, -0.01823444291949272, -0.004307270981371403, 0.07359600067138672, 0.039556264877319336, 0.015856796875596046, 0.013546944595873356, 0.023680493235588074, -0.010500400327146053, -0.019291382282972336, 0.0022203789558261633, 0.04192561283707619, -0.06496366113424301, -0...
AtomNAS: Fine-Grained End-to-End Neural Architecture Search
https://openreview.net/forum?id=BylQSxHFwr
[ "Jieru Mei", "Yingwei Li", "Xiaochen Lian", "Xiaojie Jin", "Linjie Yang", "Alan Yuille", "Jianchao Yang" ]
Poster
null
Search space design is very critical to neural architecture search (NAS) algorithms. We propose a fine-grained search space comprised of atomic blocks, a minimal search unit that is much smaller than the ones used in recent NAS algorithms. This search space allows a mix of operations by composing different types of ato...
[ "Neural Architecture Search", "Image Classification" ]
A new state-of-the-art on Imagenet for mobile setting
2,280
1912.09640
title_snapshot
[ -0.005287536419928074, 0.014883184805512428, -0.04682017117738724, 0.05578272044658661, 0.0537809394299984, 0.03881615027785301, 0.0022475975565612316, 0.005288738291710615, -0.019657807424664497, -0.05023787170648575, 0.00896555557847023, -0.04189393296837807, -0.05647918954491615, 0.0007...
AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty
https://openreview.net/forum?id=S1gmrxHFvB
[ "Dan Hendrycks*", "Norman Mu*", "Ekin Dogus Cubuk", "Barret Zoph", "Justin Gilmer", "Balaji Lakshminarayanan" ]
Poster
null
Modern deep neural networks can achieve high accuracy when the training distribution and test distribution are identically distributed, but this assumption is frequently violated in practice. When the train and test distributions are mismatched, accuracy can plummet. Currently there are few techniques that improve robu...
[ "robustness", "uncertainty" ]
We obtain state-of-the-art on robustness to data shifts, and we maintain calibration under data shift even though even when accuracy drops
2,278
1912.02781
title_snapshot
[ 0.01822768524289131, -0.027777530252933502, -0.053768038749694824, 0.06648135185241699, 0.022023571655154228, 0.018262363970279694, 0.014945789240300655, 0.0052144392393529415, -0.012723308987915516, -0.0836232602596283, -0.012262695468962193, -0.02703949436545372, -0.08050350099802017, 0....
Disentanglement by Nonlinear ICA with General Incompressible-flow Networks (GIN)
https://openreview.net/forum?id=rygeHgSFDH
[ "Peter Sorrenson", "Carsten Rother", "Ullrich Köthe" ]
Spotlight
null
A central question of representation learning asks under which conditions it is possible to reconstruct the true latent variables of an arbitrarily complex generative process. Recent breakthrough work by Khemakhem et al. (2019) on nonlinear ICA has answered this question for a broad class of conditional generative proc...
[ "disentanglement", "nonlinear ICA", "representation learning", "feature discovery", "theoretical justification" ]
null
2,272
2001.04872
title_snapshot
[ -0.01039203442633152, -0.024348795413970947, 0.003109411336481571, 0.03967965766787529, 0.03456423804163933, 0.03569439798593521, 0.0509476438164711, 0.021727746352553368, -0.026708323508501053, -0.01744033396244049, -0.010894388891756535, -0.033269625157117844, -0.0706230103969574, 0.0097...
Memory-Based Graph Networks
https://openreview.net/forum?id=r1laNeBYPB
[ "Amir Hosein Khasahmadi", "Kaveh Hassani", "Parsa Moradi", "Leo Lee", "Quaid Morris" ]
Poster
null
Graph neural networks (GNNs) are a class of deep models that operate on data with arbitrary topology represented as graphs. We introduce an efficient memory layer for GNNs that can jointly learn node representations and coarsen the graph. We also introduce two new networks based on this layer: memory-based GNN (MemGNN)...
[ "Graph Neural Networks", "Memory Networks", "Hierarchial Graph Representation Learning" ]
We introduce an efficient memory layer to jointly learn representations and coarsen the input graphs.
