Split (7)

neurips · 18.5k rows

forum_id stringlengths 9 13	conference stringclasses 1 value	year int32 2.02k 2.02k ⌀	track stringclasses 3 values	venue_id stringclasses 6 values	paper_number int32 1 29.3k	title stringlengths 14 183	abstract stringlengths 246 3.6k	authors listlengths 0 56	keywords listlengths 0 32	venue stringclasses 19 values	decision stringclasses 8 values	decision_comment stringlengths 0 9.28k	num_reviews int32 2 8	reviews_json stringlengths 1.96k 78.7k	markdown stringlengths 0 1.05M	markdown_chars int64 0 1.05M
tW7L9dKZ0OM	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,527	SILG: The Multi-domain Symbolic Interactive Language Grounding Benchmark	Existing work in language grounding typically study single environments. How do we build unified models that apply across multiple environments? We propose the multi-environment Symbolic Interactive Language Grounding benchmark (SILG), which unifies a collection of diverse grounded language learning environments under ...	[ "Victor Zhong", "Austin W. Hanjie", "Sida Wang", "Karthik R Narasimhan", "Luke Zettlemoyer" ]	[ "language grounding", "reinforcement learning" ]	NeurIPS 2021 Poster	Accept (Poster)	Contributing a unified API and benchmark for language grounding in symbolically represented environments can potentially be valuable for the community. Such a benchmark can focus efforts, spark discourse about appropriate evaluation, and avoid bespoke, over-engineered models from dominating individual tasks and obscuri...	4	[{"review_id": "hK-1n77-Jx", "reviewer": "Reviewer_UhjS", "summary": "This paper introduces a unified, OpenAI Gym-based interface to five language-and-action grounding benchmarks. This interface is designed to make it easy for researchers to evaluate whether language grounding approaches work well across a variety of d...	SILG: The Multi-enviromment Symbolic Interactive Language Grounding Benchmark Victor Zhong! Austin W. Hanjie,, Sida I. Wang?, Karthik Narasimhan' and Luke Zettlemoyer Department Computer Science, University of Washington Department of Computer Science, Princeton University 'Facebook Al Research Abstract Existing work i...	51,179
tCB-SCt5wWG	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,209	Actively Identifying Causal Effects with Latent Variables Given Only Response Variable Observable	In many real tasks, it is generally desired to study the causal effect on a specific target (response variable) only, with no need to identify the thorough causal effects involving all variables. In this paper, we attempt to identify such effects by a few active interventions where only the response variable is observa...	[ "Tian-Zuo Wang", "Zhi-hua Zhou" ]	[ "causal effect identification", "maximal ancestral graph", "latent confounders", "causal discovery" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper explores a causal setting where we are interested in understanding the impact of possible interventions on a single response variable, using observational data (which might suffer from latent confounding) and limited interventional data. The limit on the interventional data is that we only observe the respons...	4	[{"review_id": "oFVEC4RlEf", "reviewer": "Reviewer_WArj", "summary": "Based on the ancestral graph framework, the authors tackle the problem of identifying an interventional distribution $f (Y \| do (X))$ by only observing the outcome $Y$ after each possible intervention. This work extended previous work by Wang et al. ...	Actively Identifying Causal Effects with Latent Variables Given Only Response Variable Observable Tian-Zuo Wang and Zhi-Hua Zhou' National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, 210023, China, Pazhou Lab, Guangzhou, 510330, China. (wangtz, (oouzh)@ammie nju. edu. cn Abstract In many ...	48,220
t-0eCf8L4-a	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,820	Learning in Non-Cooperative Configurable Markov Decision Processes	The Configurable Markov Decision Process framework includes two entities: a Reinforcement Learning agent and a configurator that can modify some environmental parameters to improve the agent's performance. This presupposes that the two actors have the same reward functions. What if the configurator does not have the sa...	[ "Giorgia Ramponi", "Alberto Maria Metelli", "Alessandro Concetti", "Marcello Restelli" ]	[ "Reinforcement Learning", "Configurable MDP", "Multi-agent learning" ]	NeurIPS 2021 Poster	Accept (Poster)	Overall, the reviewers are fairly positive on this paper, which offers a novel take on the “environment design” problem in the non-cooperative (but not full adversarial) Stackelberg-type setting, using the framework of configurable MDPs. This new formulation is rather interesting, the (problem-dependent) regret analysi...	4	[{"review_id": "kHT8RVBZdCf", "reviewer": "Reviewer_zQVR", "summary": "The paper studies learning in configurable MDPs with two-agents, a reinforcement learning agent and a configurator that can modify the environment parameters. In contrast to prior work, the paper considers a setting in which the configurator and the...	Learning in Non-Cooperative Configurable Markov Decision Processes Giorgia Ramponi" Alberto Maria Metelli Politecnico di Milano Milan, Italy al beerromaria..mmtllllppo..... ETH AI Center Zurich, Switzerland gramponi ietpoon@@ethzz.c. ch Alessandro Concetti Politecnico di Milano Marcello Restelli Politecnico di Milano M...	49,988
tDqef76wFaO	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,420	Integrating Expert ODEs into Neural ODEs: Pharmacology and Disease Progression	Modeling a system's temporal behaviour in reaction to external stimuli is a fundamental problem in many areas. Pure Machine Learning (ML) approaches often fail in the small sample regime and cannot provide actionable insights beyond predictions. A promising modification has been to incorporate expert domain knowledge i...	[ "Zhaozhi Qian", "William R. Zame", "Lucas M Fleuren", "Paul Elbers", "Mihaela van der Schaar" ]	[ "COVID-19", "Neural ODE", "Hybrid Model", "Expert-designed ODE" ]	NeurIPS 2021 Poster	Accept (Poster)	First, thanks to the authors for this engaging submission on an important topic. While the reviewers expressed some concern about understanding the method, the back and forth between the reviewers and authors appears to have addressed most of those concerns. Reviewer 62Te was concerned about the source of the improv...	4	[{"review_id": "xeSlvl60UOW", "reviewer": "Reviewer_62Te", "summary": "This paper proposes an architecture to mix Neural ODE models and more structured ODE coming from expert knowledge. They evaluate their approach on a dataset from COVID-19 patients using an expert ODE of the dexamethasone.", "questions": "", "limitat...	Integrating Expert ODEs into Neural ODEs: Pharmacology and Disease Progression Zhaozhi Qian University of Cambridge zq2240cam.ae ac. William R. Zame UCLA zame0econ. ucla. edu Lucas M. Fleuren Amsterdam UMC w.Flluren@anstteddamucc Paul Elbers Amsterdam UMC Mihaela van der Schaar University of Cambridge p elberr@@amsterd...	71,056
trNDfee72NQ	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,197	RLlib Flow: Distributed Reinforcement Learning is a Dataflow Problem	Researchers and practitioners in the field of reinforcement learning (RL) frequently leverage parallel computation, which has led to a plethora of new algorithms and systems in the last few years. In this paper, we re-examine the challenges posed by distributed RL and try to view it through the lens of an old idea: dis...	[ "Eric Liang", "Zhanghao Wu", "Michael Luo", "Sven Mika", "Joseph E. Gonzalez", "Ion Stoica" ]	[ "Distributed Reinforcement Learning", "Dataflow Programming Model" ]	NeurIPS 2021 Poster	Accept (Poster)	The reviewers and the AC appreciate this papers contribution of a nice advance in coding machinery to improve processing for parallel streams of RL data. The authors are encouraged to carefully address the multiple suggestions for further strengthening the paper in their camera ready version.	4	[{"review_id": "yteqfei1X5", "reviewer": "Reviewer_c8zk", "summary": "This paper propose new library for simplifying the implementation of distributed RL algorithms. It identifies the difference between RL workloads and stream processing workloads, and creates a number of abstractions for handling asynchronous parallel...	RLlib Flow: Distributed Reinforcement Learning is a Dataflow Problem Eric Liang" UC Berkeley Zhanghao Wu" UC Berkeley Michael Luo UC Berkeley Sven Mika Anyscale Joseph E. Gonzalez UC Berkeley Ion Stoica UC Berkeley Abstract Researchers and practitioners in the field reinforcement learning (RL. frequently leverage paral...	42,605
syIj5ggwCYJ	neurips	2,021	main	NeurIPS.cc/2021/Conference	5,689	Information Directed Sampling for Sparse Linear Bandits	Stochastic sparse linear bandits offer a practical model for high-dimensional online decision-making problems and have a rich information-regret structure. In this work we explore the use of information-directed sampling (IDS), which naturally balances the information-regret trade-off. We develop a class of information...	[ "Botao Hao", "Tor Lattimore", "Wei Deng" ]	[ "Information-directed sampling", "sparse linear bandits", "Bayesian regret" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	This paper studies sparse linear bandits in the Bayesian setting. Theoretically, it shows that information-directed sampling (IDS) has regret guarantees that adapt to (1) small/large action sets; (2) explorative action sets. Empirically, it gives a practical approximate implementation of IDS based on an empirical Bayes...	4	[{"review_id": "wNxS-9xrI2", "reviewer": "Reviewer_p7Zh", "summary": "The paper studies a high-dimensional linear bandit problem with sparse structure. The authors propose information-directed sampling (IDS), which naturally 4 balances the information-regret trade-off. They establish information-theoretic Bayesian regr...	Information Directed Sampling for Sparse Linear Bandits Botao Hao Tor Lattimore Deepmind latt.imore@google.comm com Deepmind haobotao000@gmail.com Wei Deng Department of Mathematics Purdue University weideng0566@gail.. com Abstract Stochastic sparse linear bandits offer practical model for high-dimensional online decis...	38,000
