title
stringlengths
15
153
paper_url
stringlengths
41
45
authors
listlengths
1
21
type
stringclasses
3 values
primary_area
stringclasses
0 values
abstract
large_stringlengths
310
2.43k
keywords
listlengths
1
32
TL;DR
large_stringlengths
23
250
submission_number
int64
1
11.7k
arxiv_id
stringlengths
10
10
arxiv_id_source
stringclasses
2 values
embedding
listlengths
768
768
Optimization-Based Algebraic Multigrid Coarsening Using Reinforcement Learning
https://openreview.net/forum?id=xXYjxli-2i
[ "Ali Taghibakhshi", "Scott MacLachlan", "Luke Olson", "Matthew West" ]
Poster
null
Large sparse linear systems of equations are ubiquitous in science and engineering, such as those arising from discretizations of partial differential equations. Algebraic multigrid (AMG) methods are one of the most common methods of solving such linear systems, with an extensive body of underlying mathematical theory....
[ "Algebraic Multigrid", "Reinforcement Learning", "Graph Partitioning" ]
Providing a reinforcement learning method utilizing graph neural networks for algebraic multigrid coarsening, outperforming existing algorithms.
7,640
2106.01854
title_snapshot
[ -0.05815253034234047, -0.016800224781036377, 0.032957665622234344, 0.033165041357278824, 0.02838747389614582, 0.024072987958788872, 0.006627056282013655, -0.019798411056399345, -0.03334624320268631, -0.05930153653025627, 0.02226252853870392, 0.00974844116717577, -0.07999870926141739, 0.039...
Deep inference of latent dynamics with spatio-temporal super-resolution using selective backpropagation through time
https://openreview.net/forum?id=9pt6F8w1Jgs
[ "Feng Zhu", "Andrew R Sedler", "Harrison A Grier", "Nauman Ahad", "Mark A. Davenport", "Matthew Kaufman", "Andrea Giovannucci", "Chethan Pandarinath" ]
Poster
null
Modern neural interfaces allow access to the activity of up to a million neurons within brain circuits. However, bandwidth limits often create a trade-off between greater spatial sampling (more channels or pixels) and the temporal frequency of sampling. Here we demonstrate that it is possible to obtain spatio-temporal ...
[ "Computational neuroscience", "Systems neuroscience", "Neural population dynamics", "intermittent sampling", "Electrophysiology", "Calcium imaging", "Brain-computer interfaces", "Neuroscience", "Neuroprosthetics", "Neural coding", "Motor control", "Sequential autoencoders" ]
We develop a novel learning rule for backpropagating loss in neuroscientific time series data with intermittent sampling, enabling sequential autoencoders to increase spatiotemporal resolution in electrophysiology and calcium imaging datasets.
11,666
2111.00070
title_snapshot
[ -0.029901579022407532, -0.014953955076634884, 0.004664988722652197, 0.019893571734428406, 0.042666833847761154, 0.022424062713980675, 0.04348713532090187, 0.004437449853867292, -0.03885837271809578, -0.046318233013153076, 0.010619421489536762, -0.0000788250908954069, -0.030512576922774315, ...
Skipping the Frame-Level: Event-Based Piano Transcription With Neural Semi-CRFs
https://openreview.net/forum?id=DGA8XbJ8FVd
[ "Yujia Yan", "Frank Cwitkowitz", "Zhiyao Duan" ]
Poster
null
Piano transcription systems are typically optimized to estimate pitch activity at each frame of audio. They are often followed by carefully designed heuristics and post-processing algorithms to estimate note events from the frame-level predictions. Recent methods have also framed piano transcription as a multi-task lea...
[ "Music", "Audio", "Piano Transcrition", "Music Transcription", "Semi-Markov", "CRFs", "Sound Event Detection", "Music Information Retrieval" ]
We propose a novel piano transcription system that is simple, fast, and well-performing.
5,577
null
null
[ -0.0071775903925299644, -0.002185099758207798, -0.004759327042847872, 0.020165584981441498, 0.04325293004512787, 0.009234301745891571, 0.028095025569200516, 0.010471214540302753, -0.04530242457985878, -0.054397713392972946, -0.01102480199187994, 0.02224799431860447, -0.03778969496488571, 0...
Active Learning of Convex Halfspaces on Graphs
https://openreview.net/forum?id=O-fOgeI_D-B
[ "Maximilian Thiessen", "Thomas Gärtner" ]
Poster
null
We systematically study the query complexity of learning geodesically convex halfspaces on graphs. Geodesic convexity is a natural generalisation of Euclidean convexity and allows the definition of convex sets and halfspaces on graphs. We prove an upper bound on the query complexity linear in the treewidth and the mini...
[ "active learning", "learning theory", "semi-supervised learning", "transduction", "vertex classification", "graphs", "convexity theory", "geodesic convexity", "shortest paths", "halfspaces", "query complexity" ]
We systematically study the query complexity of learning geodesically convex halfspaces on a vertex-labelled graph.
9,657
null
null
[ -0.019333655014634132, -0.00702654616907239, 0.012278370559215546, 0.025773804634809494, 0.004383044317364693, 0.005591831170022488, 0.016778524965047836, 0.015986386686563492, -0.02684909850358963, -0.04323064535856247, -0.02638588659465313, -0.010291771963238716, -0.07552620023488998, 0....
Nested Variational Inference
https://openreview.net/forum?id=kBrHzFtwdp
[ "Heiko Zimmermann", "Hao Wu", "Babak Esmaeili", "Jan-Willem van de Meent" ]
Poster
null
We develop nested variational inference (NVI), a family of methods that learn proposals for nested importance samplers by minimizing an forward or reverse KL divergence at each level of nesting. NVI is applicable to many commonly-used importance sampling strategies and provides a mechanism for learning intermediate den...
[ "Variational Inference", "Importance Sampling", "Monte Carlo methods" ]
null
9,656
2106.11302
title_snapshot
[ 0.0062095290049910545, 0.007749610580503941, 0.009917126968502998, 0.04968733340501785, 0.03509967774152756, 0.06785953044891357, 0.0350043848156929, -0.0032388465479016304, -0.025131355971097946, -0.059136372059583664, 0.02140752226114273, -0.0010632036719471216, -0.06816934049129486, 0.0...
Row-clustering of a Point Process-valued Matrix
https://openreview.net/forum?id=YXy_2b5wufe
[ "Lihao Yin", "Ganggang Xu", "Huiyan Sang", "Yongtao Guan" ]
Poster
null
Structured point process data harvested from various platforms poses new challenges to the machine learning community. To cluster repeatedly observed marked point processes, we propose a novel mixture model of multi-level marked point processes for identifying potential heterogeneity in the observed data. Specifically,...
[ "Expectation-Solution Algorithm", "Functional Principal Component Analysis", "Marked Point Processes", "Model-based Clustering", "Semiparametric Model" ]
We propose a mixture model of multi-level marked point processes for clustering repeatedly observed marked event sequences
11,369
2110.01207
title_snapshot
[ 0.0001055505927070044, -0.03336646407842636, 0.008674166165292263, 0.0077547044493258, 0.0321834459900856, 0.052496105432510376, 0.015623662620782852, 0.0001377825828967616, -0.02088133618235588, -0.0400078259408474, 0.007339754607528448, -0.02460338920354843, -0.0506255216896534, -0.01117...
On learning sparse vectors from mixture of responses
https://openreview.net/forum?id=6k0bAbb6m6
[ "Nikita Polyanskii" ]
Poster
null
In this paper, we address two learning problems. Suppose a family of $\ell$ unknown sparse vectors is fixed, where each vector has at most $k$ non-zero elements. In the first problem, we concentrate on robust learning the supports of all vectors from the family using a sequence of noisy responses. Each response to a q...
[ "sparse vectors", "mixture of binary linear classifiers", "1-bit compessed sensing", "query complexity", "noisy measurements" ]
null
11,264
null
null
[ 0.0054166847839951515, -0.009777014143764973, 0.015591283328831196, 0.013316408731043339, 0.033237628638744354, 0.0245237834751606, 0.011013010516762733, 0.005920823663473129, -0.03335139900445938, -0.047530122101306915, -0.004039970226585865, 0.007664846256375313, -0.06537048518657684, -0...
Contextual Recommendations and Low-Regret Cutting-Plane Algorithms
https://openreview.net/forum?id=45GfBQYtYlp
[ "Sreenivas Gollapudi", "Guru Guruganesh", "Kostas Kollias", "Pasin Manurangsi", "Renato Paes Leme", "Jon Schneider" ]
Poster
null
We consider the following variant of contextual linear bandits motivated by routing applications in navigational engines and recommendation systems. We wish to learn a hidden $d$-dimensional value $w^*$. Every round, we are presented with a subset $\mathcal{X}_t \subseteq \mathbb{R}^d$ of possible actions. If we choos...
[ "Online Learning", "Convex Geometry", "Separation Oracles", "Linear Bandits", "Contextual Bandits" ]
We consider variants of the linear contextual bandit problem where we only receive the best arm as feedback.
11,257
2106.04819
title_snapshot
[ 0.006289813667535782, -0.00027359373052604496, -0.011817608959972858, 0.03223099932074547, 0.03693832829594612, 0.02677099034190178, 0.020536620169878006, 0.018490344285964966, -0.009772997349500656, -0.06558340042829514, -0.024069972336292267, 0.0062891459092497826, -0.05008324980735779, ...
A Stochastic Newton Algorithm for Distributed Convex Optimization
https://openreview.net/forum?id=5BD4_awH4Fd
[ "Brian Bullins", "Kumar Kshitij Patel", "Ohad Shamir", "Nathan Srebro", "Blake Woodworth" ]
Poster
null
We propose and analyze a stochastic Newton algorithm for homogeneous distributed stochastic convex optimization, where each machine can calculate stochastic gradients of the same population objective, as well as stochastic Hessian-vector products (products of an independent unbiased estimator of the Hessian of the popu...
[ "Distributed Optimization", "Stochastic Optimization", "Federated Learning", "Newton's Method" ]
We propose and analyze a stochastic Newton algorithm for homogeneous distributed convex optimization based on efficiently solving quadratic objectives in parallel with only a single round of communication.
9,632
2110.02954
title_snapshot
[ -0.020830200985074043, -0.0017938094679266214, 0.005929398816078901, 0.039739515632390976, 0.015977591276168823, 0.05366450920701027, 0.04679068922996521, 0.02907431498169899, -0.03391202539205551, -0.05124108865857124, 0.003372112289071083, -0.018531454727053642, -0.04872636869549751, -0....
Efficient Learning of Discrete-Continuous Computation Graphs
https://openreview.net/forum?id=TLIHuw3gcQB
[ "David Friede", "Mathias Niepert" ]
Poster
null
Numerous models for supervised and reinforcement learning benefit from combinations of discrete and continuous model components. End-to-end learnable discrete-continuous models are compositional, tend to generalize better, and are more interpretable. A popular approach to building discrete-continuous computation graphs...
[ "Discrete-Continuous Learning", "Stochastic Computation Graphs", "Gumbel-softmax Trick" ]
We propose two new strategies to enable efficient learning of discrete-continuous computation graphs with multiple stochastic nodes.
10,964
2307.14193
title_snapshot
[ -0.02039969339966774, -0.01310370396822691, -0.007638563401997089, 0.06171854957938194, 0.025712179020047188, 0.021656213328242302, 0.01899847574532032, 0.018821194767951965, 0.0004753793473355472, -0.033992499113082886, -0.006644419860094786, -0.0007366795325651765, -0.07784101366996765, ...
Constrained Two-step Look-Ahead Bayesian Optimization
https://openreview.net/forum?id=oVEGzC7ieOB
[ "Duke Zhang", "Xiangyu Zhang", "Peter I. Frazier" ]
Poster
null
Recent advances in computationally efficient non-myopic Bayesian optimization offer improved query efficiency over traditional myopic methods like expected improvement, with only a modest increase in computational cost. These advances have been largely limited to unconstrained BO methods with only a few exceptions whic...
[ "Bayesian optimization", "Non-myopic", "Gaussian processes", "Derivative-free optimization" ]
This paper gives a new method for constrained Bayesian optimization, 2-OPT-C, that is more query efficient than past methods.
7,573
null
null
[ -0.03565923869609833, 0.020916825160384178, 0.0005576190305873752, 0.023902539163827896, 0.043662793934345245, 0.04479862377047539, 0.01951654441654682, -0.003721022978425026, -0.017141638323664665, -0.038116514682769775, 0.00021791634208057076, 0.028346968814730644, -0.04733765497803688, ...
Entropic Desired Dynamics for Intrinsic Control
https://openreview.net/forum?id=Juk1LKbFvd
[ "Steven Stenberg Hansen", "Guillaume Desjardins", "Kate Baumli", "David Warde-Farley", "Nicolas Heess", "Simon Osindero", "Volodymyr Mnih" ]
Poster
null
An agent might be said, informally, to have mastery of its environment when it has maximised the effective number of states it can reliably reach. In practice, this often means maximizing the number of latent codes that can be discriminated from future states under some short time horizon (e.g. \cite{eysenbach2018diver...
[ "intrinsic control", "skill discovery", "exploration", "unsupervised reinforcement learning" ]
Discriminating between globally consistent codes yields exploratory behavior and a map of the controllable environment without the requirement of supervisory signals or rewards.
10,906
null
null
[ -0.021622713655233383, -0.037359192967414856, -0.012209572829306126, 0.06704194098711014, 0.03742202743887901, 0.037899188697338104, 0.024061741307377815, 0.00848877802491188, -0.010611995123326778, -0.05946189537644386, -0.028675509616732597, -0.004994657356292009, -0.0506589412689209, -0...