2,266
2002.09518
title_snapshot
[ -0.031019359827041626, 0.018652740865945816, 0.013792631216347218, 0.046555839478969574, 0.029326561838388443, -0.0014142758445814252, 0.008601595647633076, 0.0370115227997303, -0.01690089702606201, -0.04024561122059822, 0.03944532945752144, -0.005176635924726725, -0.07363902032375336, 0.0...
Variational Template Machine for Data-to-Text Generation
https://openreview.net/forum?id=HkejNgBtPB
[ "Rong Ye", "Wenxian Shi", "Hao Zhou", "Zhongyu Wei", "Lei Li" ]
Poster
null
How to generate descriptions from structured data organized in tables? Existing approaches using neural encoder-decoder models often suffer from lacking diversity. We claim that an open set of templates is crucial for enriching the phrase constructions and realizing varied generations.Learning such templates is prohibi...
[]
null
2,261
2002.01127
title_snapshot
[ 0.01406897697597742, 0.010042106732726097, -0.04688914120197296, 0.07688596099615097, 0.03231756016612053, 0.033060409128665924, 0.020031679421663284, 0.028055861592292786, -0.01992831751704216, -0.017262648791074753, -0.03295601159334183, 0.023302190005779266, -0.04911778122186661, 0.0207...
Phase Transitions for the Information Bottleneck in Representation Learning
https://openreview.net/forum?id=HJloElBYvB
[ "Tailin Wu", "Ian Fischer" ]
Poster
null
In the Information Bottleneck (IB), when tuning the relative strength between compression and prediction terms, how do the two terms behave, and what's their relationship with the dataset and the learned representation? In this paper, we set out to answer these questions by studying multiple phase transitions in the IB...
[ "Information Theory", "Representation Learning", "Phase Transition" ]
We give a theoretical analysis of the Information Bottleneck objective to understand and predict observed phase transitions in the prediction vs. compression tradeoff.
2,260
2001.01878
title_snapshot
[ -0.027348220348358154, -0.02253870852291584, -0.013234010897576809, 0.02156018279492855, 0.04146481677889824, 0.004558041226118803, 0.020003294572234154, 0.01873893477022648, -0.0454479455947876, -0.01880274899303913, -0.019205424934625626, -0.0033085711766034365, -0.04985397681593895, 0.0...
Continual learning with hypernetworks
https://openreview.net/forum?id=SJgwNerKvB
[ "Johannes von Oswald", "Christian Henning", "Benjamin F. Grewe", "João Sacramento" ]
Spotlight
null
Artificial neural networks suffer from catastrophic forgetting when they are sequentially trained on multiple tasks. To overcome this problem, we present a novel approach based on task-conditioned hypernetworks, i.e., networks that generate the weights of a target model based on task identity. Continual learning (CL) i...
[ "Continual Learning", "Catastrophic Forgetting", "Meta Model", "Hypernetwork" ]
null
2,252
1906.00695
title_snapshot
[ 0.008598052896559238, -0.015422928147017956, 0.009734435006976128, 0.02630843222141266, 0.05296720936894417, 0.010818428359925747, 0.00945729948580265, -0.0020500675309449434, -0.050013866275548935, -0.042274754494428635, -0.006027588155120611, -0.0021001831628382206, -0.05619273707270622, ...
Permutation Equivariant Models for Compositional Generalization in Language
https://openreview.net/forum?id=SylVNerFvr
[ "Jonathan Gordon", "David Lopez-Paz", "Marco Baroni", "Diane Bouchacourt" ]
Poster
null
Humans understand novel sentences by composing meanings and roles of core language components. In contrast, neural network models for natural language modeling fail when such compositional generalization is required. The main contribution of this paper is to hypothesize that language compositionality is a form of group...
[ "Compositionality", "Permutation Equivariance", "Language Processing" ]
We propose a link between permutation equivariance and compositional generalization, and provide equivariant language models
2,245
null
null
[ 0.010938561521470547, 0.007151956669986248, -0.016681460663676262, 0.03460445627570152, 0.016081789508461952, 0.04244491085410118, 0.039792321622371674, 0.051299408078193665, -0.03247155249118805, -0.008038563653826714, -0.013313512317836285, 0.005405709613114595, -0.06321997195482254, 0.0...
Training binary neural networks with real-to-binary convolutions
https://openreview.net/forum?id=BJg4NgBKvH
[ "Brais Martinez", "Jing Yang", "Adrian Bulat", "Georgios Tzimiropoulos" ]
Poster
null
This paper shows how to train binary networks to within a few percent points (~3-5%) of the full precision counterpart. We first show how to build a strong baseline, which already achieves state-of-the-art accuracy, by combining recently proposed advances and carefully adjusting the optimization procedure. Secondly, we...