t-7Jx48oaG	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,137	Analyzing the Generalization Capability of SGLD Using Properties of Gaussian Channels	Optimization is a key component for training machine learning models and has a strong impact on their generalization. In this paper, we consider a particular optimization method---the stochastic gradient Langevin dynamics (SGLD) algorithm---and investigate the generalization of models trained by SGLD. We derive a new g...	[ "Hao Wang", "Yizhe Huang", "Rui Gao", "Flavio Calmon" ]	[ "Information theory", "statistical learning theory" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper provides new generalization guarantees for SGLD, which incorporate the gradient variance, giving a more adaptive flavor. The paper also considers differentially private stochastic optimization as an application of their results. We had quite a bit of discussion on this paper and we thank the authors for pro...	4	[{"review_id": "mpFIG0ms2oG", "reviewer": "Reviewer_QtYi", "summary": "This paper derives upper bounds on the generalization error of SGLD and DP-SGD using an information theoretic analysis. Compared to past work on this problem the authors derive upper bounds that depend on the sum of variance of the gradients along t...	Analyzing the Generalization Capability of SGLD Using Properties of Gaussian Channels Hao Wang Harvard University Yizhe Huang The University of Texas at Austin yizhehuang@utexas. edu hao_wangog. harvard. Rui Gao Flavio P. Calmon The University of Texas at Austin rui rui.gao maii.gom@mmommss utexas edu Harvard Universit...	43,581
t1czgrQOrwW	neurips	2,021	main	NeurIPS.cc/2021/Conference	5,762	The functional specialization of visual cortex emerges from training parallel pathways with self-supervised predictive learning	The visual system of mammals is comprised of parallel, hierarchical specialized pathways. Different pathways are specialized in so far as they use representations that are more suitable for supporting specific downstream behaviours. In particular, the clearest example is the specialization of the ventral ("what") and d...	[ "Shahab Bakhtiari", "Patrick J Mineault", "Tim Lillicrap", "Christopher C Pack", "Blake Aaron Richards" ]	[ "neuroscience", "visual system", "self-supervised learning", "visual cortex" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	This is a clear accept. All the reviewers were clearly positive, and I largely agree with their judgements. I think the reviews speak for themselves, and I don't have much to add here.	4	[{"review_id": "dPcNMlQ6ezI", "reviewer": "Reviewer_3Khp", "summary": "Motivated by the literature showing that the primate visual system has two pathways (\"ventral\" and \"dorsal\") with distinct functional specializations, the authors try to build a model of a visual system that also has representationally and funct...	The functional specialization of visual cortex emerges from training parallel pathways with self-supervised predictive learning Shahab Bakhtiari Patrick Mineault patrick. miw..climaiiaaattmmaaalla com Mila & McGill University bakhtias@mila quebec Tim Lillicrap Christopher C. Pack Deep McGill University cimothy111llicca...	55,090
sxjpM-kvVv_	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,104	Center Smoothing: Certified Robustness for Networks with Structured Outputs	The study of provable adversarial robustness has mostly been limited to classification tasks and models with one-dimensional real-valued outputs. We extend the scope of certifiable robustness to problems with more general and structured outputs like sets, images, language, etc. We model the output space as a metric spa...	[ "Aounon Kumar", "Tom Goldstein" ]	[ "Adversarial Robustness", "Certified Robustness", "Randomized Smoothing", "Structured Outputs" ]	NeurIPS 2021 Poster	Accept (Poster)	Thank you for your submission to NeurIPS. The reviewers and I are all in agreement that the proposed work presents a nice (if incremental, at least from a technical perspective), extension to randomize smoothing, to apply to more general output spaces than previous approaches. The results are technically sound, even ...	4	[{"review_id": "VWJN-V0GNKJ", "reviewer": "Reviewer_t5kY", "summary": "This paper aims to extend certifiable robustness methods based on randomized smoothing into the situation when the output space is a general metric space. The basic idea is to output the center of the smallest ball in the output space that can enclo...	Center Smoothing: Certified Robustness for Networks with Structured Outputs Aounon Kumar University of Maryland aounon@und. edu Tom Goldstein University of Maryland tomg@cs umd. edu Abstract The study of provable adversarial robustness has mostly been limited to classi- fication tasks and models with one-dimensional re...	55,954
swur4c3YSyF	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,249	Self-Adaptable Point Processes with Nonparametric Time Decays	Many applications involve multi-type event data. Understanding the complex influences of the events on each other is critical to discover useful knowledge and to predict future events and their types. Existing methods either ignore or partially account for these influences. Recent works use recurrent neural networks to...	[ "Zhimeng Pan", "Zheng Wang", "Jeff Phillips", "Shandian Zhe" ]	[ "Nonparametric", "Point Process", "Gaussian process" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper proposes a new marked temporal point process model for continuous-time event sequences. While the standard neural stochastic point process models (e.g., Neural Hawkes Processes) parameterized the time-varying intensity functions using recurrent neural networks, these models are often black boxes and don't al...	4	[{"review_id": "e-ToKSjWTLM", "reviewer": "Reviewer_T9br", "summary": "This paper proposes a multi-variate self-adaptable point process with nonparametric time decays. First, to flexibly represent arbitrary types of influences between events, they introduce an embedding to represent each event type, and model the influ...	Self-Adaptable Point Processes with Nonparametric Time Decays Zhimeng Pan, Zheng Wang, Jeff M. Phillips, and Shandian Zhe School of Computing. University of Utah Salt Lake City, UT 84112 z zzpan, wzhut testah.edt, jeffpécs .ut.ttthhaao edu, zhe@cs. cateess..ahh..e edu Abstract Many applications involve multi-type event...	45,976
sthiz9zeXGG	neurips	2,021	main	NeurIPS.cc/2021/Conference	11,400	DRONE: Data-aware Low-rank Compression for Large NLP Models	The representations learned by large-scale NLP models such as BERT have been widely used in various tasks. However, the increasing model size of the pre-trained models also brings efficiency challenges, including inference speed and model size when deploying models on mobile devices. Specifically, most operations in BE...	[ "Patrick CHen", "Hsiang-Fu Yu", "Inderjit S Dhillon", "Cho-Jui Hsieh" ]	[ "Acceleration", "low-rank" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper proposes DRONE, a data-aware low-rank compression algorithm for BERT, achieving a significant speed-up with marginal accuracy loss. The authors show that DRONE can be used orthogonally with knowledge distillation and quantization. In this sense, the paper has its clear value to the community. The reviewers r...	4	[{"review_id": "kwlc_si1-Je", "reviewer": "Reviewer_i8zU", "summary": "The paper proposed a data-aware low rank compression for BERT. It is stated that data-aware low rank compression could perform better than standard low rank compression like SVD since weights for matrix multiplications in BERT are not low-rank and S...	DRONE: Data-aware Low-rank Compression for Large NLP Models Patrick H. Chen Hsian-fu, Yu Amazon Inderjit S. Dhillon UT Austin & Amazon UCLA Los Angels, CA Palo Alto, CA rofu. yofu.yu@@mail com Palo Alto, CA inderjit@cs wtw.us. edu patrickchendg ucla. aela.edu Cho-jui, Hsieh UCLA Amazon Los Angels, CA chohsiehêcs ucla. ...	47,808
swdfQTe_X9	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,611	Network-to-Network Regularization: Enforcing Occam's Razor to Improve Generalization	What makes a classifier have the ability to generalize? There have been a lot of important attempts to address this question, but a clear answer is still elusive. Proponents of complexity theory find that the complexity of the classifier's function space is key to deciding generalization, whereas other recent work reve...	[ "Rohan Ghosh", "Mehul Motani" ]	[ "Generalization", "Deep Learning", "Complexity", "Label Noise" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper introduced a novel measure of the complexity of a class of functions. Reviewers agreed that the proposal was novel, and theoretically sound. Reviewers also felt that experiments exploring regularizing using (a proxy for) Kolmogorov Growth were convincing. One reviewer increased their score, and another incre...	4	[{"review_id": "zSvfxEYuSjm", "reviewer": "Reviewer_NwZ7", "summary": "This paper proposes a measure of function complexity that is based on a function itself rather than the function class it belongs to. It uses this complexity measure to establish generalization upper bounds for classifiers. In addition, an empirical...	Network-to-Network Regularization: Enforcing Occam's Razor to Improve Generalization Rohan Ghosh and Mehul Motani Department of Electrical and Computer Engineering N.1 Institute for Health Institute of Data Science National University of Singapore rghosh920gmail.. com, motani Guu edu.sp Abstract What makes classifier h...	46,299