Collaborating with Humans without Human Data
https://openreview.net/forum?id=79zWncwO2p
[ "DJ Strouse", "Kevin R. McKee", "Matthew Botvinick", "Edward Hughes", "Richard Everett" ]
Spotlight
null
Collaborating with humans requires rapidly adapting to their individual strengths, weaknesses, and preferences. Unfortunately, most standard multi-agent reinforcement learning techniques, such as self-play (SP) or population play (PP), produce agents that overfit to their training partners and do not generalize well to...
[ "multi-agent", "zero-shot coordination", "collaboration", "cooperation", "common-payoff", "human-ai interaction", "reinforcement learning", "deep reinforcement learning" ]
We train state-of-the-art agents for zero-shot coordination with humans without using human data in the training pipeline.
3,465
2110.08176
title_snapshot
[ -0.01968255825340748, -0.01927327923476696, -0.01135100144892931, 0.05762343481183052, 0.06186534836888313, -0.011471182107925415, 0.016505593433976173, 0.01331899594515562, -0.003975627478212118, -0.05579977110028267, -0.03782334923744202, 0.017264528200030327, -0.07419528812170029, -0.01...
On the Role of Optimization in Double Descent: A Least Squares Study
https://openreview.net/forum?id=32eyjxaRxp
[ "Ilja Kuzborskij", "Csaba Szepesvari", "Omar Rivasplata", "Amal Rannen-Triki", "Razvan Pascanu" ]
Poster
null
Empirically it has been observed that the performance of deep neural networks steadily improves with increased model size, contradicting the classical view on overfitting and generalization. Recently, the double descent phenomenon has been proposed to reconcile this observation with theory, suggesting that the test err...
[ "Double Descent", "Optimization Error", "Excess Risk", "Generalization", "Least Squares" ]
null
10,874
2107.12685
title_snapshot
[ -0.0434282049536705, -0.013662160374224186, 0.012143604457378387, 0.007652135565876961, 0.047829899936914444, 0.03578168898820877, 0.033953338861465454, -0.010255304165184498, -0.03881449997425079, -0.019298655912280083, -0.005791380535811186, 0.015113181434571743, -0.03765963762998581, 0....
Machine learning structure preserving brackets for forecasting irreversible processes
https://openreview.net/forum?id=ntAkYRaIfox
[ "Kookjin Lee", "Nathaniel Trask", "Panos Stinis" ]
Poster
null
Forecasting of time-series data requires imposition of inductive biases to obtain predictive extrapolation, and recent works have imposed Hamiltonian/Lagrangian form to preserve structure for systems with \emph{reversible} dynamics. In this work we present a novel parameterization of dissipative brackets from metriplec...
[ "structure preserving machine learning", "neural odes", "forecasting", "dissipative systems" ]
We design an architecture for learning dissipative ODEs preserving algebraic structure guaranteeing compatibility with 1st and 2nd law of thermodynamics.
10,873
2106.12619
title_snapshot
[ -0.00046021537855267525, -0.0013793624239042401, 0.019834481179714203, 0.026956288143992424, 0.03557422012090683, 0.0004070392460562289, 0.021088499575853348, -0.002621573396027088, -0.0335681214928627, -0.050224270671606064, 0.04662586376070976, 0.02196306362748146, -0.0636058822274208, -...
Latent Equilibrium: A unified learning theory for arbitrarily fast computation with arbitrarily slow neurons
https://openreview.net/forum?id=an8FSGbuCw
[ "Paul Haider", "Benjamin Ellenberger", "Laura Kriener", "Jakob Jordan", "Walter Senn", "Mihai A. Petrovici" ]
Oral
null
The response time of physical computational elements is finite, and neurons are no exception. In hierarchical models of cortical networks each layer thus introduces a response lag. This inherent property of physical dynamical systems results in delayed processing of stimuli and causes a timing mismatch between network ...
[ "dynamical systems", "cortical microcircuits", "deep learning", "prospective coding", "noise robustness", "synaptic plasticity" ]
A unified theory of neuronal dynamics and synaptic plasticity that solves the relaxation problem in networks with slow components.
10,777
2110.14549
title_snapshot
[ -0.0274093858897686, 0.0006479694275185466, -0.009804045781493187, 0.040800780057907104, 0.011404870077967644, 0.0202478040009737, 0.013921844772994518, 0.01697530597448349, -0.053416844457387924, -0.028852956369519234, 0.017103062942624092, -0.022968715056777, -0.055429037660360336, 0.006...
Pooling by Sliced-Wasserstein Embedding
https://openreview.net/forum?id=ienTaVMRtl
[ "Navid Naderializadeh", "Joseph F. Comer", "Reed W Andrews", "Heiko Hoffmann", "Soheil Kolouri" ]
Poster
null
Learning representations from sets has become increasingly important with many applications in point cloud processing, graph learning, image/video recognition, and object detection. We introduce a geometrically-interpretable and generic pooling mechanism for aggregating a set of features into a fixed-dimensional repres...
[ "Pooling", "feature aggregation", "permutation invariance", "optimal transport", "set learning", "representation learning" ]
We present a novel pooling method, called Pooling by Sliced-Wasserstein Embedding (PSWE), which leverages the (generalized) sliced-Wasserstein distances to map an input set of arbitrary cardinality to a fixed-size representation.
5,491
null
null
[ 0.03023984096944332, -0.042433544993400574, 0.019404038786888123, 0.06023985147476196, 0.019070424139499664, 0.04442921653389931, 0.005071246065199375, 0.01788260042667389, -0.003945931792259216, -0.050451528280973434, -0.04098327085375786, -0.009603109210729599, -0.07141170650720596, -0.0...
Asymptotics of the Bootstrap via Stability with Applications to Inference with Model Selection
https://openreview.net/forum?id=5JPPOluv-bp
[ "Morgane Austern", "Vasilis Syrgkanis" ]
Poster
null
One of the most commonly used methods for forming confidence intervals is the empirical bootstrap, which is especially expedient when the limiting distribution of the estimator is unknown. However, despite its ubiquitous role in machine learning, its theoretical properties are still not well understood. Recent developm...
[ "bootstrap", "asymptotics", "confidence intervals", "model selection" ]
null
10,735
null
null
[ -0.026015395298600197, -0.02465156652033329, -0.02753089927136898, 0.021088827401399612, 0.03697890788316727, 0.04786060377955437, 0.04164312034845352, 0.004197912290692329, -0.023815907537937164, -0.0454329214990139, 0.006269169971346855, 0.005183471832424402, -0.08838865160942078, -0.020...
Risk Minimization from Adaptively Collected Data: Guarantees for Supervised and Policy Learning
https://openreview.net/forum?id=6KcBgHQz3sJ
[ "Aurelien Bibaut", "Nathan Kallus", "Maria Dimakopoulou", "Antoine Chambaz", "Mark van der Laan" ]
Poster
null
Empirical risk minimization (ERM) is the workhorse of machine learning, whether for classification and regression or for off-policy policy learning, but its model-agnostic guarantees can fail when we use adaptively collected data, such as the result of running a contextual bandit algorithm. We study a generic importanc...
[ "empirical risk minimization", "maximal inequality", "metric entropy", "fast rates", "policy learning", "empirical process theory", "adaptively collected data" ]
We provide first-of-their-kind generalization guarantees and fast convergence rates for importance sampling weighted empirical risk minimization from adaptively collected data, such as data collected by a contextual bandit algorithm.
7,533
2106.01723
title_snapshot
[ -0.030167385935783386, -0.020010000094771385, 0.02109428495168686, 0.06603261083364487, 0.04905145615339279, 0.03588809445500374, 0.007992822676897049, -0.03159289062023163, -0.04396595433354378, -0.044228553771972656, -0.041301604360342026, 0.029531560838222504, -0.05879334360361099, -0.0...
Evaluation of Human-AI Teams for Learned and Rule-Based Agents in Hanabi
https://openreview.net/forum?id=bB-l0cnS3E
[ "Ho Chit Siu", "Jaime Daniel Pena", "Edenna Chen", "Yutai Zhou", "Victor Lopez", "Kyle Palko", "Kimberlee Chestnut Chang", "Ross Emerson Allen" ]
Poster
null
Deep reinforcement learning has generated superhuman AI in competitive games such as Go and StarCraft. Can similar learning techniques create a superior AI teammate for human-machine collaborative games? Will humans prefer AI teammates that improve objective team performance or those that improve subjective metrics of ...
[ "Human-AI Teaming", "Deep Learning", "Reinforcement Learning" ]
When working with AI agents in the cooperative card game Hanabi, humans strongly prefer a rules-based agent over a state-of-the-art learning-based agent, despite achieving similar scores.
1,385
2107.07630
title_snapshot
[ -0.03714987263083458, -0.03420070558786392, -0.023892126977443695, 0.02328658476471901, 0.034523896872997284, -0.015620413236320019, 0.028179924935102463, 0.01575562171638012, 0.00007311387889785692, -0.04694908484816551, -0.014935816638171673, 0.025265872478485107, -0.05728686600923538, -...
Learnable Fourier Features for Multi-dimensional Spatial Positional Encoding
https://openreview.net/forum?id=YXjhRGvqfFN
[ "Yang Li", "Si Si", "Gang Li", "Cho-Jui Hsieh", "Samy Bengio" ]
Poster
null
Attentional mechanisms are order-invariant. Positional encoding is a crucial component to allow attention-based deep model architectures such as Transformer to address sequences or images where the position of information matters. In this paper, we propose a novel positional encoding method based on learnable Fourier f...
[ "Positional encoding", "Transformer", "Fourier features", "multi-dimensional spatial tasks", "image classification", "image generation", "object detection", "widget captioning", "UI modeling" ]
The work presents a method for encoding multi-dimensional positions using learnable Fourier features for Transformer models.
3,431
2106.02795
title_snapshot
[ -0.017546910792589188, 0.006729618646204472, 0.024043995887041092, 0.011857437901198864, 0.02097409963607788, 0.0476151742041111, 0.011732963845133781, 0.022702939808368683, -0.040844399482011795, -0.044775426387786865, -0.022125961259007454, -0.023769034072756767, -0.02790072374045849, -0...
Charting and Navigating the Space of Solutions for Recurrent Neural Networks
https://openreview.net/forum?id=SQm_poGrlj
[ "Elia Turner", "Kabir Vinay Dabholkar", "Omri Barak" ]
Poster
null
In recent years Recurrent Neural Networks (RNNs) were successfully used to model the way neural activity drives task-related behavior in animals, operating under the implicit assumption that the obtained solutions are universal. Observations in both neuroscience and machine learning challenge this assumption. Animals c...
[ "RNN", "underspecification", "variability", "space of solutions", "neuroscience", "reverse engineering", "task optimized networks" ]
RNNs can produce a diverse set of qualitatively different solutions to the same computational tasks
10,593
2111.09356
title_snapshot
[ -0.04724884405732155, 0.00045998895075172186, -0.02947026491165161, 0.03327476978302002, 0.06171972677111626, 0.04474356770515442, 0.044070854783058167, 0.015883585438132286, -0.07486629486083984, -0.045733314007520676, -0.010823026299476624, -0.00207294593565166, -0.06485218554735184, -0....
Pipeline Combinators for Gradual AutoML
https://openreview.net/forum?id=wnAN2ZU7br
[ "Guillaume Baudart", "Martin Hirzel", "Kiran Kate", "Parikshit Ram", "Avraham Shinnar", "Jason Tsay" ]
Poster
null
Automated machine learning (AutoML) can make data scientists more productive. But if machine learning is totally automated, that leaves no room for data scientists to apply their intuition. Hence, data scientists often prefer not total but gradual automation, where they control certain choices and AutoML explores the...
[ "AutoML", "scikit-learn", "programming models", "functional programming" ]
Combinator-based syntax for ML pipelines; translation scheme to AutoML optimizers; and open-source library with user study.
9,564
null
null
[ 0.005347164813429117, -0.054545074701309204, -0.024415859952569008, 0.011836251243948936, 0.036274950951337814, 0.0280620064586401, 0.03131897374987602, 0.0005677502485923469, -0.013671624474227428, -0.01999034360051155, -0.021385908126831055, -0.0027095002587884665, -0.05359314754605293, ...
Sliced Mutual Information: A Scalable Measure of Statistical Dependence
https://openreview.net/forum?id=27qon5Ut4PSl
[ "Ziv Goldfeld", "Kristjan Greenewald" ]
Spotlight
null
Mutual information (MI) is a fundamental measure of statistical dependence, with a myriad of applications to information theory, statistics, and machine learning. While it possesses many desirable structural properties, the estimation of high-dimensional MI from samples suffers from the curse of dimensionality. Motiva...
[ "mutual information", "sliced mutual information", "curse of dimensionality", "feature extraction", "independence tests" ]
null
9,562
2110.05279
title_snapshot
[ -0.030530337244272232, -0.002387524116784334, 0.003772362833842635, 0.018019257113337517, 0.05893052741885185, 0.02873525582253933, 0.03132254257798195, -0.013890901580452919, -0.022957801818847656, -0.03219122067093849, -0.010233629494905472, 0.007070389110594988, -0.061946310102939606, -...