[ "binary networks" ]
null
2,244
2003.11535
title_snapshot
[ 0.017677320167422295, -0.03515421971678734, 0.0020657768473029137, 0.058072831481695175, 0.01667649671435356, 0.03597165271639824, 0.002615598263218999, 0.03676663711667061, -0.011392880231142044, -0.03245002031326294, -0.011652910150587559, -0.011055687442421913, -0.06604529172182083, -0....
StructBERT: Incorporating Language Structures into Pre-training for Deep Language Understanding
https://openreview.net/forum?id=BJgQ4lSFPH
[ "Wei Wang", "Bin Bi", "Ming Yan", "Chen Wu", "Jiangnan Xia", "Zuyi Bao", "Liwei Peng", "Luo Si" ]
Poster
null
Recently, the pre-trained language model, BERT (and its robustly optimized version RoBERTa), has attracted a lot of attention in natural language understanding (NLU), and achieved state-of-the-art accuracy in various NLU tasks, such as sentiment classification, natural language inference, semantic textual similarity an...
[]
null
2,242
1908.04577
title_snapshot
[ -0.012676275335252285, -0.03782126307487488, 0.01700666733086109, 0.05342712998390198, 0.015310400165617466, 0.04128298535943031, 0.015573997981846333, 0.021509118378162384, -0.011495708487927914, -0.014444726519286633, -0.02448733150959015, 0.046440307050943375, -0.03694425895810127, 0.00...
Smooth markets: A basic mechanism for organizing gradient-based learners
https://openreview.net/forum?id=B1xMEerYvB
[ "David Balduzzi", "Wojciech M. Czarnecki", "Tom Anthony", "Ian Gemp", "Edward Hughes", "Joel Leibo", "Georgios Piliouras", "Thore Graepel" ]
Poster
null
With the success of modern machine learning, it is becoming increasingly important to understand and control how learning algorithms interact. Unfortunately, negative results from game theory show there is little hope of understanding or controlling general n-player games. We therefore introduce smooth markets (SM-game...
[ "game theory", "optimization", "gradient descent", "adversarial learning" ]
We introduce a class of n-player games suited to gradient-based methods.
2,240
2001.04678
title_snapshot
[ -0.051159318536520004, -0.03776136785745621, 0.014335477724671364, 0.049187738448381424, 0.037522900849580765, 0.020949846133589745, 0.01018553227186203, 0.007258691359311342, -0.019892821088433266, -0.01459193229675293, -0.028223052620887756, 0.03812810406088829, -0.07744023203849792, -0....
Fair Resource Allocation in Federated Learning
https://openreview.net/forum?id=ByexElSYDr
[ "Tian Li", "Maziar Sanjabi", "Ahmad Beirami", "Virginia Smith" ]
Poster
null
Federated learning involves training statistical models in massive, heterogeneous networks. Naively minimizing an aggregate loss function in such a network may disproportionately advantage or disadvantage some of the devices. In this work, we propose q-Fair Federated Learning (q-FFL), a novel optimization objective ins...
[ "federated learning", "fairness", "distributed optimization" ]
We propose a novel optimization objective that encourages fairness in heterogeneous federated networks, and develop a scalable method to solve it.
2,235
1905.10497
title_snapshot
[ -0.010555820539593697, -0.05143513157963753, 0.0012830077903345227, 0.02141978219151497, 0.043671295046806335, 0.01269568596035242, 0.005021651275455952, -0.007200987543910742, -0.0362153984606266, -0.030542323365807533, 0.014503519050776958, 0.006909667514264584, -0.06883533298969269, -0....
Convolutional Conditional Neural Processes
https://openreview.net/forum?id=Skey4eBYPS
[ "Jonathan Gordon", "Wessel P. Bruinsma", "Andrew Y. K. Foong", "James Requeima", "Yann Dubois", "Richard E. Turner" ]
Talk
null
We introduce the Convolutional Conditional Neural Process (ConvCNP), a new member of the Neural Process family that models translation equivariance in the data. Translation equivariance is an important inductive bias for many learning problems including time series modelling, spatial data, and images. The model embeds ...