svlanLvYsTd	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,551	Deep Neural Networks as Point Estimates for Deep Gaussian Processes	Neural networks and Gaussian processes are complementary in their strengths and weaknesses. Having a better understanding of their relationship comes with the promise to make each method benefit from the strengths of the other. In this work, we establish an equivalence between the forward passes of neural networks and ...	[ "Vincent Dutordoir", "James Hensman", "Mark van der Wilk", "Carl Henrik Ek", "Zoubin Ghahramani", "Nicolas Durrande" ]	[ "Gaussian processes", "Deep Gaussian processes", "variational inference" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper shows an equivalence between forward passes of a neural network and deep Gaussian processes. It received a positive review, and low-borderline reviews that were appreciative of the method but concerned about novelty over a workshop paper appearing this year. This paper was discussed extensively with the AC, ...	3	[{"review_id": "hF-A8MpDcrW", "reviewer": "Reviewer_AwCU", "summary": "This paper explores the connection between neural networks and Gaussian processes (GP). Specifically, this paper interprets the neural network activations as the inter-domain inducing variables for a GP. Specifically, this paper adopts the proposed ...	Deep Neural Networks as Point Estimates for Deep Gaussian Processes Vincent Dutordoir University of Cambridge Secondmind James Hensman" Amazon Mark van der Wilk Imperial College London Carl Henrik Ek University of Cambridge Zoubin Ghahramani University of Cambridge Nicolas Durrande Secondmind Google Brain Abstract Neur...	45,891
syu7m80S_CA	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,353	Learning Equivariant Energy Based Models with Equivariant Stein Variational Gradient Descent	We focus on the problem of efficient sampling and learning of probability densities by incorporating symmetries in probabilistic models. We first introduce Equivariant Stein Variational Gradient Descent algorithm -- an equivariant sampling method based on Stein's identity for sampling from densities with symmetries. Eq...	[ "Priyank Jaini", "Lars Holdijk", "Max Welling" ]	[ "Energy Based Models", "Equivariance", "Equivariant Sampling", "Stein Variational Gradient Descent", "Unsupervised Learning", "Self-Supervised Learning" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper proposes an equivariant version of the Stein variational gradient descent algorithm for sampling from densities with symmetries. Using this sampling algorithm, the authors developed equivariant versions of energy-based models for learning invariant densities. The reviewers all agree that this is an interestin...	4	[{"review_id": "apffxtSrfP", "reviewer": "Reviewer_FRDu", "summary": "This paper worked on efficient sampling from distributions that have special structures regarding symmetries. Focusing on the group invariance of the target distribution, the authors developed a new SVGD algorithm, of which kernel function incorporat...	Learning Equivariant Energy Based Models with Equivariant Stein Variational Gradient Descent Priyank Jaini" Bosch-Delds Lab University of Amsterdam Lars Holdijk" Bosch-Delta Lab University of Amsterdam Max Welling Bosch-Delta Lab University of Amsterdam Abstract We focus on the problem of efficient sampling and learnin...	40,065
t0B9XQwRDi	neurips	2,021	main	NeurIPS.cc/2021/Conference	5,579	Sim and Real: Better Together	Simulation is used extensively in autonomous systems, particularly in robotic manipulation. By far, the most common approach is to train a controller in simulation, and then use it as an initial starting point for the real system. We demonstrate how to learn simultaneously from both simulation and interaction with the ...	[ "Shirli Di-Castro Shashua", "Dotan Di Castro", "Shie Mannor" ]	[ "Reinforcement Learning", "Robotics", "Replay buffer" ]	NeurIPS 2021 Poster	Accept (Poster)	The subject of this paper is sim2real transfer, and it tackles this with two different parts. Firstly, it proposes an algorithm for sampling interactions from two different replay buffers, a simulated and a real one, the objective being to sample far less samples from the real environment, where interactions are costly...	4	[{"review_id": "qwM2Jsh8kJq", "reviewer": "Reviewer_rbtG", "summary": "The paper addresses the challenge of sim-to-real transfer, i.e., how to use simulations for robot learning in the real world. It proposes the use of information from rollouts in both the simulated and the real world. The underlying technique is remi...	Sim and Real: Better Together Shirli Di Castro Shashua Technion Institute Technology Haifa, Israel shirlidi@techmion Shie Mannor Dotan Di Castro Technion and NVIDIA Research Israel shie@technion. Bosch Center of AI Haifa, Israel dotan. ietanddicasto@il. htt.:iii..osc..ccoooooo bosch. smannor@nvidia. com Abstract Simula...	43,750
ssohLcmn4-r	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,394	Choose a Transformer: Fourier or Galerkin	In this paper, we apply the self-attention from the state-of-the-art Transformer in Attention Is All You Need for the first time to a data-driven operator learning problem related to partial differential equations. An effort is put together to explain the heuristics of, and to improve the efficacy of the attention mech...	[ "Shuhao Cao" ]	[ "Transformer", "attention", "operator learning", "partial differential equations", "Galerkin methods" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper studies Transformer variants for data-driven operator learning problems of parametric partial differential equations: the Fourier-type and the Galerkin-type. This paper includes both theoretical and empirical evidence to show that these two Transformer variants are accurate and efficient for solving various ...	4	[{"review_id": "wFPd2dTHUAf", "reviewer": "Reviewer_xknS", "summary": "This paper aims to improve the well-established attention mechanism, and challenges the necessity of its softmax normalization through an integral perspective.", "questions": "", "limitations": "", "rating": 7, "confidence": 3, "soundness": null, "p...	Choose a Transformer: Fourier or Galerkin Shuhao Cao Department of Mathematics and Statistics Washington University in St. Louis s. caa@wustl..eu Abstract In this paper, we apply the self-attention from the state-of-the-art Transformer Attention Is All You Need [88] for the first time data-driven operator learning prob...	65,769
sygvo7ctb_	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,467	Relative Flatness and Generalization	Flatness of the loss curve is conjectured to be connected to the generalization ability of machine learning models, in particular neural networks. While it has been empirically observed that flatness measures consistently correlate strongly with generalization, it is still an open theoretical problem why and under whic...	[ "Henning Petzka", "Michael Kamp", "Linara Adilova", "Cristian Sminchisescu", "Mario Boley" ]	[ "Flatness", "generalization", "interpolation" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper provides a theoretical account of the relation between flatness of minima and generalization in deep neural networks. This is an important problem that attracted recent attention. The results presented here (Theorem 6) provide novel insight and are validated by experiments. The presentation is clear.	3	[{"review_id": "qnwyTGfgJ9", "reviewer": "Reviewer_5qca", "summary": "This paper describes a connection between flatness of minima and generalization in deep neural networks. The authors define a concept called feature robustness and show that it is related to flatness. This is derived through a straightforward observa...	Relative Flatness and Generalization Henning Petzka" Michael Kamp" Lund University, Sweden henning. petzka@math. tttimatthhhhtts..... se CISPA Helmholtz Center for Information Security, Germany and Monash University, Australia michael kamp@monash. edu Linara Adilova Ruhr University Bochum, Germany and Fraumhofer IAIS C...	45,107
sneJD9juaNl	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,497	Re-ranking for image retrieval and transductive few-shot classification	In the problems of image retrieval and few-shot classification, the mainstream approaches focus on learning a better feature representation. However, directly tackling the distance or similarity measure between images could also be efficient. To this end, we revisit the idea of re-ranking the top-k retrieved images in ...	[ "Xi SHEN", "Yang Xiao", "Shell Xu Hu", "Othman Sbai", "Mathieu Aubry" ]	[ "meta-learning", "few-shot classification", "synthetic gradient", "image retrieval" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper presents an interesting approach for re-ranking based on learning similarity graph between data points to be ranked. Reviewers agree that the paper presents interesting ideas. A concern remains which is: comparison with our GNN based ranking methods. It would be important for authors to present strong GNN bas...	3	[{"review_id": "YBmY8r57os", "reviewer": "Reviewer_STkv", "summary": "This paper applies a type of graph neural network on top of fixed image representations to learn to refine similarity graphs for image retrieval and transductive few-shot classification. A contrastive loss over image pairs is used to optimize the mod...	Re-ranking for image retrieval and transductive few-shot classification Xi Shen', Yang Xiao?, Shell Xu Hu Othman Sbai*. and Mathieu Aubry² 1.2.4.5 (UMR 8049), École des Ponts ParisTech Saasunng AI Center, Cambridge Abstract In the problems of image retrieval and few-shot classification, the mainstream approaches focus ...	48,135
t4485RO6O8P	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,015	Object-aware Contrastive Learning for Debiased Scene Representation	Contrastive self-supervised learning has shown impressive results in learning visual representations from unlabeled images by enforcing invariance against different data augmentations. However, the learned representations are often contextually biased to the spurious scene correlations of different objects or object an...	[ "Sangwoo Mo", "Hyunwoo Kang", "Kihyuk Sohn", "Chun-Liang Li", "Jinwoo Shin" ]	[ "self-supervised learning", "contrastive learning", "debiased representation" ]	NeurIPS 2021 Poster	Accept (Poster)	All reviewers are positive about this paper. The idea of using CAM to localize the most salient object in an image and use that to reduce background and contextual bias in scene-centric (as opposed to object-centered) datasets for selfsupervised learning is interesting and seems to indeed alleviate the issues. I recomm...	3	[{"review_id": "wCjXcrWB4S", "reviewer": "Reviewer_1LkH", "summary": "This paper proposes an object-aware contrastive learning framework which can localize the objects in a self-supervised manner (ContraCAM) and then uses these localized objects to learn de-biased representations. The authors motivate the proposed appr...	Object-aware Contrastive Learning for Debiased Scene Representation Sangwoo Mo*!, Hyunwoo Kang"\|, Kihyuk Sohn', Chun-Liang Li², Jinwoo Shin KKAIST "Google Cloud Al hyunsoonun, (inooos)).iiioos))mmmm 0kaist. ac.kr, (kihyuks, whpnlcchnngpaaa braiauusss.cuulllnggggoooooo...o Abstract Contrastive self-supervised learning h...	55,857