Forster Decomposition and Learning Halfspaces with Noise
https://openreview.net/forum?id=l4DQWgjbZg
[ "Ilias Diakonikolas", "Daniel Kane", "Christos Tzamos" ]
Spotlight
null
A Forster transform is an operation that turns a multivariate distribution into one with good anti-concentration properties. While a Forster transform does not always exist, we show that any distribution can be efficiently decomposed as a disjoint mixture of few distributions for which a Forster transform exists and ca...
[ "learning theory", "Forster transform", "halfspaces", "Massart noise" ]
First efficient learning algorithm for Massart halfspaces with sample complexity independent of the bit complexity of the examples.
10,354
2107.05582
title_snapshot
[ -0.007740294095128775, 0.011856278404593468, 0.00003842110527330078, 0.045603394508361816, 0.028913751244544983, 0.03391379117965698, -0.011314089410007, 0.0042314594611525536, -0.02698657289147377, -0.03240208700299263, -0.009415996260941029, 0.0049875034019351006, -0.07336528599262238, 0...
Shared Independent Component Analysis for Multi-Subject Neuroimaging
https://openreview.net/forum?id=24-ZY0UZVKD
[ "Hugo Richard", "Pierre Ablin", "Bertrand Thirion", "Alexandre Gramfort", "Aapo Hyvarinen" ]
Poster
null
We consider shared response modeling, a multi-view learning problem where one wants to identify common components from multiple datasets or views. We introduce Shared Independent Component Analysis (ShICA) that models each view as a linear transform of shared independent components contaminated by additive Gaussian noi...
[ "neuroimaging", "fMRI", "MEG", "shared response modeling", "component analysis", "independent component analysis", "multi-view learning" ]
A new probabilistic model and estimation algorithms for finding components shared by multiple subjects in neuroimaging data, combining theories of ICA, CCA, and shared response modelling
1,360
2110.13502
title_snapshot
[ 0.005375606473535299, 0.007785589899867773, -0.00795203447341919, 0.02988438867032528, 0.021887529641389847, 0.04982028901576996, 0.0305502749979496, 0.022008227184414864, -0.027424460276961327, -0.053496066480875015, 0.00007912015280453488, -0.008225981146097183, -0.05571012198925018, 0.0...
ParK: Sound and Efficient Kernel Ridge Regression by Feature Space Partitions
https://openreview.net/forum?id=reOnED4N_P-
[ "Luigi Carratino", "Stefano Vigogna", "Daniele Calandriello", "Lorenzo Rosasco" ]
Poster
null
We introduce ParK, a new large-scale solver for kernel ridge regression. Our approach combines partitioning with random projections and iterative optimization to reduce space and time complexity while provably maintaining the same statistical accuracy. In particular, constructing suitable partitions directly in the fea...
[ "Large-scale kernel methods", "kernel ridge regression", "random projections", "partitions" ]
We introduce a new large-scale algorithm to solve kernel ridge regression.
9,550
2106.12231
title_snapshot
[ -0.03127143159508705, -0.027368692681193352, 0.015859220176935196, 0.03825762867927551, 0.044490132480859756, 0.08731294423341751, -0.008701073937118053, -0.019886482506990433, -0.03597389906644821, -0.023710358887910843, -0.004149907268583775, 0.0070586418733000755, -0.06556205451488495, ...
Adversarial Feature Desensitization
https://openreview.net/forum?id=IIo3Ew4v5tk
[ "Pouya Bashivan", "Reza Bayat", "Adam Ibrahim", "Kartik Ahuja", "Mojtaba Faramarzi", "Touraj Laleh", "Blake Aaron Richards", "Irina Rish" ]
Poster
null
Neural networks are known to be vulnerable to adversarial attacks -- slight but carefully constructed perturbations of the inputs which can drastically impair the network's performance. Many defense methods have been proposed for improving robustness of deep networks by training them on adversarially perturbed inputs....
[ "adversarial robustness", "adversarial learning", "adversarial examples", "domain adaptation" ]
A defense method against adversarial attacks based on domain adaptation theory, achieving superior generalization towards a wide range of attack types and strengths.
5,444
2006.04621
title_snapshot
[ -0.02875414304435253, -0.002846750197932124, 0.0013901286292821169, 0.019220169633626938, 0.033573366701602936, 0.03642389550805092, 0.04092094302177429, -0.03109387494623661, -0.02820628508925438, -0.02255748026072979, -0.014636872336268425, -0.008701415732502937, -0.07003692537546158, 0....
Generalized Depthwise-Separable Convolutions for Adversarially Robust and Efficient Neural Networks
https://openreview.net/forum?id=jNq-i1zd0t9
[ "Hassan Dbouk", "Naresh Shanbhag" ]
Spotlight
null
Despite their tremendous successes, convolutional neural networks (CNNs) incur high computational/storage costs and are vulnerable to adversarial perturbations. Recent works on robust model compression address these challenges by combining model compression techniques with adversarial training. But these methods are un...
[ "adversarial robustness", "efficient inference", "generalized depthwise-seprarable", "convolutions", "neural networks" ]
We propose Generalized Depthwise-Separable convolutions as an efficient approximation of standard 2D convolutions that dramatically improve the throughput of an arbitrary pre-trained network on real-life hardware while preserving its robustness.
9,536
2110.14871
title_snapshot
[ -0.005861340556293726, -0.022536156699061394, -0.012327121570706367, 0.052905116230249405, 0.02378975972533226, 0.027823621407151222, 0.022831594571471214, -0.01276545599102974, -0.004614792764186859, -0.06028275564312935, 0.011261222884058952, -0.03419258072972298, -0.06690546125173569, 0...
Locally differentially private estimation of functionals of discrete distributions
https://openreview.net/forum?id=o1njPnYnttK
[ "Cristina Butucea", "Yann Issartel" ]
Poster
null
We study the problem of estimating non-linear functionals of discrete distributions in the context of local differential privacy. The initial data $x_1,\ldots,x_n \in[K]$ are supposed i.i.d. and distributed according to an unknown discrete distribution $p = (p_1,\ldots,p_K)$. Only $\alpha$-locally differentially priva...
[ "high-dimensional statistics", "local differential privacy", "power sum functional", "minimax estimation", "interactive privacy mechanism", "non-interactive privacy mechanism", "plug-in estimator", "threshold estimation" ]
null
5,436
2107.03940
title_judge
[ -0.013357306830585003, 0.017714565619826317, -0.0017222084570676088, 0.036916911602020264, 0.06205344945192337, 0.04584990441799164, 0.033096447587013245, -0.046933844685554504, -0.0009781882399693131, -0.024861836805939674, 0.03532926365733147, -0.02201571688055992, -0.06115807965397835, ...
Reusing Combinatorial Structure: Faster Iterative Projections over Submodular Base Polytopes
https://openreview.net/forum?id=961kvwqhR05
[ "Jai Moondra", "Hassan Mortagy", "Swati Gupta" ]
Poster
null
Optimization algorithms such as projected Newton's method, FISTA, mirror descent and its variants enjoy near-optimal regret bounds and convergence rates, but suffer from a computational bottleneck of computing ``projections" in potentially each iteration (e.g., $O(T^{1/2})$ regret of online mirror descent). On the oth...
[ "Convex optimization", "Conditional Gradients", "Bregman Projections", "Submodular base polytopes", "Online learning" ]
We bridge discrete and continuous optimization approaches to speed up iterative Bregman projections over submodular base polytopes.
9,529
2106.11943
title_snapshot
[ -0.025009892880916595, -0.027384260669350624, 0.022689174860715866, 0.04643896967172623, 0.03011133335530758, 0.05458850786089897, 0.0014324119547381997, -0.013055824674665928, -0.00903533585369587, -0.030502211302518845, -0.013758487068116665, -0.011443636380136013, -0.07334695756435394, ...
Reinforcement Learning in Linear MDPs: Constant Regret and Representation Selection
https://openreview.net/forum?id=t0r2M-ndcaJ
[ "Matteo Papini", "Andrea Tirinzoni", "Aldo Pacchiano", "Marcello Restelli", "Alessandro Lazaric", "Matteo Pirotta" ]
Poster
null
We study the role of the representation of state-action value functions in regret minimization in finite-horizon Markov Decision Processes (MDPs) with linear structure. We first derive a necessary condition on the representation, called universally spanning optimal features (UNISOFT), to achieve constant regret in any ...
[ "reinforcement learning", "regret minimization", "exploration", "problem-dependent analysis", "representation selection" ]
null
7,481
2110.14798
title_snapshot
[ -0.04706919938325882, -0.009749266318976879, -0.02838420681655407, 0.04874834418296814, 0.06482469290494919, 0.054633729159832, 0.012947963550686836, 0.01473927590996027, -0.018712647259235382, -0.05828012526035309, -0.01542445458471775, -0.0014962865971028805, -0.08850804716348648, -0.011...
Residual2Vec: Debiasing graph embedding with random graphs
https://openreview.net/forum?id=Z9K7sds_-jC
[ "Sadamori Kojaku", "Jisung Yoon", "Isabel Constantino", "Yong-Yeol Ahn" ]
Poster
null
Graph embedding maps a graph into a convenient vector-space representation for graph analysis and machine learning applications. Many graph embedding methods hinge on a sampling of context nodes based on random walks. However, random walks can be a biased sampler due to the structural properties of graphs. Most notably...
[ "Negative Sampling", "Debiasing", "Graph embedding", "Random walks" ]
We propose residual2vec, a general graph embedding method that can debias specified structural biases in graphs by using random graphs.
7,479
2110.07654
title_snapshot
[ 0.029171347618103027, -0.04279965162277222, 0.011291436851024628, 0.05026299133896828, 0.012892907485365868, 0.014780890196561813, 0.041801344603300095, 0.0021748137660324574, 0.013661830686032772, -0.07117295265197754, 0.03043479472398758, -0.0420091487467289, -0.07580383867025375, -0.004...
A Highly-Efficient Group Elastic Net Algorithm with an Application to Function-On-Scalar Regression
https://openreview.net/forum?id=KrAVI2AhNJh
[ "Tobia Boschi", "Matthew Reimherr", "Francesca Chiaromonte" ]
Poster
null
Feature Selection and Functional Data Analysis are two dynamic areas of research, with important applications in the analysis of large and complex data sets. Straddling these two areas, we propose a new highly efficient algorithm to perform Group Elastic Net with application to function-on-scalar feature selection, whe...
[ "Feature selection", "Group Elastic Net", "Function-on-Scalar regression", "Functional Principal Components", "Augmented Lagrangian", "GWAS" ]
We propose a new highly efficient dual augmented Lagrangian method to solve Group Elastic Net and Function-on-Scalar feature selection
10,159
null
null
[ -0.022541983053088188, -0.04648837819695473, 0.04908052459359169, -0.01213422603905201, 0.032795559614896774, 0.0489342100918293, 0.04354770854115486, -0.011823156848549843, -0.024469122290611267, -0.0547000952064991, 0.016247238963842392, -0.002155345631763339, -0.061446502804756165, 0.01...
Efficient Active Learning for Gaussian Process Classification by Error Reduction
https://openreview.net/forum?id=B9NOIHl2Z6K
[ "Guang Zhao", "Edward Dougherty", "Byung-Jun Yoon", "Francis Alexander", "Xiaoning Qian" ]
Poster
null
Active learning sequentially selects the best instance for labeling by optimizing an acquisition function to enhance data/label efficiency. The selection can be either from a discrete instance set (pool-based scenario) or a continuous instance space (query synthesis scenario). In this work, we study both active learnin...
[ "Bayesian Active Learning", "Gaussian Process Classification", "Expected Error Reduction", "Query Synthesis" ]
null
10,131
null
null
[ -0.0000448396967840381, -0.0012180025223642588, 0.0069552334025502205, 0.02628680318593979, 0.012986170127987862, 0.015530682168900967, -0.0035643167793750763, -0.013294383883476257, -0.007880828343331814, -0.0292324498295784, -0.0336247980594635, 0.01854505017399788, -0.06981506198644638, ...
Relational Self-Attention: What's Missing in Attention for Video Understanding
https://openreview.net/forum?id=DLKakJ2W-In
[ "Manjin Kim", "Heeseung Kwon", "Chunyu Wang", "Suha Kwak", "Minsu Cho" ]
Poster
null
Convolution has been arguably the most important feature transform for modern neural networks, leading to the advance of deep learning. Recent emergence of Transformer networks, which replace convolution layers with self-attention blocks, has revealed the limitation of stationary convolution kernels and opened the do...
[ "Action recognition", "Video understanding", "Motion analysis", "Self-attention" ]
We introduce relational self-attention (RSA) that effectively learns both visual appearance and motion for video understanding. Our model achieves state-of-the-art on several motion-centric action recognition benchmarks.
10,090
2111.01673
title_snapshot
[ 0.013762271031737328, -0.010381052270531654, 0.0037358361296355724, 0.03538713604211807, 0.019137723371386528, 0.04700161889195442, 0.036987170577049255, 0.018907107412815094, -0.014778469689190388, -0.0012952468823641539, -0.026741134002804756, -0.008214406669139862, -0.040210578590631485, ...
Play to Grade: Testing Coding Games as Classifying Markov Decision Process
https://openreview.net/forum?id=hjBEEXWNFH3
[ "Allen Nie", "Emma Brunskill", "Christopher J Piech" ]
Poster
null
Contemporary coding education often presents students with the task of developing programs that have user interaction and complex dynamic systems, such as mouse based games. While pedagogically compelling, there are no contemporary autonomous methods for providing feedback. Notably, interactive programs are impossible ...