[ "Neural Processes", "Deep Sets", "Translation Equivariance" ]
We extend deep sets to functional embeddings and Neural Processes to include translation equivariant members
2,232
1910.13556
manual
[ 0.00928612519055605, -0.022418219596147537, -0.007026576437056065, 0.03237256035208702, 0.026971299201250076, 0.04684910178184509, 0.0059923999942839146, 0.00782615877687931, 0.005012992303818464, -0.03624589741230011, -0.01915573514997959, -0.0005695989239029586, -0.06201418116688728, 0.0...
Meta-Learning with Warped Gradient Descent
https://openreview.net/forum?id=rkeiQlBFPB
[ "Sebastian Flennerhag", "Andrei A. Rusu", "Razvan Pascanu", "Francesco Visin", "Hujun Yin", "Raia Hadsell" ]
Talk
null
Learning an efficient update rule from data that promotes rapid learning of new tasks from the same distribution remains an open problem in meta-learning. Typically, previous works have approached this issue either by attempting to train a neural network that directly produces updates or by attempting to learn better i...
[ "meta-learning", "transfer learning" ]
We propose a novel framework for meta-learning a gradient-based update rule that scales to beyond few-shot learning and is applicable to any form of learning, including continual learning.
2,223
1909.00025
title_snapshot
[ 0.004917536862194538, -0.036438144743442535, 0.012821357697248459, 0.05313395708799362, 0.0308857224881649, 0.030626673251390457, 0.034663863480091095, 0.011883450672030449, -0.060525842010974884, -0.047128502279520035, -0.01751786842942238, 0.03731556609272957, -0.06437516957521439, 0.000...
Never Give Up: Learning Directed Exploration Strategies
https://openreview.net/forum?id=Sye57xStvB
[ "Adrià Puigdomènech Badia", "Pablo Sprechmann", "Alex Vitvitskyi", "Daniel Guo", "Bilal Piot", "Steven Kapturowski", "Olivier Tieleman", "Martin Arjovsky", "Alexander Pritzel", "Andrew Bolt", "Charles Blundell" ]
Poster
null
We propose a reinforcement learning agent to solve hard exploration games by learning a range of directed exploratory policies. We construct an episodic memory-based intrinsic reward using k-nearest neighbors over the agent's recent experience to train the directed exploratory policies, thereby encouraging the agent to...
[ "deep reinforcement learning", "exploration", "intrinsic motivation" ]
We propose a reinforcement learning agent to solve hard exploration games by learning a range of directed exploratory policies.
2,222
2002.06038
title_snapshot
[ -0.05215553194284439, -0.014833755791187286, 0.0008927127346396446, 0.04377979785203934, 0.044981200248003006, 0.02154180407524109, 0.009811694733798504, -0.006909404415637255, -0.035797830671072006, -0.029269501566886902, -0.016473067924380302, 0.014893868938088417, -0.05269046872854233, ...
AdvectiveNet: An Eulerian-Lagrangian Fluidic Reservoir for Point Cloud Processing
https://openreview.net/forum?id=H1eqQeHFDS
[ "Xingzhe He", "Helen Lu Cao", "Bo Zhu" ]
Poster
null
This paper presents a novel physics-inspired deep learning approach for point cloud processing motivated by the natural flow phenomena in fluid mechanics. Our learning architecture jointly defines data in an Eulerian world space, using a static background grid, and a Lagrangian material space, using moving particles. ...
[ "Point Cloud Processing", "Physical Reservoir Learning", "Eulerian-Lagrangian Method", "PIC/FLIP" ]
We present a new grid-particle learning method to process point clouds motivated by computational fluid dynamics.
2,221
2002.00118
title_snapshot
[ 0.002564795780926943, -0.019699042662978172, 0.03289608657360077, 0.024200856685638428, 0.00493357190862298, 0.03329945728182793, -0.00758892297744751, 0.01858489029109478, -0.0525757260620594, -0.05662013962864876, 0.0038200439885258675, -0.056610312312841415, -0.05915471166372299, 0.0309...
SEED RL: Scalable and Efficient Deep-RL with Accelerated Central Inference
https://openreview.net/forum?id=rkgvXlrKwH
[ "Lasse Espeholt", "Raphaël Marinier", "Piotr Stanczyk", "Ke Wang", "Marcin Michalski‎" ]
Talk
null
We present a modern scalable reinforcement learning agent called SEED (Scalable, Efficient Deep-RL). By effectively utilizing modern accelerators, we show that it is not only possible to train on millions of frames per second but also to lower the cost. of experiments compared to current methods. We achieve this with a...