sri2nuQC4Y	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,055	Learning Generalized Gumbel-max Causal Mechanisms	To perform counterfactual reasoning in Structural Causal Models (SCMs), one needs to know the causal mechanisms, which provide factorizations of conditional distributions into noise sources and deterministic functions mapping realizations of noise to samples. Unfortunately, the causal mechanism is not uniquely identifi...	[ "Guy Lorberbom", "Daniel D. Johnson", "Chris J. Maddison", "Daniel Tarlow", "Tamir Hazan" ]	[ "Gumbel max", "reparameterization trick", "structural causal model" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	This paper explores an understudied connection between causal inference and couplings learning, building upon very recent work from Oberst & Sontag. More specifically, it considers the general problem of constructing a SCM that allows counterfactual (level 3) reasoning, when given access to all interventional (level 2)...	4	[{"review_id": "l3lWoZXZ6ce", "reviewer": "Reviewer_eGhr", "summary": "The authors propose a method extending Gumbel SCM for doing counterfactual reasoning.", "questions": "", "limitations": "", "rating": 5, "confidence": 2, "soundness": null, "presentation": null, "contribution": null, "strengths": "", "weaknesses": "...	Learning Generalized Gumbel-max Causal Mechanisms Guy Lorberbom" Daniel D. JJohnson" Google Research Toronto, ON, Canada dd jaijohonnggogle... com Technion Haifa, Israel guy juy_lorber@camuus hww.obbrrrrammp.tteeeiiiooooo. technion. ac. il Chris J. Maddison Daniel Tarlow Google Research Montreal, QC, Canada Tamir Hazan...	45,487
srzTZmjko0N	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,141	Large-Scale Learning with Fourier Features and Tensor Decompositions	Random Fourier features provide a way to tackle large-scale machine learning problems with kernel methods. Their slow Monte Carlo convergence rate has motivated the research of deterministic Fourier features whose approximation error can decrease exponentially in the number of basis functions. However, due to their ten...	[ "Frederiek Wesel", "kim batselier" ]	[ "Fourier Features", "Kernel Methods", "Kernel Machines", "Tensors", "Tensor Decomposition", "Tensor Networks", "Kernel Ridge Regression", "Gaussian Process Regression", "Least-Squares Support Vector Machine" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper gives a deterministic version of the random feature model by constructing the features explicitly using certain tensor structure. The additional structure allows more efficient computation and performs well in practice. The reviewers found that the experiments are convincing and the ideas are sound. There we...	4	[{"review_id": "jz3C1qQN2r2", "reviewer": "Reviewer_pffh", "summary": "The authors propose a method for inference of large-scale gaussian process (GP) regressio problems with relatively high-dimensional inputs. The method achieves state of the art performance on real-world datasets using 2 innovations 1) formulation of...	Large-Scale Learning with Fourier Features and Tensor Decompositions Frederiek Wesel Kim Batselier Delft Center for Systems and Control Delft University of Technology f. wesel @tudelft.... ftf.tt... Delft Center for Systems and Control Delft University of Technology k. fatsueler@tudelft. Abstract Random Fourier feature...	42,675
sn0wj3Dci2J	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,936	Revisiting Smoothed Online Learning	In this paper, we revisit the problem of smoothed online learning, in which the online learner suffers both a hitting cost and a switching cost, and target two performance metrics: competitive ratio and dynamic regret with switching cost. To bound the competitive ratio, we assume the hitting cost is known to the learne...	[ "Lijun Zhang", "Wei Jiang", "Shiyin Lu", "Tianbao Yang" ]	[ "Smoothed online learning", "competitive ratio", "dynamic regret with switching cost", "polyhedral functions", "quadratic growth functions" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper considers the analysis of competitive ratio/dynamic regret for problems with hitting cost and switching cost. They provide concrete improvements in terms of nailing down the constants for the ratios in the case of polyhedral function and quadratic growth functions achieving this with a relatively simple algor...	4	[{"review_id": "iNTejGqIv_", "reviewer": "Reviewer_Njc7", "summary": "This paper studies smoothed online learning and considers two performance metrics: the competitive ratio and the dynamic regret. For the competitive ratio part, the authors study the greedy algorithm which minimizes the weighted sum of the hitting co...	Revisiting Smoothed Online Learning Lijun Zhang Wei Jiang', Shiyin Lu', Tianbao Yang² National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, China "Peng Cheng Laboratory, Shenzhen, Guangdong. China 3 Department of Computer Science, The University of lowa, lowa City, IA 52242, USA {zhanglj, ...	44,381
sjLs5OXcL7j	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,490	On the Suboptimality of Thompson Sampling in High Dimensions	In this paper we consider Thompson Sampling for combinatorial semi-bandits. We demonstrate that, perhaps surprisingly, Thompson Sampling is sub-optimal for this problem in the sense that its regret scales exponentially in the ambient dimension, and its minimax regret scales almost linearly. This phenomenon occurs under...	[ "Raymond Zhang", "Richard Combes" ]	[ "Combinatorial Bandits", "Thompson sampling" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper has received mixed reviews in the first round, but the author response has successfully addressed the concerns of the reviewers. After some internal discussion, the reviewers all agreed that the paper offers a strong and interesting contribution and should thus be accepted for publication at the conference.	4	[{"review_id": "wxEqDBBXS5v", "reviewer": "Reviewer_nhUQ", "summary": "This work considers combinatorial semi-bandit problems, and theoretically investigates the sub-obtimality of Thompson sampling in the high-dimensional feature space and empirically shows the poor performance of Thompson sampling compared to other ex...	On the Suboptimality of Thompson Sampling in High Dimensions Raymond Zhang ENS Paris Saclay, Département de Mathématiques, Gif-sur- Yvette, France raymond, peefeees-paiisseeee paris-saclay..F Richard Combes Centrale-Supelec, Laboratoire des Signaux et Systèmes (L2S), Gif-sur- France richard. cen Abstract In this paper ...	28,327
tgdoUMqlwMv	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,079	Disrupting Deep Uncertainty Estimation Without Harming Accuracy	Deep neural networks (DNNs) have proven to be powerful predictors and are widely used for various tasks. Credible uncertainty estimation of their predictions, however, is crucial for their deployment in many risk-sensitive applications. In this paper we present a novel and simple attack, which unlike adversarial attack...	[ "Ido Galil", "Ran El-Yaniv" ]	[ "uncertainty estimation", "adversarial attacks", "deep neural networks", "selective prediction", "attacking uncertainty estimation" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper proposes a simple heuristic for manipulating the uncertainty estimation of a neural network without harming its accuracy by perturbing the inputs. Two reviewers (omcp and UycQ) suggest reject, while one reviewer (PoYC) does acceptance. In the discussion period, the authors provide additional experimental resu...	3	[{"review_id": "q2ZBBrsefwd", "reviewer": "Reviewer_PoYC", "summary": "Contribute an attack where the DNN is more confident of its incorrect predictions than about its correct ones without having its accuracy reduced. Attacks contributed in the white-box and black-box regime, where the black-box regime here constitutes...	Disrupting Deep Uncertainty Estimation Without Harming Accuracy Ido Galil Ran El-Yaniv Technion, Deci.AI rani technion, ac i1 Technion idogalil igggmaii........ com Abstract Deep neural networks (DNNs) have proven to be powerful predictors and widely used for various tasks. Credible uncertainty estimation their predict...	41,645
sqZ-b0a6Wm	neurips	2,021	main	NeurIPS.cc/2021/Conference	10,358	Learning Student-Friendly Teacher Networks for Knowledge Distillation	We propose a novel knowledge distillation approach to facilitate the transfer of dark knowledge from a teacher to a student. Contrary to most of the existing methods that rely on effective training of student models given pretrained teachers, we aim to learn the teacher models that are friendly to students and, consequ...	[ "Dae Young Park", "Moon-Hyun Cha", "Changwook Jeong", "Daesin Kim", "Bohyung Han" ]	[ "deep learning", "transfer learning", "knowledge distillation" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper proposes an auxiliary loss that balances independent teacher and student training and joint training. It is largely an empirical submission, but helps address an interesting open question --- since it has been shown recently that the agreement between student and teacher are often very poor. The authors also...	4	[{"review_id": "lEtFHbeAunm", "reviewer": "Reviewer_2smh", "summary": "This paper proposes a new KD framework named SFTN that optimizes both the teacher and the student at the same time for better knowledge transfer. The experiments show that the proposed method can improve the performance of many KD methods.", "questi...	Learning Student-Friendly Teacher Networks for Knowledge Distillation Dae Young Park.l, Moon-Hyun Cha', Changwook Jeong', Dae Sin Kim!, and Bohyung Han DITT Center, Samsung Electronics, Korea ECE & ASRI, Seoul National University, Korea (p30. daeyoung, moonhyun. cha, thris.jeone daesin. kim Asamsung. com bhhan@snu.c ...	43,220