[ "reinforcement learning", "computational education", "automated testing", "LSTM", "MDP equivalence" ]
Trained an RL agent to simultaneously learn to play a game and recognize bugs
9,512
2110.14615
title_snapshot
[ -0.03040374256670475, -0.01883704587817192, -0.05073860287666321, 0.04732406884431839, 0.0662289410829544, 0.024130141362547874, 0.02384136989712715, 0.011371877044439316, -0.011988796293735504, -0.05119375139474869, -0.018445534631609917, 0.0061118570156395435, -0.06604646146297455, -0.02...
Improving Coherence and Consistency in Neural Sequence Models with Dual-System, Neuro-Symbolic Reasoning
https://openreview.net/forum?id=uyKk_avJ-p4
[ "Maxwell Nye", "Michael Henry Tessler", "Joshua B. Tenenbaum", "Brenden M. Lake" ]
Poster
null
Human reasoning can be understood as an interplay between two systems: the intuitive and associative ("System 1") and the deliberative and logical ("System 2"). Neural sequence models---which have been increasingly successful at performing complex, structured tasks---exhibit the advantages and failure modes of System 1...
[ "System 2", "cognitive science", "dual-system", "coherence", "consistency", "dual-process" ]
Inspired by dual-system theories in cognitive science, we combine neural generation and symbolic checking to improve coherence and consistency in neural generations.
5,414
2107.02794
title_snapshot
[ -0.01976325735449791, 0.008172869682312012, -0.03271929547190666, 0.040686290711164474, 0.05363781005144119, 0.014115431345999241, 0.021351832896471024, 0.02937174029648304, -0.04145309701561928, -0.03106755018234253, -0.008574905805289745, 0.05151170492172241, -0.05336799472570419, -0.011...
Statistical Query Lower Bounds for List-Decodable Linear Regression
https://openreview.net/forum?id=7pU_P1IbePx
[ "Ilias Diakonikolas", "Daniel Kane", "Ankit Pensia", "Thanasis Pittas", "Alistair Stewart" ]
Spotlight
null
We study the problem of list-decodable linear regression, where an adversary can corrupt a majority of the examples. Specifically, we are given a set $T$ of labeled examples $(x, y) \in \mathbb{R}^d \times \mathbb{R}$ and a parameter $0< \alpha <1/2$ such that an $\alpha$-fraction of the points in $T$ are i.i.d. sample...
[ "learning theory", "high-dimensional robust statistics", "statistical query model", "list-decodable learning", "linear regression" ]
We prove a superpolynomial statistical query lower bound for the problem of learning the regression vector of a Gaussian linear model when outliers constitute the majority of the dataset.
10,006
2106.09689
title_snapshot
[ -0.007200584281235933, 0.00023908338334877044, -0.014742358587682247, 0.020389825105667114, 0.06012396514415741, 0.037541285157203674, 0.044611040502786636, -0.038558002561330795, -0.02235499583184719, -0.020906992256641388, -0.04036823660135269, 0.004385984502732754, -0.07492770254611969, ...
Learning to Iteratively Solve Routing Problems with Dual-Aspect Collaborative Transformer
https://openreview.net/forum?id=63pC59XOZLZ
[ "Yining Ma", "Jingwen Li", "Zhiguang Cao", "Wen Song", "Le Zhang", "Zhenghua Chen", "Jing Tang" ]
Poster
null
Recently, Transformer has become a prevailing deep architecture for solving vehicle routing problems (VRPs). However, it is less effective in learning improvement models for VRP because its positional encoding (PE) method is not suitable in representing VRP solutions. This paper presents a novel Dual-Aspect Collaborati...
[ "Transformer", "positional encoding", "learning to optimize", "vehicle routing problem", "combinatorial optimization" ]
We present a Dual-Aspect Collaborative Transformer to solve vehicle routing problems, which delivers superior performance.
9,497
2110.02544
title_snapshot
[ 0.0022949189879000187, -0.04633748531341553, -0.0088276956230402, 0.07006838172674179, 0.01353326253592968, 0.039242979139089584, 0.014249861240386963, 0.018590625375509262, -0.007270405068993568, -0.040707413107156754, -0.0057654716074466705, -0.025072811171412468, -0.04893389344215393, 0...
Gradient Starvation: A Learning Proclivity in Neural Networks
https://openreview.net/forum?id=h8flNv9x8v-
[ "Mohammad Pezeshki", "Sékou-Oumar Kaba", "Yoshua Bengio", "Aaron Courville", "Doina Precup", "Guillaume Lajoie" ]
Poster
null
We identify and formalize a fundamental gradient descent phenomenon resulting in a learning proclivity in over-parameterized neural networks. Gradient Starvation arises when cross-entropy loss is minimized by capturing only a subset of features relevant for the task, despite the presence of other predictive features th...
[ "generalization", "neural networks", "dynamics", "OOD" ]
This paper studies Gradient Starvation, a phenomenon that learning of one (spurious) feature may prevent the learning of other (generalizable) features..
3,352
2011.09468
title_snapshot
[ -0.047445159405469894, -0.002650463953614235, -0.009588952176272869, 0.036416102200746536, 0.047351423650979996, 0.021826745942234993, 0.0234568752348423, 0.003531919326633215, -0.05544149875640869, -0.014902101829648018, 0.011204240843653679, -0.0027480588760226965, -0.06481264531612396, ...
Extending Lagrangian and Hamiltonian Neural Networks with Differentiable Contact Models
https://openreview.net/forum?id=pZQrKCkbas
[ "Yaofeng Desmond Zhong", "Biswadip Dey", "Amit Chakraborty" ]
Poster
null
The incorporation of appropriate inductive bias plays a critical role in learning dynamics from data. A growing body of work has been exploring ways to enforce energy conservation in the learned dynamics by encoding Lagrangian or Hamiltonian dynamics into the neural network architecture. These existing approaches are b...
[ "Deep Model Learning", "Physics-based Priors", "Contact Models" ]
By introducing a differentiable contact model, this work extends the applicability of Lagrangian/Hamiltonian-inspired neural networks to enable the learning of hybrid dynamics.
9,830
2102.06794
title_snapshot
[ -0.02919778972864151, 0.025100212544202805, -0.017025265842676163, 0.03399871662259102, 0.04033711925148964, 0.02595270425081253, 0.003713083453476429, 0.002731827087700367, -0.06478138267993927, -0.0616333968937397, 0.02806861326098442, -0.01720876805484295, -0.05729524791240692, -0.00400...
Learning where to learn: Gradient sparsity in meta and continual learning
https://openreview.net/forum?id=8p46f7pYckL
[ "Johannes von Oswald", "Dominic Zhao", "Seijin Kobayashi", "Simon Schug", "Massimo Caccia", "Nicolas Zucchet", "Joao Sacramento" ]
Poster
null
Finding neural network weights that generalize well from small datasets is difficult. A promising approach is to learn a weight initialization such that a small number of weight changes results in low generalization error. We show that this form of meta-learning can be improved by letting the learning algorithm decide ...
[ "meta-learning", "continual learning", "few-shot learning", "deep learning", "sparsity", "sparse gradients" ]
null
5,392
2110.14402
title_snapshot
[ -0.0008188586798496544, -0.006472980137914419, 0.014696665108203888, 0.0368049219250679, 0.029161307960748672, 0.028519578278064728, 0.027317995205521584, 0.015520037151873112, -0.05939396098256111, -0.015796858817338943, -0.0015114482957869768, 0.01584004797041416, -0.056904327124357224, ...
Provably Strict Generalisation Benefit for Invariance in Kernel Methods
https://openreview.net/forum?id=HPG6TxihC1Y
[ "Bryn Elesedy" ]
Poster
null
It is a commonly held belief that enforcing invariance improves generalisation. Although this approach enjoys widespread popularity, it is only very recently that a rigorous theoretical demonstration of this benefit has been established. In this work we build on the function space perspective of Elesedy and Zaidi [8] t...
[ "generalization", "kernel methods", "invariance", "equivariance", "symmetry", "geometric deep learning", "statistical learning theory" ]
Strict generalisation benefit for invariance in kernel ridge regression
7,423
2106.02346
title_snapshot
[ -0.042011745274066925, -0.0112460320815444, 0.04460208863019943, 0.0072675105184316635, 0.023192834109067917, 0.042446162551641464, 0.05653563141822815, -0.031717125326395035, -0.01320622581988573, -0.02531735599040985, -0.04704051837325096, 0.015281119383871555, -0.07285718619823456, -0.0...
What training reveals about neural network complexity
https://openreview.net/forum?id=nTfnB6CvPJ
[ "Andreas Loukas", "Marinos Poiitis", "Stefanie Jegelka" ]
Poster
null
This work explores the Benevolent Training Hypothesis (BTH) which argues that the complexity of the function a deep neural network (NN) is learning can be deduced by its training dynamics. Our analysis provides evidence for BTH by relating the NN's Lipschitz constant at different regions of the input space with the beh...
[ "neural networks", "training behavior", "DNN complexity", "Lipschitz constants", "Benevolent Training Hypothesis" ]
One can deduce a neural network's complexity (i.e., its Lipschitz constant close and far from the training data) from its training dynamics.
5,373
2106.04186
title_snapshot
[ -0.03037576749920845, -0.011266982182860374, -0.003523726249113679, 0.04298456758260727, 0.03228777274489403, 0.03237852826714516, 0.04315151646733284, 0.00014996125537436455, -0.03406110405921936, -0.03401162102818489, 0.002107591601088643, 0.02629983425140381, -0.049997176975011826, 0.00...
Periodic Activation Functions Induce Stationarity
https://openreview.net/forum?id=gRwh5HkdaTm
[ "Lassi Meronen", "Martin Trapp", "Arno Solin" ]
Poster
null
Neural network models are known to reinforce hidden data biases, making them unreliable and difficult to interpret. We seek to build models that `know what they do not know' by introducing inductive biases in the function space. We show that periodic activation functions in Bayesian neural networks establish a connecti...
[ "Bayesian deep learning", "Gaussian process", "uncertainty quantification" ]
Periodic activation functions induce a connection between the prior on the network weights and stationary Gaussian process priors.
9,638
2110.13572
title_snapshot
[ -0.025441743433475494, 0.015508433803915977, -0.012255159206688404, 0.04230606555938721, 0.018580837175250053, 0.021055886521935463, 0.03917861729860306, 0.008446919731795788, -0.03556882217526436, -0.03797564283013344, 0.008622478693723679, 0.026984943076968193, -0.0478784553706646, -0.00...
Scaling Neural Tangent Kernels via Sketching and Random Features
https://openreview.net/forum?id=vIRFiA658rh
[ "Amir Zandieh", "Insu Han", "Haim Avron", "Neta Shoham", "Chaewon Kim", "Jinwoo Shin" ]
Poster
null
The Neural Tangent Kernel (NTK) characterizes the behavior of infinitely-wide neural networks trained under least squares loss by gradient descent. Recent works also report that NTK regression can outperform finitely-wide neural networks trained on small-scale datasets. However, the computational complexity of kernel m...
[ "Neural Tangent Kernel", "NTK", "Convolutional Neural Tangent Kernel", "CNTK", "Sketching", "Random Features", "Kernel Approximation", "Kernel Methods" ]
null
9,577
2106.07880
title_snapshot
[ -0.0030747184064239264, -0.044628024101257324, 0.00002044620669039432, 0.05489005893468857, 0.00529590155929327, 0.058105986565351486, 0.007523905951529741, 0.01906685344874859, -0.024163713678717613, -0.05354713276028633, -0.028925864025950432, -0.018148133531212807, -0.04671638086438179, ...
Neural Routing by Memory
https://openreview.net/forum?id=rHNF8Kq3u2P
[ "Kaipeng Zhang", "Zhenqiang Li", "Zhifeng Li", "Wei Liu", "Yoichi Sato" ]
Poster
null
Recent Convolutional Neural Networks (CNNs) have achieved significant success by stacking multiple convolutional blocks, named procedures in this paper, to extract semantic features. However, they use the same procedure sequence for all inputs, regardless of the intermediate features. This paper proffers a simple yet e...
[ "convolutional neural networks", "routing", "memory" ]
null
9,449
null
null
[ -0.008207978680729866, -0.02986852452158928, -0.03671234846115112, 0.027686508372426033, 0.033901963382959366, 0.03677515313029289, -0.0018964940682053566, 0.02177942357957363, -0.02034526877105236, -0.06319989264011383, 0.010662795975804329, -0.038490332663059235, -0.037550393491983414, 0...
Improved Regret Bounds for Tracking Experts with Memory
https://openreview.net/forum?id=x_sdq4ZYSOl
[ "James Robinson", "Mark Herbster" ]
Poster
null
We address the problem of sequential prediction with expert advice in a non-stationary environment with long-term memory guarantees in the sense of Bousquet and Warmuth [4]. We give a linear-time algorithm that improves on the best known regret bound [27]. This algorithm incorporates a relative entropy projection step....
[ "online learning", "prediction with expert advice" ]
We give an efficient projection-based algorithm for switching with memory in the experts setting and prove the best known regret bound.
7,361
2106.13021
title_snapshot
[ -0.02960207872092724, 0.0002773673622868955, 0.005494577810168266, 0.020016930997371674, 0.05987723544239998, 0.019612818956375122, 0.020950287580490112, 0.006578293163329363, -0.03780801221728325, -0.045985568314790726, -0.02810279093682766, 0.017702864482998848, -0.051159583032131195, -0...
Non-asymptotic convergence bounds for Wasserstein approximation using point clouds
https://openreview.net/forum?id=GoWFBE2-aLt
[ "Quentin Mérigot", "Filippo Santambrogio", "Clément SARRAZIN" ]
Poster
null
Several issues in machine learning and inverse problems require to generate discrete data, as if sampled from a model probability distribution. A common way to do so relies on the construction of a uniform probability distribution over a set of $N$ points which minimizes the Wasserstein distance to the model distributi...