[ "machine learning", "reinforcement learning", "scalability", "distributed", "DeepMind Lab", "ALE", "Atari-57", "Google Research Football" ]
SEED RL, a scalable and efficient deep reinforcement learning agent with accelerated central inference. State of the art results, reduces cost and can process millions of frames per second.
2,213
1910.06591
title_snapshot
[ -0.02687111496925354, -0.06352359056472778, 0.00318334111943841, 0.06258069723844528, 0.03564095497131348, 0.03399505838751793, 0.005395232234150171, 0.02159939892590046, -0.0270905252546072, -0.03344627097249031, 0.024316810071468353, 0.003730485215783119, -0.0787021592259407, -0.01108853...
You CAN Teach an Old Dog New Tricks! On Training Knowledge Graph Embeddings
https://openreview.net/forum?id=BkxSmlBFvr
[ "Daniel Ruffinelli", "Samuel Broscheit", "Rainer Gemulla" ]
Poster
null
Knowledge graph embedding (KGE) models learn algebraic representations of the entities and relations in a knowledge graph. A vast number of KGE techniques for multi-relational link prediction have been proposed in the recent literature, often with state-of-the-art performance. These approaches differ along a number of ...
[ "knowledge graph embeddings", "hyperparameter optimization" ]
We study the impact of training strategies on the performance of knowledge graph embeddings.
2,209
null
null
[ 0.0018668730044737458, -0.020038099959492683, 0.004068789537996054, 0.053722795099020004, 0.04717644304037094, -0.003724541049450636, 0.04349927976727486, -0.03203700855374336, 0.017303824424743652, -0.018721813336014748, -0.010770679451525211, 0.0305315051227808, -0.06702904403209686, 0.0...
High Fidelity Speech Synthesis with Adversarial Networks
https://openreview.net/forum?id=r1gfQgSFDr
[ "Mikołaj Bińkowski", "Jeff Donahue", "Sander Dieleman", "Aidan Clark", "Erich Elsen", "Norman Casagrande", "Luis C. Cobo", "Karen Simonyan" ]
Talk
null
Generative adversarial networks have seen rapid development in recent years and have led to remarkable improvements in generative modelling of images. However, their application in the audio domain has received limited attention, and autoregressive models, such as WaveNet, remain the state of the art in generative mode...
[ "texttospeech", "speechsynthesis", "audiosynthesis", "gans", "generativeadversarialnetworks", "implicitgenerativemodels" ]
We introduce GAN-TTS, a Generative Adversarial Network for Text-to-Speech, which achieves Mean Opinion Score (MOS) 4.2.
2,203
1909.11646
title_snapshot
[ -0.023439768701791763, -0.020167212933301926, -0.024182341992855072, 0.05614796653389931, 0.0030149435624480247, 0.044845208525657654, 0.0405925028026104, 0.02178375795483589, -0.002304760506376624, -0.05228148773312569, -0.03633642569184303, 0.03503952920436859, -0.05507069081068039, 0.00...
At Stability's Edge: How to Adjust Hyperparameters to Preserve Minima Selection in Asynchronous Training of Neural Networks?
https://openreview.net/forum?id=Bkeb7lHtvH
[ "Niv Giladi", "Mor Shpigel Nacson", "Elad Hoffer", "Daniel Soudry" ]
Spotlight
null
Background: Recent developments have made it possible to accelerate neural networks training significantly using large batch sizes and data parallelism. Training in an asynchronous fashion, where delay occurs, can make training even more scalable. However, asynchronous training has its pitfalls, mainly a degradation in...
[ "implicit bias", "stability", "neural networks", "generalization gap", "asynchronous SGD" ]
How to prevent stale gradients (in asynchronous SGD) from changing minima stability and degrade steady state generalization?
2,199
1909.12340
title_snapshot
[ -0.03262736648321152, -0.026632651686668396, -0.025875641033053398, 0.03319048136472702, 0.00602565286681056, 0.034313928335905075, 0.03597981482744217, -0.003968117758631706, -0.058696240186691284, -0.04065883159637451, 0.0011939124669879675, -0.007943284697830677, -0.04929543659090996, 0...