t0r2M-ndcaJ	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,481	Reinforcement Learning in Linear MDPs: Constant Regret and Representation Selection	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 ...	[ "Matteo Papini", "Andrea Tirinzoni", "Aldo Pacchiano", "Marcello Restelli", "Alessandro Lazaric", "Matteo Pirotta" ]	[ "reinforcement learning", "regret minimization", "exploration", "problem-dependent analysis", "representation selection" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper studies "fine-grained" properties of linear representations in reinforcement learning, in particular when such a representation can admit constant regret guarantees, going beyond the worst case bounds of \sqrt{T}-type. The paper also consider a representation selection setting, where one representation among ...	4	[{"review_id": "_gp6kuKqP4P", "reviewer": "Reviewer_rJnu", "summary": "This paper studies representation learning in reinforcement learning. The authors prove that the proposed UNI-SOFT assumption is sufficient and necessary for achieving the constant regret bound for a single representation. The authors also propose a...	Reinforcement Learning in Linear MDPs: Constant Regret and Representation Selection Matteo Papini Universitat Pompeu Fabra matteo. patteeoopppiii@uppt edu Andrea Tirinzoni INRIA Lille andrea. iiarraa.trrizzoniinniii... Aldo Pacchiano" Microsoft Research opacchiano@miiccoooff.ooo com Marcello Restilli Alessandro Lazaric...	43,964
sesLIFMA9x9	neurips	2,021	main	NeurIPS.cc/2021/Conference	734	Automated Dynamic Mechanism Design	We study Bayesian automated mechanism design in unstructured dynamic environments, where a principal repeatedly interacts with an agent, and takes actions based on the strategic agent's report of the current state of the world. Both the principal and the agent can have arbitrary and potentially different valuations fo...	[ "Hanrui Zhang", "Vincent Conitzer" ]	[ "mechanism design", "dynamic mechanism design", "automated mechanism design" ]	NeurIPS 2021 Poster	Accept (Poster)	SUMMARY The authors consider a dynamic mechanism design problem. In this problem a single agent interacts with a single principal over a time horizon of T rounds. The problem is as follows: There is a distribution over initial states. A start state is drawn from this distribution. In each round, the agent observes the...	4	[{"review_id": "t5nonynWlJc", "reviewer": "Reviewer_AaYK", "summary": "This paper presents and proves a linear program for unstructured dynamic mechanism design. The LP supports payments and different individual-rationality constraints. The key contributions are:\n1) Presenting an LP that provably yields an optimal mec...	Automated Dynamic Mechanism Design Hanrui Zhang Vincent Conitzer Duke University Duke University conitzer duke. .dd hrzhang@cs. duke. edu Abstract We study Bayesian automated mechanism design in unstructured dynamic environ- ments, where principal repeatedly interacts with an agent, and takes actions based on the strat...	45,493
sZu0b4WrElD	neurips	2,021	main	NeurIPS.cc/2021/Conference	8,436	Faster Non-asymptotic Convergence for Double Q-learning	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...	[ "Lin Zhao", "Huaqing Xiong", "Yingbin Liang" ]	[ "Reinforcement learning theory", "Markov decision process", "stochastic approximation" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper received borderline reviews. Some main issues were about lack of clarity, and of a lack of empirical validation of the theoretical benefits of the proposed constant step sizes. I felt the rebuttal did help clarify some of the raised issues. I would also encourage the authors to include the numerical exper...	4	[{"review_id": "vOr7_17GK4B", "reviewer": "Reviewer_Z175", "summary": "Thanks to the practical applicability of double Q-learning, this paper denoted to analyzing the finite-time convergence of double Q-learning algorithms with constant stepsizes. Building on existing analysis in Xiong etal'2020, the paper brings in so...	Faster Non-asymptotic Convergence for Double Q-learning Lin Zhao Huaqing Xiong The Ohio State University xiong. 3098osu. edu Yingbin Liang The Ohio State University liang. 8890osu. edu National University of Singapore elezhl 1unui.iuuu... edu.sg Abstract Double Q-learning (Hasselt 2010) has gained significant success i...	43,169
tHzvH4Rv1Qa	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,720	Generative Occupancy Fields for 3D Surface-Aware Image Synthesis	The advent of generative radiance fields has significantly promoted the development of 3D-aware image synthesis. The cumulative rendering process in radiance fields makes training these generative models much easier since gradients are distributed over the entire volume, but leads to diffused object surfaces. In the me...	[ "Xudong XU", "Xingang Pan", "Dahua Lin", "Bo Dai" ]	[ "Generative model", "Generative Radiance Fields", "3D-aware image synthesis" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper received reviews from four expert reviewers in the community. All reviewers agree that the proposed hybrid volume-surface approach is novel in the context of 3D generative modeling. In the initial review, there were some concerns regarding the qualitative evaluation (reviewer mS1W), results on synthetic cars ...	4	[{"review_id": "z_G8cJ5C4__", "reviewer": "Reviewer_mS1W", "summary": "The paper proposes GOF, a method to improve the geometry and quality of generative methods that use a neural radiance field representation. The core idea is to modify the network's output to predict occupancy which can be used to find the surface of...	Generative Occupancy Fields for 3D Surface-Aware Image Synthesis Xudong Xut Xingang Pan Dahua Lin't Bo Dai CUHK SenseTime Joint Lab, The Chinese University of Hong Kong Max Planck Institute for Informatics 8$ - Lab, Nanyang Technological University [xx0188 dhlin]@ie. cuhk. edu. hk "xpan@mpi- -nnf mpg. bbo.dai doo.dai@t...	43,133
tu5Wg41hWl_	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,907	Unsupervised Foreground Extraction via Deep Region Competition	We present Deep Region Competition (DRC), an algorithm designed to extract foreground objects from images in a fully unsupervised manner. Foreground extraction can be viewed as a special case of generic image segmentation that focuses on identifying and disentangling objects from the background. In this work, we rethin...	[ "Peiyu Yu", "Sirui Xie", "Xiaojian Ma", "Yixin Zhu", "Ying Nian Wu", "Song-Chun Zhu" ]	[ "unsupervised learning", "foreground extraction", "generative model", "generalization" ]	NeurIPS 2021 Poster	Accept (Poster)	All reviewers rate this work as interesting and unanimously recommend acceptance of the paper but still see room for improvement.	4	[{"review_id": "wKDMRSG12Bc", "reviewer": "Reviewer_GuBW", "summary": "This work proposes a novel unsupervised foreground segmentation technique based on Mixture of Experts energy based modelling with a generative neural network. They introduce an interesting inductive bias, pixel reassignment, which works as the main ...	Unsupervised Foreground Extraction via Deep Region Competition Peiyu Yu yupeiyu980cs. ucla. edu Sirui Xie srxieducla.edu Xiaojian Ma' xiaojian www.mla edu Yixin Zhu y@bigai ai Ving Nian Wu² ywußstat. ucla.clloooooooo edu Song-Chun Zhu 1.2.3 sczhu@stat. ucla. edu UCLA Department of Computer Science *Beijing Institute fo...	59,471
sf2BxJNXC3K	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,958	Ultrahyperbolic Neural Networks	Riemannian space forms, such as the Euclidean space, sphere and hyperbolic space, are popular and powerful representation spaces in machine learning. For instance, hyperbolic geometry is appropriate to represent graphs without cycles and has been used to extend Graph Neural Networks. Recently, some pseudo-Riemannian sp...	[ "Marc T Law" ]	[ "differential geometry", "pseudo-Riemannian manifolds", "hyperbolic geometry", "elliptic geometry", "ultrahyperbolic geometry", "optimization on manifolds", "pseudo-Riemannian optimization", "quotient manifolds", "graph neural networks" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	There is a lot of support for this paper, which proposes a mathematical elegant and convincing method for embedding graphs in non-Euclidean spaces, namely ultrahyperbolic spaces. The experiments - despite some caveats mentioned - provide sufficient evidence for the validity of the approach.	4	[{"review_id": "aYo0GQwPylz", "reviewer": "Reviewer_J8on", "summary": "This paper presents a (graph) neural network in ultrahyperbolic (semi-Riemannian) space that can be used to model hierarchical graphs with cycles. The motivation is that ultrahyperbolic manifold generalizes hyperbolic and spherical manifolds, thus p...	Utrahyperbolic Neural Networks Marc T. Law NVIDIA Abstract Riemannian space forms, such as the Euclidean space, sphere and hyperbolic space, are popular and powerful representation spaces in machine learning. For instance, hyperbolic geometry is appropriate to represent graphs without cycles and has been used to extend...	44,818