[ "Optimal transport", "Wasserstein distance", "Non-convex optimization", "Polyak-Lojasiewicz inequality", "Lloyd's algorithm", "Point clouds" ]
We provide explicit error estimates for the output of the Lloyd-type algorithm used in uniform optimal quantization with respect to the sample size.
1,217
2106.07911
title_snapshot
[ -0.029942654073238373, -0.0010157125070691109, 0.006391321774572134, 0.043989963829517365, 0.028522806242108345, 0.03154711797833443, 0.01964755542576313, 0.006527171935886145, -0.03203194588422775, -0.05241040885448456, -0.01815488375723362, -0.03282080963253975, -0.0568537563085556, -0.0...
Bandit Quickest Changepoint Detection
https://openreview.net/forum?id=mxowVJFe8D5
[ "Aditya Gopalan", "Braghadeesh Lakshminarayanan", "Venkatesh Saligrama" ]
Poster
null
Many industrial and security applications employ a suite of sensors for detecting abrupt changes in temporal behavior patterns. These abrupt changes typically manifest locally, rendering only a small subset of sensors informative. Continuous monitoring of every sensor can be expensive due to resource constraints, and s...
[ "Bandit", "Changepoint detection", "CUSUM", "Quickest change detection", "Detection delay", "Adaptive sensing" ]
Approaches for adaptively monitoring a system to rapidly detect local change
9,260
2107.10492
title_snapshot
[ -0.017598435282707214, -0.03608512878417969, -0.02022363431751728, 0.04502280056476593, 0.055510014295578, 0.006375335622578859, 0.051080431789159775, 0.013869201764464378, -0.027450570836663246, -0.046428289264440536, -0.013214761391282082, 0.0056220549158751965, -0.05512970685958862, -0....
Early-stopped neural networks are consistent
https://openreview.net/forum?id=rMKTq-ca0qu
[ "Ziwei Ji", "Justin D. Li", "Matus Telgarsky" ]
Spotlight
null
This work studies the behavior of shallow ReLU networks trained with the logistic loss via gradient descent on binary classification data where the underlying data distribution is general, and the (optimal) Bayes risk is not necessarily zero. In this setting, it is shown that gradient descent with early stopping achie...
[ "Neural Networks", "Deep Networks", "calibration", "consistency", "nonseparable", "gradient descent" ]
For general classification problems, including those with noise, gradient descent with early stopping on shallow ReLU networks achieves the optimal risk amongst all measurable predictors
3,245
2106.05932
title_snapshot
[ -0.032837387174367905, -0.02916037291288376, -0.02938326634466648, 0.03943072631955147, 0.06071697920560837, 0.06786121428012848, 0.017856722697615623, 0.023202408105134964, -0.05135481804609299, -0.04052908346056938, 0.00479985773563385, 0.007917789742350578, -0.06199672818183899, -0.0171...
Adversarial Attacks on Graph Classifiers via Bayesian Optimisation
https://openreview.net/forum?id=5j_lH4OpZBl
[ "Xingchen Wan", "Henry Kenlay", "Binxin Ru", "Arno Blaas", "Michael Osborne", "Xiaowen Dong" ]
Poster
null
Graph neural networks, a popular class of models effective in a wide range of graph-based learning tasks, have been shown to be vulnerable to adversarial attacks. While the majority of the literature focuses on such vulnerability in node-level classification tasks, little effort has been dedicated to analysing adversar...
[ "adversarial attack", "Bayesian optimisation", "graph neural networks" ]
We present an effective black-box and query-efficient adversarial attack on graph classification models
3,243
2111.02842
title_judge
[ -0.0029205221217125654, -0.01741560362279415, 0.012346080504357815, 0.0453399196267128, 0.013189248740673065, 0.010545989498496056, 0.04820922389626503, -0.020213324576616287, -0.005541265942156315, -0.03537357226014137, -0.014136671088635921, -0.003690166864544153, -0.07165981084108353, 0...
Efficient Statistical Assessment of Neural Network Corruption Robustness
https://openreview.net/forum?id=IBHP61avv0R
[ "Karim TIT", "Teddy Furon", "Mathias ROUSSET" ]
Poster
null
We quantify the robustness of a trained network to input uncertainties with a stochastic simulation inspired by the field of Statistical Reliability Engineering. The robustness assessment is cast as a statistical hypothesis test: the network is deemed as locally robust if the estimated probability of failure is lower t...
[ "deep learning", "robustess", "reliability", "Monte Carlo" ]
Using a sequential Monte Carlo algorithm we assess efficiently the reliability of neural networks.
9,181
null
null
[ -0.017441492527723312, -0.002550793346017599, -0.034287337213754654, 0.050127990543842316, 0.05001097172498703, 0.02102951891720295, 0.00448421947658062, -0.03055918775498867, -0.03294478729367256, -0.02783733606338501, 0.022666169330477715, -0.0057145049795508385, -0.037396140396595, -0.0...
Scaling Vision with Sparse Mixture of Experts
https://openreview.net/forum?id=FrIDgjDOH1u
[ "Carlos Riquelme Ruiz", "Joan Puigcerver", "Basil Mustafa", "Maxim Neumann", "Rodolphe Jenatton", "André Susano Pinto", "Daniel Keysers", "Neil Houlsby" ]
Poster
null
Sparsely-gated Mixture of Experts networks (MoEs) have demonstrated excellent scalability in Natural Language Processing. In Computer Vision, however, almost all performant networks are "dense", that is, every input is processed by every parameter. We present a Vision MoE (V-MoE), a sparse version of the Vision Transfo...
[ "sparse", "model", "mixture", "experts", "sparsity", "vision", "conditional", "computation", "routing", "router", "adaptive", "compute" ]
We introduce a sparse mixture of experts architecture for vision classification and a new routing algorithm, leading to strong upstream and transfer results at the largest scale.
5,286
2106.05974
title_snapshot
[ 0.010274273343384266, -0.015577332116663456, 0.008392942138016224, 0.03650348260998726, 0.01994520053267479, 0.04838588833808899, 0.02634192630648613, 0.013635286130011082, -0.040778812021017075, -0.03077281080186367, 0.0073736668564379215, 0.022735025733709335, -0.05286922678351402, 0.016...
Combinatorial Optimization for Panoptic Segmentation: A Fully Differentiable Approach
https://openreview.net/forum?id=70eD741FHyI
[ "Ahmed Abbas", "Paul Swoboda" ]
Poster
null
We propose a fully differentiable architecture for simultaneous semantic and instance segmentation (a.k.a. panoptic segmentation) consisting of a convolutional neural network and an asymmetric multiway cut problem solver. The latter solves a combinatorial optimization problem that elegantly incorporates semantic and bo...
[ "Combinatorial optimization", "Backpropagation", "Panoptic segmentation", "end-to-end learning", "CNNs", "Gradient estimation", "Fully differentiable" ]
Show insights and benefits of training a pipeline containing neural network and large scale combinatorial optimization problem for panoptic segmentation
9,124
2106.03188
title_snapshot
[ 0.004231109283864498, -0.007724554277956486, 0.009824597276747227, 0.030516399070620537, 0.030006852000951767, 0.05802619829773903, -0.004929594695568085, 0.0035893567837774754, -0.031747959554195404, -0.03935268893837929, -0.04858030751347542, 0.007894081063568592, -0.04984616860747337, 0...
Leveraging Recursive Gumbel-Max Trick for Approximate Inference in Combinatorial Spaces
https://openreview.net/forum?id=UmSGhq5GuK
[ "Kirill Struminsky", "Artyom Gadetsky", "Denis Rakitin", "Danil Karpushkin", "Dmitry P. Vetrov" ]
Poster
null
Structured latent variables allow incorporating meaningful prior knowledge into deep learning models. However, learning with such variables remains challenging because of their discrete nature. Nowadays, the standard learning approach is to define a latent variable as a perturbed algorithm output and to use a different...
[ "structured discrete latent variable", "gumbel-max trick", "perturb-and-map", "gradient estimation", "score function" ]
Unbiased score-function-based gradient estimators for structured discrete latent variables
5,280
2110.15072
title_snapshot
[ 0.00207617343403399, -0.03640136495232582, -0.005247073248028755, 0.036346402019262314, 0.046502478420734406, 0.011713351123034954, 0.03453058376908302, -0.03108540177345276, -0.018550405278801918, -0.032798316329717636, -0.009174127131700516, -0.006494823377579451, -0.061606016010046005, ...
Open Rule Induction
https://openreview.net/forum?id=Tku-9lhJC5
[ "Wanyun Cui", "Xingran Chen" ]
Poster
null
Rules have a number of desirable properties. It is easy to understand, infer new knowledge, and communicate with other inference systems. One weakness of the previous rule induction systems is that they only find rules within a knowledge base (KB) and therefore cannot generalize to more open and complex real-world ru...
[ "Logic rules", "Open rules", "Language model" ]
We revisit the differences between KB-based rule induction and LM-based rule generation methods, and propose a noval method to induce open rules from language models.
9,092
2110.13577
title_snapshot
[ -0.03248811885714531, -0.018499134108424187, 0.0005734012229368091, 0.01693922095000744, 0.057304300367832184, -0.015071457251906395, 0.03470355272293091, -0.012699970975518227, -0.01571487821638584, 0.02119702473282814, -0.01108117587864399, 0.065375916659832, -0.07210669666528702, -0.004...
Class-agnostic Reconstruction of Dynamic Objects from Videos
https://openreview.net/forum?id=2Y8ikpBg6N8
[ "Zhongzheng Ren", "Xiaoming Zhao", "Alex Schwing" ]
Poster
null
We introduce REDO, a class-agnostic framework to REconstruct the Dynamic Objects from RGBD or calibrated videos. Compared to prior work, our problem setting is more realistic yet more challenging for three reasons: 1) due to occlusion or camera settings an object of interest may never be entirely visible, but we aim to...
[ "Dynamic reconstruction", "implicit network", "3D vision" ]
We introduce REDO, a class-agnostic framework to reconstruct the dynamic objects from videos
5,262
2112.02091
title_snapshot
[ 0.023950161412358284, -0.0220854002982378, 0.012078648433089256, 0.03960677608847618, 0.014860791154205799, 0.045231327414512634, 0.009715763852000237, 0.00954429991543293, -0.0615137442946434, -0.06029725819826126, -0.0030862886924296618, -0.01548425666987896, -0.06544884294271469, -0.007...
Multiclass Boosting and the Cost of Weak Learning
https://openreview.net/forum?id=fJWmx5i5lOv
[ "Nataly Brukhim", "Elad Hazan", "Shay Moran", "Indraneel Mukherjee", "Robert E. Schapire" ]
Poster
null
Boosting is an algorithmic approach which is based on the idea of combining weak and moderately inaccurate hypotheses to a strong and accurate one. In this work we study multiclass boosting with a possibly large number of classes or categories. Multiclass boosting can be formulated in various ways. He...
[ "Boosting" ]
null
8,893
null
null
[ -0.013481334783136845, -0.03994053229689598, -0.005711973179131746, 0.053797632455825806, 0.014701811596751213, 0.008902886882424355, 0.010248620994389057, -0.006701197475194931, 0.001180045772343874, -0.024900486692786217, -0.02814975380897522, 0.039019402116537094, -0.0752985030412674, -...
Fast Certified Robust Training with Short Warmup
https://openreview.net/forum?id=Qh-fwFsrEz
[ "Zhouxing Shi", "Yihan Wang", "Huan Zhang", "Jinfeng Yi", "Cho-Jui Hsieh" ]
Poster
null
Recently, bound propagation based certified robust training methods have been proposed for training neural networks with certifiable robustness guarantees. Despite that state-of-the-art (SOTA) methods including interval bound propagation (IBP) and CROWN-IBP have per-batch training complexity similar to standard neural ...
[ "Certified robustness", "Adversarial robustness" ]
We propose several improvements to Interval Bound Propagation (IBP) based certified robust training, which leads to faster training with short warmup schedules.
8,739
2103.17268
title_snapshot
[ -0.010450194589793682, -0.021369028836488724, -0.022053882479667664, 0.025238672271370888, 0.028926795348525047, 0.03021848574280739, 0.05474042147397995, -0.0349075011909008, -0.00876100454479456, -0.011155644431710243, 0.002158157993108034, -0.002425186801701784, -0.08008673787117004, -0...
Grounding Spatio-Temporal Language with Transformers
https://openreview.net/forum?id=2WnjXcymLtP
[ "Tristan Karch", "Laetitia Teodorescu", "Katja Hofmann", "Clément Moulin-Frier", "Pierre-Yves Oudeyer" ]
Poster
null
Language is an interface to the outside world. In order for embodied agents to use it, language must be grounded in other, sensorimotor modalities. While there is an extended literature studying how machines can learn grounded language, the topic of how to learn spatio-temporal linguistic concepts is still largely unch...
[ "Language Grounding", "Embodied Autonomous Agents" ]
We propose a novel Embodied Language Grounding task in order to study the generalization capabilities of Multi-modal Transformers on spatio-temporal language describing behavioral traces of an agent.
5,240
2106.08858
title_snapshot
[ -0.018187319859862328, 0.014950779266655445, -0.003310639411211014, 0.030809123069047928, 0.007093650288879871, 0.027948709204792976, 0.044687289744615555, 0.048524051904678345, -0.0216714795678854, -0.001423464622348547, -0.03419479727745056, 0.04352017119526863, -0.06222245469689369, -0....