sl_0rQmHxQk	neurips	2,021	main	NeurIPS.cc/2021/Conference	9,144	Sparse Quadratic Optimisation over the Stiefel Manifold with Application to Permutation Synchronisation	We address the non-convex optimisation problem of finding a sparse matrix on the Stiefel manifold (matrices with mutually orthogonal columns of unit length) that maximises (or minimises) a quadratic objective function. Optimisation problems on the Stiefel manifold occur for example in spectral relaxations of various co...	[ "Florian Bernard", "Daniel Cremers", "Anders Johan Thunberg" ]	[ "Stiefel manifold", "quadratic optimisation", "permutation synchronisation", "sparsity", "multi-matching", "correspondence problems", "manifold optimisation", "QR decomposition", "orthogonal iteration algorithm" ]	NeurIPS 2021 Poster	Accept (Poster)	The proposed method for sparse quadratic optimization over the Stiefel manifold and its analysis are interesting. However, the technical presentation and theoretical results have to be strengthened.	4	[{"review_id": "TUOncEBifgA", "reviewer": "Reviewer_fYRj", "summary": "They address sparse quadratic optimization over the Stiefel manifold with application to permutation synchronization.", "questions": "", "limitations": "", "rating": 7, "confidence": 5, "soundness": null, "presentation": null, "contribution": null, ...	Sparse Quadratic Optimisation over the Stiefel Manifold with Application to Permutation Synchronisation Florian Bernard TU Munich, University of Bonn Daniel Cremers TU Munich Johan Thunberg Halmstad University Abstract We address the non-convex optimisation problem of finding sparse matrix the Stiefel manifold (matrice...	38,593
slvWAZohje	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,091	Fuzzy Clustering with Similarity Queries	The fuzzy or soft $k$-means objective is a popular generalization of the well-known $k$-means problem, extending the clustering capability of the $k$-means to datasets that are uncertain, vague and otherwise hard to cluster. In this paper, we propose a semi-supervised active clustering framework, where the learner is a...	[ "Wasim Huleihel", "Arya Mazumdar", "Soumyabrata Pal" ]	[ "Fuzzy clustering", "oracle", "similarity queries", "k-means", "soft clustering." ]	NeurIPS 2021 Poster	Accept (Poster)	The authors study the fuzzy k-means problem with oracle queries and present algorithm that outperform previous work. The reviewers find the result interesting and novel and so, after the discussion phase, we think that the paper should be accepted. One limitation of the paper is that it is a bit hard to read but the a...	4	[{"review_id": "acDFLHZzeDi", "reviewer": "Reviewer_E3Xr", "summary": "This work gives algorithm for the soft k-means problem using similarity queries. This means that the clustering algorithm is allowed to make queries of the form \"how similar are two data items\". There have been recent results in this setting for k...	Fuzzy Clustering with Similarity Queries Wasim Huleihel Arya Mazumdar Department Electrical Engineering Tel Aviv University Haltcroglu Data Science Institute University of California, San Diego La Jolla, CA 92093 arya@ucsd. edu Tel Aviv 6997801, Israel wasimhStauex. aa i1 Soumyabrata Pal College of Information & Comput...	49,939
scn3RYn1DYx	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,077	Transformers Generalize DeepSets and Can be Extended to Graphs & Hypergraphs	We present a generalization of Transformers to any-order permutation invariant data (sets, graphs, and hypergraphs). We begin by observing that Transformers generalize DeepSets, or first-order (set-input) permutation invariant MLPs. Then, based on recently characterized higher-order invariant MLPs, we extend the concep...	[ "Jinwoo Kim", "Saeyoon Oh", "Seunghoon Hong" ]	[ "transformer", "graph", "hypergraph", "self-attention", "graph transformer", "kernel attention", "graph regression", "graph prediction", "hyperedge prediction", "graph neural network" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper develops transformers for order permutation invariant data such as sets, graphs and hypergaphs. Overall, the reviewers and I have enjoyed reading the paper since moving transformers towards structured data is an important research direction. The experimental results presented in the rolling discussion on sha...	4	[{"review_id": "xTxDwbn4uCh", "reviewer": "Reviewer_tk1s", "summary": "The authors propose a framework to generalize transformers to any order permutation invariant data, especially hypergraphs. While the generalization itself is incremental compare to [20], the naïve approach gives unbearable computational complexity....	Transformers Generalize DeepSets and Can be Extended to Graphs and Hypergraphs Jinwoo Kim, Saeyoon Oh, Seunghoon Hong School of Computing, KAIST [jinwoo-kiie saeyoon17, seunghoon. (etpg/oonn.oog))ooo.. hong) ac. Abstract We present a generalization of Transformers to any-order permutation invariant data (sets, graphs, ...	45,619
sYNr-OqGC9m	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,983	VigDet: Knowledge Informed Neural Temporal Point Process for Coordination Detection on Social Media	Recent years have witnessed an increasing use of coordinated accounts on social media, operated by misinformation campaigns to influence public opinion and manipulate social outcomes. Consequently, there is an urgent need to develop an effective methodology for coordinated group detection to combat the misinformation o...	[ "Yizhou Zhang", "Karishma Sharma", "Yan Liu" ]	[ "Coordinated Influence Campaigns", "Disinformation", "Social Media", "Fake News", "Temporal Point Process", "Variational Inference" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper proposes a method for detecting groups of coordinated social media accounts which are engaged in misinformation campaigns based on temporal point processes. All the reviewers recognize that the application is very important as well as challenging and the methodology, even if not groundbreaking, is reasonable...	4	[{"review_id": "uBDV2nEgY6", "reviewer": "Reviewer_w5yn", "summary": "This paper propose a method for coordinated group detection in social networks. For group detection they use existing neural temporal point process to acquire account embeddings from the observed data and then apply group detection in that embedding ...	VigDet: Knowledge Informed Neural Temporal Point Process for Coordination Detection on Social Media Yizhou Zhang', Karishma Sharma", Yan Liu Department of Computer Science Viterbi School of Engineering University of Southern California {zhangyiz, krsharma, yanl iu. cs @usc..e Abstract Recent years have witnessed increa...	48,115
sjRlHsawmRf	neurips	2,021	main	NeurIPS.cc/2021/Conference	11,720	On the Representation Power of Set Pooling Networks	Point clouds and sets are input data-types which pose unique problems to deep learning. Since sets can have variable cardinality and are unchanged by permutation, the input space for these problems naturally form infinite-dimensional non-Euclidean spaces. Despite these mathematical difficulties, PointNet (Qi et al. 201...	[ "Christian Bueno", "Alan Hylton" ]	[ "universal approximation", "point clouds", "sets", "topology", "functional analysis", "non-euclidean", "invariance" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper gives a novel theoretical understanding on what functions "normalized" DeepSets and PointNets can represent, from a topological point of view. The work is novel and provides meaningful guidance to the community about the difference between these approaches; many of the (numerous) reviewers, however, had comp...	6	[{"review_id": "xbzPcQKdTo", "reviewer": "Reviewer_prSH", "summary": "This paper develops universal approximation theory for deep sets / pointNet architectures which consist of the composition of a deep network on a permutation invariant aggregation of embeddings (i.e. $f(x) = \\rho \\circ \\text{agg} \\circ \\phi \\ci...	On the Representation Power of Set Pooling Networks Christian Bucno Alan G. Hylton Department of Mathematics University of California, Santa Barbara Space Communications and Navigation NASA Glenn Research Center Cleveland, OH 44135, USA alan. g hyltonenasaa gov Santa Barbara. CA 93106 christianbueeno@ucss.. edu Abstrac...	49,113
soDi-HkzC1	neurips	2,021	main	NeurIPS.cc/2021/Conference	9,313	Generative vs. Discriminative: Rethinking The Meta-Continual Learning	Deep neural networks have achieved human-level capabilities in various learning tasks. However, they generally lose performance in more realistic scenarios like learning in a continual manner. In contrast, humans can incorporate their prior knowledge to learn new concepts efficiently without forgetting older ones. In t...	[ "Mohammad Amin Banayeean Zade", "Rasoul Mirzaiezadeh", "Hosein Hasani", "Mahdieh Soleymani Baghshah" ]	[ "Continual Learning", "Meta-Learning", "Meta-Continual Learning", "Generative Classifier", "Bayesian Classifier", "Neuro-Inspired Artificial Intelligence" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper addresses catastrophic forgetting in continual learning using inspirations from cognitive and neuroscience. The drawbacks of discriminative models are first demonstrated, and then a probabilistic generative method is presented, leveraging prototypical learning by Bayesian approach. Experimental results conf...	4	[{"review_id": "xU1pucWDV5", "reviewer": "Reviewer_vJhk", "summary": "This paper proposes a probabilistic generative classifier that is immune to catastrophic forgetting. The proposed lightweight generative classifier achieves high accuracy for new unseen continual learning problems.", "questions": "", "limitations": "...	Generative vs Discriminative: Rethinking The Meta-Continual Learning Mohammadamin Banayeeanzade, Rasoul Mirzaiezadeh', Hosein Hasani", Mahdieh Soleymani Baghshah Department of Computer Engineering Sharif University of Technology m. banayeean@gmail. com, hasanih@ce. sharif edu, sollymani@sharif..om mizzaierasoul7@gmai.....	45,186
spjlJ4jeM_	neurips	2,021	main	NeurIPS.cc/2021/Conference	496	Deceive D: Adaptive Pseudo Augmentation for GAN Training with Limited Data	Generative adversarial networks (GANs) typically require ample data for training in order to synthesize high-fidelity images. Recent studies have shown that training GANs with limited data remains formidable due to discriminator overfitting, the underlying cause that impedes the generator's convergence. This paper intr...	[ "Liming Jiang", "Bo Dai", "Wayne Wu", "Chen Change Loy" ]	[ "image and video synthesis", "generative adversarial networks", "limited data", "discriminator overfitting" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper proposes a very simple method of adaptively feeding generated instances into real data to prevent the overfitting of GAN Discriminator in the situation where the training data is limited. It is somewhat incremental in terms of novelty, considering existing label-noise techniques, and the algorithm in decidin...	4	[{"review_id": "oniArssDvt", "reviewer": "Reviewer_GodP", "summary": "This paper tries to solve discriminator overfitting problem.\n\nThe authors propose adaptive pseudo augmentation (APA).\n* APA employs the generator itself to augment the real data distribution with fake images.\n * i.e., Fake images are presented a...	Deceive D: Adaptive Pseudo Augmentation for GAN Training with Limited Data Liming Jiang' Bo Dai Wayne Wu² Chen Change Loy! S-LLa,, Nanyang Technological University (1iming002, da.d ecloy)@ntu.ee 2 Sensee Time Research wuvenyan@sensetingg com Abstract Generative adversarial networks (GANs) typically require ample data f...	42,663