Rebooting ACGAN: Auxiliary Classifier GANs with Stable Training
https://openreview.net/forum?id=r7UC-b67YkO
[ "Minguk Kang", "Woohyeon Joseph Shim", "Minsu Cho", "Jaesik Park" ]
Poster
null
Conditional Generative Adversarial Networks (cGAN) generate realistic images by incorporating class information into GAN. While one of the most popular cGANs is an auxiliary classifier GAN with softmax cross-entropy loss (ACGAN), it is widely known that training ACGAN is challenging as the number of classes in the data...
[ "Generative Adversarial Networks", "Conditional Image Generation", "Adversarial Learning", "Image Synthesis" ]
We propose a new conditional GAN: ReACGAN that can enjoy stable training and exploit data-to-data relationships during training.
5,238
2111.01118
title_snapshot
[ -0.008401961997151375, -0.0433691143989563, -0.028778590261936188, 0.032084811478853226, 0.0003839047276414931, 0.010443259961903095, -0.00478241965174675, -0.018857156857848167, -0.015901731327176094, -0.07029940187931061, -0.0327797494828701, -0.013738559558987617, -0.08553341031074524, ...
Evaluating State-of-the-Art Classification Models Against Bayes Optimality
https://openreview.net/forum?id=K9WlOVPEpnM
[ "Ryan Theisen", "Huan Wang", "Lav R. Varshney", "Caiming Xiong", "richard socher" ]
Poster
null
Evaluating the inherent difficulty of a given data-driven classification problem is important for establishing absolute benchmarks and evaluating progress in the field. To this end, a natural quantity to consider is the \emph{Bayes error}, which measures the optimal classification error theoretically achievable for a g...
[ "Bayes error", "normalizing flows" ]
null
8,696
2106.03357
title_snapshot
[ -0.009855019859969616, 0.0064841932617127895, -0.014706021174788475, 0.05621116980910301, 0.034416232258081436, 0.03595009446144104, 0.03396957740187645, -0.010617407038807869, -0.007046758662909269, -0.03253672644495964, -0.026511574164032936, -0.0016711431089788675, -0.0788683146238327, ...
Low-Rank Constraints for Fast Inference in Structured Models
https://openreview.net/forum?id=Mcldz4OJ6QB
[ "Justin T Chiu", "Yuntian Deng", "Alexander M Rush" ]
Poster
null
Structured distributions, i.e. distributions over combinatorial spaces, are commonly used to learn latent probabilistic representations from observed data. However, scaling these models is bottlenecked by the high computational and memory complexity with respect to the size of the latent representations. Common models ...
[ "structured prediction", "generative models", "graphical models", "inference" ]
null
8,624
2201.02715
title_snapshot
[ -0.021278873085975647, -0.005065026693046093, -0.02417706325650215, 0.03800085559487343, 0.028851546347141266, 0.012950916774570942, 0.016496755182743073, -0.01724724844098091, -0.026144729927182198, -0.010409969836473465, -0.005888765677809715, 0.014142881147563457, -0.07214789092540741, ...
Decoupling the Depth and Scope of Graph Neural Networks
https://openreview.net/forum?id=_IY3_4psXuf
[ "Hanqing Zeng", "Muhan Zhang", "Yinglong Xia", "Ajitesh Srivastava", "Andrey Malevich", "Rajgopal Kannan", "Viktor Prasanna", "Long Jin", "Ren Chen" ]
Poster
null
State-of-the-art Graph Neural Networks (GNNs) have limited scalability with respect to the graph and model sizes. On large graphs, increasing the model depth often means exponential expansion of the scope (i.e., receptive field). Beyond just a few layers, two fundamental challenges emerge: 1. degraded expressivity due...
[ "Graph Representation Learning", "Subgraph Mining", "Expressive Power", "Scalability" ]
Old GNN models + New data perspective = Surpassing 1-WL, Avoiding oversmoothing & Overcoming neighborhood explosion
8,538
2201.07858
title_snapshot
[ -0.01018222700804472, 0.009135418571531773, 0.005824210587888956, 0.03194123134016991, 0.04288101568818092, 0.033075667917728424, 0.023092009127140045, 0.011278467252850533, 0.010898047126829624, -0.04940436780452728, 0.03780054673552513, -0.019174447283148766, -0.09127532690763474, 0.0322...
Adversarial Regression with Doubly Non-negative Weighting Matrices
https://openreview.net/forum?id=npvEdo4Ftb1
[ "Tam Le", "Truyen Nguyen", "Makoto Yamada", "Jose Blanchet", "Viet Anh Nguyen" ]
Poster
null
Many machine learning tasks that involve predicting an output response can be solved by training a weighted regression model. Unfortunately, the predictive power of this type of models may severely deteriorate under low sample sizes or under covariate perturbations. Reweighting the training samples has aroused as an ef...
[ "adversarial regression", "kernel weighted regression", "covariate perturbation", "distributional robust", "doubly non-negative matrices", "Bures-Wasserstein", "Log-determinant divergence", "low-sample-size problem" ]
Adversarial Regression with Doubly Non-negative Weighting Matrices
8,492
2109.14875
title_snapshot
[ -0.025052165612578392, -0.028681235387921333, 0.014104180969297886, 0.0278424434363842, 0.023693125694990158, 0.0404353067278862, 0.01639668457210064, -0.03923598304390907, -0.011075860820710659, -0.0443270206451416, -0.03567410632967949, 0.02579873986542225, -0.07273858785629272, -0.00504...
Meta-Adaptive Nonlinear Control: Theory and Algorithms
https://openreview.net/forum?id=nm3sOq42Gmx
[ "Guanya Shi", "Kamyar Azizzadenesheli", "Michael O'Connell", "Soon-Jo Chung", "Yisong Yue" ]
Poster
null
We present an online multi-task learning approach for adaptive nonlinear control, which we call Online Meta-Adaptive Control (OMAC). The goal is to control a nonlinear system subject to adversarial disturbance and unknown \emph{environment-dependent} nonlinear dynamics, under the assumption that the environment-depende...
[ "Online Learning", "Representation Learning", "Adaptive Control", "Nonlinear Control", "Meta-learning" ]
By integrating adaptive control and meta-learning, we show the first non-asymptotic end-to-end guarantee for multi-task nonlinear control.
7,271
2106.06098
title_snapshot
[ -0.03635061904788017, 0.010015980340540409, 0.01424702350050211, 0.011860559694468975, 0.042488981038331985, 0.02537810057401657, 0.013770749792456627, 0.018602827563881874, -0.050923507660627365, -0.016590744256973267, -0.020919864997267723, -0.00037923327181488276, -0.07592132687568665, ...
Learning to Schedule Heuristics in Branch and Bound
https://openreview.net/forum?id=fEImgFxKU63
[ "Antonia Chmiela", "Elias Boutros Khalil", "Ambros Gleixner", "Andrea Lodi", "Sebastian Pokutta" ]
Poster
null
Primal heuristics play a crucial role in exact solvers for Mixed Integer Programming (MIP). While solvers are guaranteed to find optimal solutions given sufficient time, real-world applications typically require finding good solutions early on in the search to enable fast decision-making. While much of MIP research foc...
[ "integer programming", "learning to optimize", "data-driven algorithm design", "tree search", "algorithm configuration" ]
null
8,474
2103.10294
title_snapshot
[ -0.037507541477680206, 0.005093942862004042, -0.03483649343252182, 0.05866788700222969, 0.045991379767656326, 0.03949330747127533, 0.007203521206974983, -0.007890765555202961, -0.017681604251265526, -0.015683121979236603, -0.01075313426554203, -0.02366071380674839, -0.07820834964513779, -0...
Faster Non-asymptotic Convergence for Double Q-learning
https://openreview.net/forum?id=ZLwBKuGlY_
[ "Lin Zhao", "Huaqing Xiong", "Yingbin Liang" ]
Poster
null
Double Q-learning (Hasselt, 2010) has gained significant success in practice due to its effectiveness in overcoming the overestimation issue of Q-learning. However, the theoretical understanding of double Q-learning is rather limited. The only existing finite-time analysis was recently established in (Xiong et al. 2020...
[ "Reinforcement learning theory", "Markov decision process", "stochastic approximation" ]
This paper provides sharper finite-time analysis for double Q-learning with improved convergence rate over all major parameters.
1,122
null
null
[ -0.021042397245764732, -0.04181074723601341, -0.01048105489462614, 0.013668262399733067, 0.04063618928194046, 0.0259762741625309, -0.006190222222357988, 0.010458776727318764, -0.012001915834844112, -0.03403355926275253, 0.014411266893148422, 0.004694710951298475, -0.07595860958099365, -0.0...
On the Value of Infinite Gradients in Variational Autoencoder Models
https://openreview.net/forum?id=oumDUrf2dAB
[ "Bin Dai", "Li Kevin Wenliang", "David Wipf" ]
Spotlight
null
A number of recent studies of continuous variational autoencoder (VAE) models have noted, either directly or indirectly, the tendency of various parameter gradients to drift towards infinity during training. Because such gradients could potentially contribute to numerical instabilities, and are often framed as a probl...
[ "variational autoencoders", "sparse representations", "latent variable models" ]
We demonstrate that infinite gradients, although perhaps at times difficult to address in practical, can serve a useful role in pruning the latent space of autoencoder-based models.
1,121
null
null
[ -0.0029889375437051058, -0.0044538239017128944, -0.01183283794671297, 0.03708280250430107, 0.03238537535071373, 0.053611885756254196, 0.06403826922178268, 0.010672581382095814, -0.032230354845523834, -0.05186240375041962, -0.022142885252833366, -0.0003047526115551591, -0.08649613708257675, ...
Hit and Lead Discovery with Explorative RL and Fragment-based Molecule Generation
https://openreview.net/forum?id=onyYGbBJ2Mh
[ "Soojung Yang", "Doyeong Hwang", "Seul Lee", "Seongok Ryu", "Sung Ju Hwang" ]
Poster
null
Recently, utilizing reinforcement learning (RL) to generate molecules with desired properties has been highlighted as a promising strategy for drug design. Molecular docking program -- a physical simulation that estimates protein-small molecule binding affinity -- can be an ideal reward scoring function for RL, as it i...
[ "Drug discovery", "Molecule generation" ]
Our model generates realistic, qualified molecules with high therapeutic potential.
8,424
2110.01219
title_snapshot
[ -0.01456635445356369, -0.0005575708928517997, -0.008206910453736782, 0.030738692730665207, 0.0631733313202858, -0.03343866392970085, 0.016307149082422256, -0.01987604983150959, -0.013757606036961079, -0.035417161881923676, 0.005413537845015526, 0.03158774971961975, -0.06355860829353333, -0...
STEM: A Stochastic Two-Sided Momentum Algorithm Achieving Near-Optimal Sample and Communication Complexities for Federated Learning
https://openreview.net/forum?id=J28lNO4p3ki
[ "Prashant Khanduri", "PRANAY SHARMA", "Haibo Yang", "Mingyi Hong", "Jia Liu", "Ketan Rajawat", "Pramod Varshney" ]
Poster
null
Federated Learning (FL) refers to the paradigm where multiple worker nodes (WNs) build a joint model by using local data. Despite extensive research, for a generic non-convex FL problem, it is not clear, how to choose the WNs' and the server's update directions, the minibatch sizes, and the local update frequency, so t...
[ "Distributed Non-Convex Optimization", "Minibatch SGD", "Local SGD" ]
null
8,419
2106.10435
title_snapshot
[ -0.011150790378451347, -0.06398534029722214, 0.011639328673481941, 0.049296148121356964, 0.04001469165086746, 0.028435125946998596, -0.005760143045336008, -0.013202461414039135, -0.004112845752388239, -0.05504484102129936, -0.0007577790529467165, -0.034188978374004364, -0.022688301280140877,...
Dimensionality Reduction for Wasserstein Barycenter
https://openreview.net/forum?id=d4Lo6PhbKA
[ "Zachary Izzo", "Sandeep Silwal", "Samson Zhou" ]
Poster
null
The Wasserstein barycenter is a geometric construct which captures the notion of centrality among probability distributions, and which has found many applications in machine learning. However, most algorithms for finding even an approximate barycenter suffer an exponential dependence on the dimension $d$ of the underly...
[ "wasserstein barycenter", "dimensionality reduction" ]
We study randomized dimensionality reduction for the Wasserstein barycenter problem.
7,261
2110.08991
title_snapshot
[ -0.04813956469297409, -0.02575545944273472, 0.0211256705224514, 0.0251394584774971, 0.036560527980327606, 0.04507165774703026, 0.010392691008746624, 0.005978799425065517, -0.009525571949779987, -0.05897168815135956, -0.013078023679554462, -0.03619196638464928, -0.038688089698553085, 0.0073...
Sequential Causal Imitation Learning with Unobserved Confounders
https://openreview.net/forum?id=Kvb0482Ysaf
[ "Daniel Kumor", "Junzhe Zhang", "Elias Bareinboim" ]
Oral
null
"Monkey see monkey do" is an age-old adage, referring to naive imitation without a deep understanding of a system's underlying mechanics. Indeed, if a demonstrator has access to information unavailable to the imitator (monkey), such as a different set of sensors, then no matter how perfectly the imitator models its per...
[ "causality", "reinforcement learning", "imitation" ]
We provide a complete graphical condition for determining feasibility of sequential imitation in the presence of latent confounding
5,209
2208.06276
title_snapshot
[ -0.01725403033196926, -0.015412776730954647, -0.04973375424742699, 0.03278091549873352, 0.0586690679192543, 0.03941120207309723, 0.04996134713292122, 0.051704972982406616, -0.04722852259874344, -0.026918869465589523, 0.001854405738413334, 0.023717129603028297, -0.054813966155052185, -0.004...
Efficient Training of Visual Transformers with Small Datasets
https://openreview.net/forum?id=SCN8UaetXx
[ "Yahui Liu", "Enver Sangineto", "Wei Bi", "Nicu Sebe", "Bruno Lepri", "Marco De Nadai" ]
Poster
null
Visual Transformers (VTs) are emerging as an architectural paradigm alternative to Convolutional networks (CNNs). Differently from CNNs, VTs can capture global relations between image elements and they potentially have a larger representation capacity. However, the lack of the typical convolutional inductive bias makes...
[ "Vision Transformers", "Transformers", "Computer vision", "Self-supervision" ]
null
8,236
2106.03746
title_snapshot
[ 0.03870976343750954, -0.03176877647638321, 0.010587131604552269, 0.057450249791145325, 0.008774250745773315, 0.03232460096478462, 0.016567381098866463, -0.0014241812750697136, -0.023685799911618233, -0.03171573951840401, -0.05088041350245476, 0.022243477404117584, -0.06546828150749207, -0....
DynamicViT: Efficient Vision Transformers with Dynamic Token Sparsification
https://openreview.net/forum?id=jB0Nlbwlybm
[ "Yongming Rao", "Wenliang Zhao", "Benlin Liu", "Jiwen Lu", "Jie Zhou", "Cho-Jui Hsieh" ]
Poster
null
Attention is sparse in vision transformers. We observe the final prediction in vision transformers is only based on a subset of most informative tokens, which is sufficient for accurate image recognition. Based on this observation, we propose a dynamic token sparsification framework to prune redundant tokens progressiv...
[ "vision transformer", "deep model acceleration" ]
We propose a dynamic token sparsification framework to prune redundant tokens progressively and dynamically for vision transformer acceleration.
3,140
2106.02034
title_snapshot
[ 0.02087399549782276, -0.0249475110322237, 0.020695412531495094, 0.03327289596199989, 0.013757891952991486, 0.06111716851592064, -0.004702883772552013, 0.022207137197256088, -0.05236437916755676, -0.026180291548371315, -0.008244884200394154, 0.007306545507162809, -0.052764542400836945, 0.00...
On Learning Domain-Invariant Representations for Transfer Learning with Multiple Sources
https://openreview.net/forum?id=LkNBNOut0oD
[ "Trung Quoc Phung", "Trung Le", "Long Tung Vuong", "Toan Tran", "Anh Tuan Tran", "Hung Bui", "Dinh Phung" ]
Poster
null
Domain adaptation (DA) benefits from the rigorous theoretical works that study its insightful characteristics and various aspects, e.g., learning domain-invariant representations and its trade-off. However, it seems not the case for the multiple source DA and domain generalization (DG) settings which are remarkably mor...
[ "Domain generalization", "domain adaptation", "transfer learning", "multiple-source domain adaptation", "machine learning", "deep learning" ]
Develop target loss bounds in multi-domain transfer learning setting, which motivates 2 different domain invariant representations.
8,122
2111.13822
title_snapshot
[ -0.03812325745820999, -0.0015460094437003136, -0.0006783040589652956, 0.03380624204874039, 0.05043356865644455, 0.022308388724923134, 0.01767454668879509, -0.01895783469080925, -0.0006213962333276868, -0.014184737578034401, -0.023254765197634697, 0.03617775812745094, -0.0869361087679863, 0...
SIMILAR: Submodular Information Measures Based Active Learning In Realistic Scenarios
https://openreview.net/forum?id=VGDFaLNFFk
[ "Suraj Nandkishor Kothawade", "Nathan Alexander Beck", "Krishnateja Killamsetty", "Rishabh K Iyer" ]
Poster
null
Active learning has proven to be useful for minimizing labeling costs by selecting the most informative samples. However, existing active learning methods do not work well in realistic scenarios such as imbalance or rare classes,out-of-distribution data in the unlabeled set, and redundancy. In this w...
[ "Submodularity", "Active Learning", "Information Measures", "Realistic", "Redundancy", "Rare Classes", "Out of distribution", "Robust" ]
A unified active learning framework for active learning in realistic scenarios.
8,087
2107.00717
title_snapshot
[ -0.0029012816958129406, -0.06269065290689468, 0.008589228615164757, 0.0395219586789608, 0.0372064933180809, 0.016991345211863518, -0.00806401390582323, -0.014053969644010067, -0.0195342805236578, -0.02040371485054493, -0.004398263990879059, 0.0033111078664660454, -0.07628047466278076, -0.0...
A nonparametric method for gradual change problems with statistical guarantees
https://openreview.net/forum?id=zwkj1_pxFM
[ "Lizhen Nie", "Dan L Nicolae" ]
Poster
null
We consider the detection and localization of gradual changes in the distribution of a sequence of time-ordered observations. Existing literature focuses mostly on the simpler abrupt setting which assumes a discontinuity jump in distribution, and is unrealistic for some applied settings. We propose a general method for...
[ "Change point detection", "gradual change", "nonparametric" ]
We propose a method for detecting and localizing gradual changes in sequence data, which is applicable to any data generating model, any data type, requires no prior knowledge, but still comes with theoretical guarantees.
5,161
null
null
[ -0.03602169081568718, -0.04191002994775772, -0.006885135546326637, 0.034604594111442566, 0.04936906695365906, 0.05336737260222435, 0.031186386942863464, 0.024049313738942146, -0.021738549694418907, -0.04788282886147499, -0.017422854900360107, 0.003030354855582118, -0.044063765555620193, -0...
Grammar-Based Grounded Lexicon Learning
https://openreview.net/forum?id=VJQMp5xu24
[ "Jiayuan Mao", "Freda Shi", "Jiajun Wu", "Roger P. Levy", "Joshua B. Tenenbaum" ]
Poster
null
We present Grammar-Based Grounded Language Learning (G2L2), a lexicalist approach toward learning a compositional and grounded meaning representation of language from grounded data, such as paired images and texts. At the core of G2L2 is a collection of lexicon entries, which map each word to a tuple of a syntactic typ...
[ "Grounded Language Learning", "Combinatory Categorial Grammar", "Compositional Generalization" ]
We present a lexicalist approach toward learning a compositional and grounded meaning representation of language from grounded data.
7,986
2202.08806
title_snapshot
[ -0.018277112394571304, 0.016439832746982574, -0.018734922632575035, 0.026390893384814262, 0.020650053396821022, 0.043424177914857864, 0.03391040861606598, 0.0089616309851408, -0.023664189502596855, -0.00156241818331182, -0.025865543633699417, 0.04467613250017166, -0.05758710950613022, -0.0...
On the Power of Differentiable Learning versus PAC and SQ Learning
https://openreview.net/forum?id=WYrC0Aentah
[ "Emmanuel Abbe", "Pritish Kamath", "eran malach", "Colin Sandon", "Nathan Srebro" ]
Spotlight
null
We study the power of learning via mini-batch stochastic gradient descent (SGD) on the loss of a differentiable model or neural network, and ask what learning problems can be learnt using this paradigm. We show that SGD can always simulate learning with statistical queries (SQ), but its ability to go beyond that depend...
[ "Differentiable Learning", "PAC Learning", "Statistical Query Learning", "mini-batch SGD" ]
Identifies regimes of mini-batch size and gradient precision in gradient-based learning on differentiable models under which it becomes as powerful as sample based learning, or collapses to learning in the statistical query model.
7,971
2108.04190
title_snapshot
[ -0.05030618607997894, -0.03293921798467636, -0.016814403235912323, 0.043147750198841095, 0.0456993542611599, 0.03425240516662598, 0.02679997682571411, -0.0036206322256475687, -0.027732526883482933, -0.02342464216053486, 0.0007286064210347831, -0.01048591174185276, -0.039262086153030396, 0....
User-Level Differentially Private Learning via Correlated Sampling
https://openreview.net/forum?id=aTv5-fIfGte
[ "Badih Ghazi", "Ravi Kumar", "Pasin Manurangsi" ]
Poster
null
Most works in learning with differential privacy (DP) have focused on the setting where each user has a single sample. In this work, we consider the setting where each user holds $m$ samples and the privacy protection is enforced at the level of each user's data. We show that, in this setting, we may learn with a much...
[ "User-level privacy", "Global stability", "Representation dimension", "Correlated sampling" ]
User-level private learning is possible with very few users provided each user has sufficiently many samples.
1,061
2110.11208
title_judge
[ 0.017274340614676476, 0.006736184004694223, -0.00023131523630581796, 0.06751368194818497, 0.045460257679224014, 0.00237265694886446, 0.04241697117686272, -0.04014814272522926, -0.013293570838868618, -0.02726706676185131, 0.01613297127187252, -0.028833521530032158, -0.0709715411067009, 0.00...
Dataset Distillation with Infinitely Wide Convolutional Networks
https://openreview.net/forum?id=hXWPpJedrVP
[ "Timothy Nguyen", "Roman Novak", "Lechao Xiao", "Jaehoon Lee" ]
Poster
null
The effectiveness of machine learning algorithms arises from being able to extract useful features from large amounts of data. As model and dataset sizes increase, dataset distillation methods that compress large datasets into significantly smaller yet highly performant ones will become valuable in terms of training ef...
[ "dataset distillation", "infinite-width limit", "convolutional neural networks", "kernel ridge-regression", "meta-learning", "feature learning" ]
We achieve state of the art dataset distillation results on a variety of datasets using large-scale distributed metalearning with neural kernels.
1,060
2107.13034
title_snapshot
[ 0.02898353338241577, -0.04620436578989029, 0.005475948099046946, 0.0932840034365654, 0.051744453608989716, 0.004911663942039013, 0.01910666562616825, 0.0010313858510926366, -0.0001375597930746153, -0.03712617605924606, -0.02690732851624489, -0.02021767385303974, -0.07118531316518784, 0.015...
Evolution Gym: A Large-Scale Benchmark for Evolving Soft Robots
https://openreview.net/forum?id=eoTy4ihL0W
[ "Jagdeep Singh Bhatia", "Holly Jackson", "Yunsheng Tian", "Jie Xu", "Wojciech Matusik" ]
Poster
null
Both the design and control of a robot play equally important roles in its task performance. However, while optimal control is well studied in the machine learning and robotics community, less attention is placed on finding the optimal robot design. This is mainly because co-optimizing design and control in robotics is...
[ "Co-design", "Benchmark", "Soft Robotics", "Evolutionary Algorithms", "Reinforcement Learning", "Machine Learning" ]
We built Evolution Gym, the first large-scale benchmark for co-optimizing the design and control of soft robots.
7,203
2201.09863
title_snapshot
[ -0.021533474326133728, -0.018531590700149536, -0.023965006694197655, 0.030122073367238045, 0.0441458560526371, 0.038654159754514694, 0.01940768212080002, 0.008840414695441723, -0.029804138466715813, -0.04636912792921066, -0.00806556735187769, -0.010914661921560764, -0.06633155047893524, -0...
CoAtNet: Marrying Convolution and Attention for All Data Sizes
https://openreview.net/forum?id=dUk5Foj5CLf
[ "Zihang Dai", "Hanxiao Liu", "Quoc V Le", "Mingxing Tan" ]
Poster
null
Transformers have attracted increasing interests in computer vision, but they still fall behind state-of-the-art convolutional networks. In this work, we show that while Transformers tend to have larger model capacity, their generalization can be worse than convolutional networks due to the lack of the right inductive ...
[ "Hybrid", "Transformer", "Image Recognition" ]
Properly combining Convolution and Attention is surprisingly good.
7,915
2106.04803
title_snapshot
[ 0.026420610025525093, -0.0465204231441021, 0.020057372748851776, 0.01481658685952425, 0.02197221852838993, 0.032067254185676575, -0.0029032635502517223, 0.013386796228587627, -0.02554507926106453, -0.053233612328767776, -0.010012110695242882, -0.006022240500897169, -0.058055441826581955, -...
Learning to Generate Visual Questions with Noisy Supervision
https://openreview.net/forum?id=gEGPcdiXbky
[ "Kai Shen", "Lingfei Wu", "Siliang Tang", "Yueting Zhuang", "zhen he", "Zhuoye Ding", "Yun Xiao", "Bo Long" ]
Poster
null
The task of visual question generation (VQG) aims to generate human-like neural questions from an image and potentially other side information (e.g., answer type or the answer itself). Existing works often suffer from the severe one image to many questions mapping problem, which generates uninformative and non-referent...
[ "Noisy Supervision", "Generative Adversarial Networks", "Visual Question Generation" ]
Learning to Generate Visual Questions with Noisy Supervision
3,096
null
null
[ 0.02496732771396637, -0.03360758721828461, 0.0020712013356387615, 0.06285892426967621, 0.013027047738432884, 0.023584455251693726, -0.002974285976961255, -0.00613201642408967, -0.04142938554286957, -0.015473860315978527, -0.060510069131851196, 0.016714805737137794, -0.06555286049842834, -0...
Two Sides of Meta-Learning Evaluation: In vs. Out of Distribution
https://openreview.net/forum?id=X4JfcKvvDOw
[ "Amrith Setlur", "Oscar Li", "Virginia Smith" ]
Poster
null
We categorize meta-learning evaluation into two settings: $\textit{in-distribution}$ [ID], in which the train and test tasks are sampled $\textit{iid}$ from the same underlying task distribution, and $\textit{out-of-distribution}$ [OOD], in which they are not. While most meta-learning theory and some FSL applications f...