srHp6A1c2z-	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,847	Adversarially Robust 3D Point Cloud Recognition Using Self-Supervisions	3D point cloud data is increasingly used in safety-critical applications such as autonomous driving. Thus, the robustness of 3D deep learning models against adversarial attacks becomes a major consideration. In this paper, we systematically study the impact of various self-supervised learning proxy tasks on different a...	[ "Jiachen Sun", "Yulong Cao", "Christopher Choy", "Zhiding Yu", "Anima Anandkumar", "Zhuoqing Mao", "Chaowei Xiao" ]	[ "Adversarial Training", "Point Cloud Recognition", "Self-supervised Learning" ]	NeurIPS 2021 Poster	Accept (Poster)	After some discussion, four reviews, all of them pretty thorough, recommend accepting this paper. I see nothing that would cause me to want to contradict this unanimous decision, and so I am recommending acceptance.	4	[{"review_id": "ziRV4SALWd3", "reviewer": "Reviewer_tgUE", "summary": "The paper provides a deep dive into the adversarial robustness of a 3D point cloud recognition models. Specifically, the paper considers PointNet, DynamicGraphCNN, and PointCloudTransformer based 3D point cloud recognition architectures for the anal...	Adversarially Robust 3D Point Cloud Recognition Using Self-Supervisions Jiachen Sun " Yulong Cao ' Christopher Choy?, ? Zhiding Yu 2, Anima Anandkumar 2.3 Z. Morley Mao I and Chaowei Xiao University of Michigan, NVIDIA, Caltech, ASU Abstract 3D point cloud data is increasingly used in safety-critical applications such ...	62,700
sW40wkwfsZp	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,174	Multi-modal Dependency Tree for Video Captioning	Generating fluent and relevant language to describe visual content is critical for the video captioning task. Many existing methods generate captions using sequence models that predict words in a left-to-right order. In this paper, we investigate a graph-structured model for caption generation by explicitly modeling th...	[ "Wentian Zhao", "Xinxiao Wu", "Jiebo Luo" ]	[ "video captioning", "tree-structured decoding" ]	NeurIPS 2021 Poster	Accept (Poster)	All reviewers who took part in the post rebuttal discussion recommend or lean to accept the paper. In my opinion, the concerns of reviewer c8JF have also been addressed. The paper contributes an interesting model of using dependency trees for video captioning, showing solid performance compared to SOTA, and clear expe...	4	[{"review_id": "mvRk_hD-hP", "reviewer": "Reviewer_2Mir", "summary": "The paper proposes a new architecture that combines a multimodal dependency tree approach with a tree-structured reinforcement learning method to improve upon existing methods on the video captioning task. The authors conduct experiments to understan...	Multi-modal Dependency Tree for Video Captioning Wentian Zhao, Xinxiao Wu' Jiebo Luo Department of Computer Science University of Rochester Rochester, NY 14627 USA jluoêes www.ddssrrcheeeer.oo edu Beijing Laboratory of Intelligent Information Technology School of Computer Science Beijing Institute of Technology No. 5, ...	40,427
sRojdWhXJx	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,387	Revitalizing CNN Attention via Transformers in Self-Supervised Visual Representation Learning	Studies on self-supervised visual representation learning (SSL) improve encoder backbones to discriminate training samples without labels. While CNN encoders via SSL achieve comparable recognition performance to those via supervised learning, their network attention is under-explored for further improvement. Motivated ...	[ "Chongjian GE", "Youwei Liang", "Yibing Song", "Jianbo Jiao", "Jue Wang", "Ping Luo" ]	[ "Self-Supervised Visual Representation Learning", "Vision Transformers" ]	NeurIPS 2021 Poster	Accept (Poster)	The rebuttal addressed all of the reviewers concerns, and all reviewers recommend acceptance. The AC agrees with this recommendation.	4	[{"review_id": "a_e6WlRhzz", "reviewer": "Reviewer_XzkT", "summary": "A self-supervised visual representation learning method is proposed in this paper. Besides the prevalent two-branch CNN architecture design, the two-branch transformer structure is introduced to improve CNN backbone features via attention enhancement...	Revitalizing CNN Attentions via Transformers in Self-Supervised Visual Representation Learning Chongjian Ge Youwei Liang² Yibing Song?* Jianbo Jiao Jue Wang² Ping Luo TTe University of Hong Kong Teencett AI Lab University of Oxford rhettgae@comeett.u haaeeeddoomeethhk..b hku. b Liangyoueei1@maii.cco com jianbodrobots.c...	51,521
sR1XB9-F-rv	neurips	2,021	main	NeurIPS.cc/2021/Conference	8,502	Antipodes of Label Differential Privacy: PATE and ALIBI	We consider the privacy-preserving machine learning (ML) setting where the trained model must satisfy differential privacy (DP) with respect to the labels of the training examples. We propose two novel approaches based on, respectively, the Laplace mechanism and the PATE framework, and demonstrate their effectiveness o...	[ "Mani Malek Esmaeili", "Ilya Mironov", "Karthik Prasad", "Igor Shilov", "Florian Tramer" ]	[ "differential privacy", "label differential privacy", "PATE", "ALIBI", "memorization attacks" ]	NeurIPS 2021 Poster	Accept (Poster)	In private deliberation, reviewers seemed to feel that the paper was sound, but perhaps not the most exciting (in particular, not the most novelty in PATE-FM). Nonetheless, they felt it was thorough enough and above the bar for NeurIPS.	4	[{"review_id": "wH8g4G7g7ix", "reviewer": "Reviewer_ufMh", "summary": " The paper considers the label DP setting and proposes ALIBI and PATE-FM to achieve Label DP. Their ALIBI approach combines Laplace noise and Bayesian inference and improves over the prior work. Moreover, they introduce a memorization attack in the...	Antipodes of Label Differential Privacy: PATE and ALIBI Mani Malek Iya Mironov Karthik Prasad" Igor Shilov* Florian Tramèr' Abstract We consider the crivacy-preservie machine learning (ML) setting where the trained model must satisfy differential privacy (DP) with respect to the labels of the training examples. We prop...	42,934
sNKpWhzEDWS	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,758	Uncertain Decisions Facilitate Better Preference Learning	Existing observational approaches for learning human preferences, such as inverse reinforcement learning, usually make strong assumptions about the observability of the human's environment. However, in reality, people make many important decisions under uncertainty. To better understand preference learning in these cas...	[ "Cassidy Laidlaw", "Stuart Russell" ]	[ "preference learning", "reward learning", "inverse reinforcement learning", "IRL", "statistical learning", "statistical learning theory", "PAC learning", "inverse decision theory", "IDT", "preference elicitation", "decision theory" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	This paper analyzes the inverse decision theory task of recovering the loss function of a decision maker making (observational) decisions under uncertainty. The paper's surprising insight is that uncertainty of the decision maker can enable better loss function recovery. This is supported with sample complexity bounds....	3	[{"review_id": "oJRHb7HUhJp", "reviewer": "Reviewer_FKEy", "summary": "The paper studies inverse decision theory---the problem of identifying the loss function of given optimal (under a given hypothesis class) state-action (covariates-label) pairs in a binary classification setting. The contributions of the paper are t...	Uncertain Decisions Facilitate Better Preference Learning Cassidy Laidlaw University of California, Berkeley berkeley edu www.woccsbbeeeeey...e.. Stuart Russell University of California, Berkeley russell0cs. berkeley. messell@cs.berkeeyy edu Abstract Existing observational approaches for learning human preferences, suc...	56,161
sVsqvF7ZfvI	neurips	2,021	main	NeurIPS.cc/2021/Conference	10,830	Derivatives and residual distribution of regularized M-estimators with application to adaptive tuning	This paper studies M-estimators with gradient-Lipschitz loss function regularized with convex penalty in linear models with Gaussian design matrix and arbitrary noise distribution. A practical example is the robust M-estimator constructed with the Huber loss and the Elastic-Net penalty and the noise distribution has he...	[ "Pierre C Bellec", "Yiwei Shen" ]	[ "M-estimators", "regularization", "robustness", "adaptive parameter tuning", "p/n->const", "Huber loss", "Elastic-Net" ]	NeurIPS 2021 Submitted	Reject	All the reviewers agree that the paper makes interesting contributions to the field of high-dimensional statitics. The authors provided informative answers to the questions raised by the area chair and the reviewers as well. The area chair has a positive view of the paper as well and believes that the paper should be e...	5	[{"review_id": "iU9JbYZaoiu", "reviewer": "Reviewer_ikDg", "summary": "Finding the right parameter when doing robust estimation is a recurring problem in the robust community. This article propose to choose the parameters in a robust regularized linear regression model by minimizing a completely data-driven criterion. ...	Derivatives and residual distribution of regularized M-estimators with application to adaptive tuning Anonymous Author(s) Affiliation Address email Abstract This paper studies M-estimators with gradient-Lppschitz loss function regularized with convex penalty in linear models with Gaussian design matrix and arbitrary no...	35,506