[ "meta-learning", "meta-learning evaluation", "few-shot learning", "out of distribution" ]
null
7,187
2102.11503
title_snapshot
[ -0.014016697183251381, -0.02384660579264164, -0.02340994030237198, 0.0336967408657074, 0.03978995233774185, -0.00905916653573513, 0.03697655349969864, 0.013388891704380512, -0.022338060662150383, -0.003271134803071618, 0.0025782075244933367, 0.02566170133650303, -0.07857662439346313, -0.03...
Increasing Liquid State Machine Performance with Edge-of-Chaos Dynamics Organized by Astrocyte-modulated Plasticity
https://openreview.net/forum?id=0isj8oxdQys
[ "Vladimir A Ivanov", "Konstantinos P. Michmizos" ]
Poster
null
The liquid state machine (LSM) combines low training complexity and biological plausibility, which has made it an attractive machine learning framework for edge and neuromorphic computing paradigms. Originally proposed as a model of brain computation, the LSM tunes its internal weights without backpropagation of gradie...
[ "liquid state machine", "astrocytes", "spiking neural networks", "neuromorphic computing", "neuron-astrocyte networks", "STDP", "self organized criticality" ]
We introduced astrocyte-modulated STDP to the liquid state machine (LSM) learning framework as a way to improve LSM accuracy performance and minimize dataset specific hand tuning.
7,777
2111.01760
title_snapshot
[ -0.040706757456064224, -0.029814576730132103, -0.02083796076476574, 0.017422573640942574, 0.05759020522236824, 0.018096299842000008, 0.019891273230314255, 0.0249785203486681, -0.04720789194107056, -0.03651633858680725, 0.010625012218952179, -0.03330472856760025, -0.057080283761024475, 0.01...
Bandit Learning with Delayed Impact of Actions
https://openreview.net/forum?id=4QrgRSAAroI
[ "Wei Tang", "Chien-Ju Ho", "Yang Liu" ]
Poster
null
We consider a stochastic multi-armed bandit (MAB) problem with delayed impact of actions. In our setting, actions taken in the past impact the arm rewards in the subsequent future. This delayed impact of actions is prevalent in the real world. For example, the capability to pay back a loan for people in a certain socia...
[ "action delay impact", "bandit learning" ]
Too Long; Didn't Read: We consider a stochastic multi-armed bandit (MAB) problem with delayed impact of actions. In our setting, actions taken in the past impact the arm rewards in the subsequent future
1,033
2002.10316
title_snapshot
[ -0.018810022622346878, -0.023774702101945877, -0.019801657646894455, 0.02807220071554184, 0.03651856258511543, 0.03896030783653259, 0.011382289230823517, 0.024758964776992798, -0.03849327191710472, -0.04149464890360832, -0.024233320727944374, 0.005379206035286188, -0.036420270800590515, -0...
Accelerating Quadratic Optimization with Reinforcement Learning
https://openreview.net/forum?id=xAFm5knU7Nc
[ "Jeffrey Ichnowski", "Paras Jain", "Bartolomeo Stellato", "Goran Banjac", "Michael Luo", "Francesco Borrelli", "Joseph E. Gonzalez", "Ion Stoica", "Ken Goldberg" ]
Poster
null
First-order methods for quadratic optimization such as OSQP are widely used for large-scale machine learning and embedded optimal control, where many related problems must be rapidly solved. These methods face two persistent challenges: manual hyperparameter tuning and convergence time to high-accuracy solutions. To ad...
[ "quadratic optimization", "convex optimization", "first-order methods", "reinforcement learning for optimization" ]
Using RL we train a policy to adapt internal parameters of a QP solver that allows the QP solver to converge faster
1,031
2107.10847
title_snapshot
[ -0.031357452273368835, -0.04712419584393501, 0.020071057602763176, 0.046460896730422974, 0.04333601891994476, 0.04911360144615173, -0.00011770171113312244, -0.005951335653662682, -0.03779887780547142, -0.028264835476875305, -0.03256969153881073, 0.020619414746761322, -0.08728791773319244, ...
Learning Disentangled Behavior Embeddings
https://openreview.net/forum?id=ThbM9_6DNU
[ "Changhao Shi", "Sivan Schwartz", "Shahar Levy", "Shay Achvat", "Maisan Abboud", "Amir Ghanayim", "Jackie Schiller", "Gal Mishne" ]
Spotlight
null
To understand the relationship between behavior and neural activity, experiments in neuroscience often include an animal performing a repeated behavior such as a motor task. Recent progress in computer vision and deep learning has shown great potential in the automated analysis of behavior by leveraging large and high-...
[ "Neuroscience", "Animal Behavior", "Behavioral Videos", "Generative Models" ]
null
7,734
null
null
[ 0.033575959503650665, 0.021025346592068672, -0.04992746189236641, 0.049695249646902084, 0.033230215311050415, 0.009634608402848244, 0.06374260038137436, 0.008103976026177406, -0.01616240292787552, -0.015422272495925426, -0.020456187427043915, -0.013107343576848507, -0.06683454662561417, 0....
An Uncertainty Principle is a Price of Privacy-Preserving Microdata
https://openreview.net/forum?id=PYJnWEn-uMn
[ "John M. Abowd", "Robert Ashmead", "Ryan Cumings-Menon", "Simson L. Garfinkel", "Daniel Kifer", "Philip Leclerc", "William Sexton", "Ashley E Simpson", "Christine Task", "Pavel Zhuravlev" ]
Poster
null
Privacy-protected microdata are often the desired output of a differentially private algorithm since microdata is familiar and convenient for downstream users. However, there is a statistical price for this kind of convenience. We show that an uncertainty principle governs the trade-off between accuracy for a populati...
[ "differential privacy", "synthetic data" ]
Privacy-preserving microdata may cause accuracy loss, specifically due to the microdata format
3,074
2110.13239
title_snapshot
[ -0.003577425377443433, 0.01361081749200821, -0.0009677375783212483, 0.08789423853158951, 0.0382048562169075, 0.03758068382740021, 0.04824035242199898, -0.04861010983586311, -0.018745096400380135, -0.01964799128472805, -0.022639846429228783, -0.012096557766199112, -0.06992565095424652, -0.0...
SADGA: Structure-Aware Dual Graph Aggregation Network for Text-to-SQL
https://openreview.net/forum?id=NJg6R1ATGpe
[ "Ruichu Cai", "Jinjie Yuan", "Boyan Xu", "Zhifeng Hao" ]
Poster
null
The Text-to-SQL task, aiming to translate the natural language of the questions into SQL queries, has drawn much attention recently. One of the most challenging problems of Text-to-SQL is how to generalize the trained model to the unseen database schemas, also known as the cross-domain Text-to-SQL task. The key lies i...
[ "Graph Aggregation Network", "Graph Neural Network", "Text-to-SQL" ]
null
5,110
2111.00653
title_snapshot
[ -0.016289781779050827, -0.016395607963204384, 0.016705051064491272, 0.05035974830389023, 0.042346179485321045, 0.0298552718013525, 0.030422400683164597, 0.008242909796535969, -0.010042546316981316, -0.036946531385183334, 0.0060807280242443085, 0.01132278237491846, -0.0804009661078453, -0.0...
Learning from Inside: Self-driven Siamese Sampling and Reasoning for Video Question Answering
https://openreview.net/forum?id=lDVeaQIScg
[ "Weijiang Yu", "Haoteng Zheng", "Mengfei Li", "Lei Ji", "Lijun Wu", "Nong Xiao", "Nan Duan" ]
Poster
null
Recent advances in the video question answering (i.e., VideoQA) task have achieved strong success by following the paradigm of fine-tuning each clip-text pair independently on the pretrained transformer-based model via supervised learning. Intuitively, multiple samples (i.e., clips) should be interdependent to capture ...
[ "Video Question Answering", "Multi-view Learning", "Data-effeciency Reasoning" ]
Learning from multimodal knowledge via self-supervised learning for Video Question Answering
7,610
null
null
[ 0.035623688250780106, -0.017224108800292015, -0.008879579603672028, 0.07169030606746674, 0.04036322608590126, 0.015425853431224823, 0.02284577488899231, 0.005438452120870352, -0.01251916028559208, -0.0013293451629579067, -0.023801712319254875, 0.028537670150399208, -0.05554581806063652, -0...
Does Knowledge Distillation Really Work?
https://openreview.net/forum?id=7J-fKoXiReA
[ "Samuel Don Stanton", "Pavel Izmailov", "Polina Kirichenko", "Alexander A Alemi", "Andrew Gordon Wilson" ]
Poster
null
Knowledge distillation is a popular technique for training a small student network to emulate a larger teacher model, such as an ensemble of networks. We show that while knowledge distillation can improve student generalization, it does not typically work as it is commonly understood: there often remains a surprisingly...
[ "distillation", "deep ensembles", "optimization" ]
We show that it is difficult to achieve high agreement between student and teacher in knowledge distillation and explain why
7,579
2106.05945
title_snapshot
[ 0.01310157310217619, -0.023047393187880516, -0.010589135810732841, 0.07915107160806656, 0.08482919633388519, -0.03107592836022377, 0.045843496918678284, -0.011274335905909538, 0.0008342525688931346, 0.0016686684684827924, -0.033096883445978165, 0.01244400255382061, -0.044619835913181305, 0...
Neural Tangent Kernel Maximum Mean Discrepancy
https://openreview.net/forum?id=egBGjXWWlH3
[ "Xiuyuan Cheng", "Yao Xie" ]
Poster
null
We present a novel neural network Maximum Mean Discrepancy (MMD) statistic by identifying a new connection between neural tangent kernel (NTK) and MMD. This connection enables us to develop a computationally efficient and memory-efficient approach to compute the MMD statistic and perform NTK based two-sample tests towa...
[ "Maximum Mean Discrepancy (MMD)", "Neural Tangent Kernel (NTK)", "two-sample test", "kernel methods", "change-point detection" ]
New kernel MMD statistic computed by neural network stochastic optimization, with theoretical testing power guarantee through NTK approximation.
7,140
2106.03227
title_snapshot
[ -0.051159702241420746, -0.003938162233680487, 0.00317153288051486, 0.05375971272587776, 0.041276123374700546, 0.04255225881934166, 0.029305221512913704, 0.009514901787042618, -0.04927024245262146, -0.05662338808178902, -0.010259374044835567, 0.023849986493587494, -0.029064202681183815, -0....
Combining Latent Space and Structured Kernels for Bayesian Optimization over Combinatorial Spaces
https://openreview.net/forum?id=fxHzZlo4dxe
[ "Aryan Deshwal", "Jana Doppa" ]
Poster
null
We consider the problem of optimizing combinatorial spaces (e.g., sequences, trees, and graphs) using expensive black-box function evaluations. For example, optimizing molecules for drug design using physical lab experiments. Bayesian optimization (BO) is an efficient framework for solving such problems by intelligentl...
[ "Bayesian optimization", "Gaussian process", "Combinatorial spaces" ]
null
7,551
2111.01186
title_snapshot
[ -0.005684835370630026, -0.0012989455135539174, -0.018549581989645958, 0.05312597006559372, 0.04823391139507294, 0.037079062312841415, 0.00533108226954937, -0.034432005137205124, 0.016749953851103783, -0.04739103093743324, 0.005982383619993925, 0.01619870960712433, -0.05810530111193657, 0.0...
Online Knapsack with Frequency Predictions
https://openreview.net/forum?id=rMm9d_aDtOa
[ "Sungjin Im", "Ravi Kumar", "Mahshid Montazer Qaem", "Manish Purohit" ]
Poster
null
There has been recent interest in using machine-learned predictions to improve the worst-case guarantees of online algorithms. In this paper we continue this line of work by studying the online knapsack problem, but with very weak predictions: in the form of knowing an upper and lower bound for the number of items of ...
[ "online knapsack", "learning-augmented algorithms", "semi-online algorithms" ]
Designing optimal algorithms for online knapsack problems with range predictions on the frequency of items.
7,512
null
null
[ -0.028328338637948036, -0.02829967625439167, -0.007814832031726837, 0.012687519192695618, 0.06968335062265396, 0.032573673874139786, 0.0035183310974389315, 0.006470071617513895, -0.03593837469816208, -0.02420518361032009, -0.026654111221432686, 0.003908155020326376, -0.08219170570373535, -...
Improved Regularization and Robustness for Fine-tuning in Neural Networks
https://openreview.net/forum?id=j7buX9nsfis
[ "Dongyue Li", "Hongyang R. Zhang" ]
Poster
null
A widely used algorithm for transfer learning is fine-tuning, where a pre-trained model is fine-tuned on a target task with a small amount of labeled data. When the capacity of the pre-trained model is much larger than the size of the target data set, fine-tuning is prone to overfitting and "memorizing" the training la...
[ "fine-tuning", "transfer learning", "robustness" ]
We present a regularized self-labeling approach to improve the generalization and robustness properties of fine-tuning.
3,020
2111.04578
title_snapshot
[ 0.0001966442505363375, -0.02306799218058586, -0.00477610295638442, 0.021511705592274666, 0.058351773768663406, 0.01938236691057682, 0.03194597363471985, -0.0160131324082613, -0.031206289306282997, -0.05111084133386612, -0.004811182618141174, 0.017519168555736542, -0.049305979162454605, -0....