sUFdZqWeMM	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,920	Learning Debiased and Disentangled Representations for Semantic Segmentation	Deep neural networks are susceptible to learn biased models with entangled feature representations, which may lead to subpar performances on various downstream tasks. This is particularly true for under-represented classes, where a lack of diversity in the data exacerbates the tendency. This limitation has been address...	[ "Sanghyeok Chu", "Dongwan Kim", "Bohyung Han" ]	[ "Debiasing", "Disentanglement", "Semantic Segmentation" ]	NeurIPS 2021 Poster	Accept (Poster)	This work describes a new training scheme for semantic segmentation models (“DropClass”) that is intended to encourage the learning of feature representations that avoid entangling features from objects (e.g. car & road) that tend to co-occur/co-locate. Reviewers were generally positive about the work, and found it t...	3	[{"review_id": "kJQZl4PBjJ", "reviewer": "Reviewer_4WPr", "summary": "Recent semantic segmentation methods often learn biased representation from the dataset with highly co-located objects (e.g., motorcycle and road) as well as the class imbalance. This paper proposes a training scheme of dropping the class-specific re...	Learning Debiased and Disentangled Representations for Semantic Segmentation Sanghyeok Chu Dongwan Kim Bohyung Han ECE & ASRI, Seoul National University (sanghyeok.chu chu, dongwan 123, bhhan] Ssnu. ac.kr Abstract Deep neural networks susceptible to learn biased models with entangled feature representations, which may ...	41,863
sMIMAXqiqj3	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,113	Global Convergence of Gradient Descent for Asymmetric Low-Rank Matrix Factorization	We study the asymmetric low-rank factorization problem: \[\min_{\mathbf{U} \in \mathbb{R}^{m \times d}, \mathbf{V} \in \mathbb{R}^{n \times d}} \frac{1}{2}\\|\mathbf{U}\mathbf{V}^\top -\mathbf{\Sigma}\\|_F^2\] where $\mathbf{\Sigma}$ is a given matrix of size $m \times n$ and rank $d$. This is a canonical problem that ad...	[ "Tian Ye", "Simon Shaolei Du" ]	[ "non-convex optimization", "theory", "global convergence", "random initialization" ]	NeurIPS 2021 Poster	Accept (Poster)	We thank the authors for this submission. Overall, the paper presents the first proof that shows randomly initialized gradient descent converges to a global minimum of the asymmetric low-rank factorization problem with a polynomial rate. The paper well-motivates the approach. The authors have provided extensive respon...	4	[{"review_id": "iic2dWIv6PD", "reviewer": "Reviewer_UcqW", "summary": "This paper provides a proof of polynomial convergence rate whp of randomly initialized gradient descent on the non-convex problem of matrix factorization with the Frobenius norm. Notably, this problem was considered in previous works and only weaker...	Global Convergence of Gradient Descent for Asymmetric Low-Rank Matrix Factorization Tian Ye Institute for Interdisciplinary Information Sciences Tsinghua University yet17@maile tsinghua. edu. Simon S. Du Paul G. Allen School of Computer Science and Engineering University of Washington ssdućcs. washington, edu Abstract ...	32,137
sO4tOk2lg9I	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,386	Collaborative Uncertainty in Multi-Agent Trajectory Forecasting	Uncertainty modeling is critical in trajectory-forecasting systems for both interpretation and safety reasons. To better predict the future trajectories of multiple agents, recent works have introduced interaction modules to capture interactions among agents. This approach leads to correlations among the predicted traj...	[ "Bohan Tang", "Yiqi Zhong", "Ulrich Neumann", "Gang Wang", "Siheng Chen", "Ya Zhang" ]	[ "Multi-agent Trajectory Forecasting", "Uncertainty Analysis", "Interaction Modeling", "Deep Learning" ]	NeurIPS 2021 Poster	Accept (Poster)	All reviewers recommend accepting this paper. The work addresses the issue of estimating joint uncertainties between actors in a motion forecasting setting. Experiments are provided on some commonly used benchmarks for motion forecasting. Reviewers generally felt the paper was clearly written and the experiments were ...	3	[{"review_id": "gk4tLtlM0-", "reviewer": "Reviewer_9D3B", "summary": " Update 8/19 \n\nThanks to the authors for providing detailed responses to comments. Based on their clarifications and the other reviews, I have updated my score. \n\n** \n\nThe authors propose a collaborative uncertainty (CU) module to model th...	Collaborative Uncertainty in Multi-Agent Trajectory Forecasting Bohan Tang' Yiqi Zhong UIrich Neumann" Gang Wang" Ya Zhang' Siheng Chen' "Shanghai Jiao Tong University University of Southern California BBeijing Institute of Technology (sihenge, ya_zhang) @sju.. edu. cn 2 Lyiqizhon, meeuann]]Oueee edu "gangsang@bit. edu...	50,360
sKWgT8WppC3	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,463	Multi-Armed Bandits with Bounded Arm-Memory: Near-Optimal Guarantees for Best-Arm Identification and Regret Minimization	We study the Stochastic Multi-armed Bandit problem under bounded arm-memory. In this setting, the arms arrive in a stream, and the number of arms that can be stored in the memory at any time, is bounded. The decision-maker can only pull arms that are present in the memory. We address the problem from the perspective ...	[ "Arnab Maiti", "Vishakha Patil", "Arindam Khan" ]	[ "Multi-Armed Bandits", "Bounded Arm-Memory", "Best-Arm Identification", "Regret Minimization", "KL-Divergence", "Hoeffding's Inequality" ]	NeurIPS 2021 Poster	Accept (Poster)	The committee was divided on this paper and the discussions have highlighted that the despite several writing issues, the paper presents novel and important contributions. We suggest the authors to revise their draft for clarity but we recommend to accept this work to Neurips 2021.	4	[{"review_id": "r2pGBpNdjxW", "reviewer": "Reviewer_8qyp", "summary": "The paper studied the multi-armed bandit problem with bounded arm-memory. In this model, at each time at most $m<n$ arms (and their statistics) may be stored in memory.\nThe paper is focused on a single-pass model, in which after an arm being remove...	Multi-Armed Bandits with Bounded Arm-Memory: Near-Optimal Guarantees for Best-Arm Identification and Regret Minimization Arnab Maiti Vishakha Patil Indian Institute of Science, Bangalore, India patilv0iisc.ac. Arindam Khan Indian Institute of Technology, Indian Institute of Science, Bangalore, India Kharagpur, India ar...	47,077
sUBSPowU3L5	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,716	A PAC-Bayes Analysis of Adversarial Robustness	We propose the first general PAC-Bayesian generalization bounds for adversarial robustness, that estimate, at test time, how much a model will be invariant to imperceptible perturbations in the input. Instead of deriving a worst-case analysis of the risk of a hypothesis over all the possible perturbations, we leverage ...	[ "Paul Viallard", "Guillaume Eric VIDOT", "Amaury Habrard", "Emilie Morvant" ]	[ "Adversarial Robustness", "PAC-Bayesian", "Generalization Bound" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper studies the generalization of adversarial error using the PAC-Bayes framework. The authors provide generalization bounds that are independent of the type of perturbations. The reviewers see this direction as novel and the results of interest. The more important concerns initially raised by the reviewers we...	3	[{"review_id": "m8Ot5dG3w5", "reviewer": "Reviewer_mbWu", "summary": "Adversarial examples are test time attacks on machine learning models that add small imperceptible noise to examples with the goal of having them misclassified. This paper studies the problem in the \"PAC-Bayes\" framework where the learner produces ...	A PAC-Bayes Analysis of Adversarial Robustness Paul Viallard'; Univ Lyon, Laboratoire Hubert Curien UMR 5516, F-42023, SAINT-ETIENNE France Guillaume Vidot Amaury Habrard', Emilie Morvant' CNRS. Institut Optique Graduate School, Airbus Opération S.A.S University of Toulouse, Institut de Recherche en Informatique de Tou...	45,097
sUgpxb9QD	neurips	2,021	main	NeurIPS.cc/2021/Conference	8,595	SOSP: Efficiently Capturing Global Correlations by Second-Order Structured Pruning	Pruning neural networks reduces inference time and memory cost, as well as accelerates training when done at initialization. On standard hardware, these benefits will be especially prominent if coarse-grained structures, like feature maps, are pruned. We devise global saliency-based methods for second-order structured ...	[ "Manuel Nonnenmacher", "Thomas Pfeil", "Ingo Steinwart", "David Reeb" ]	[ "Structured Pruning", "Network Compression", "Neural Architecture Search", "Hessian Approximation", "Saliency-based Pruning", "Deep Learning", "Computer Vision" ]	NeurIPS 2021 Submitted	Reject	This paper proposes a pruning technique that exploits second order information. In order to circumvent the high computational cost, an approximate computation is proposed. The improvements brought by the presented method are mostly marginal, and the method is not really compared to the most recent state of the art. Als...	4	[{"review_id": "nuegUBVnxdz", "reviewer": "Reviewer_edNm", "summary": "This work proposes a second-order structured pruning algorithms (SOSP), which can drastically reduce training time and further improve pruned network performance by removing architecture bottlenecks", "questions": "", "limitations": "", "rating": 7,...	SOSP: Efficiently Capturing Global Correlations by Second-Order Structured Pruning Anonymous Author(s) Affiliation Address email Abstract Pruning neural networks reduces inference time and memory cost, as well accel- erates training when done at initialization. On standard hardware, these benefits will especially promi...	45,902
sS8rRmgAatA	neurips	2,021	main	NeurIPS.cc/2021/Conference	185	On Pathologies in KL-Regularized Reinforcement Learning from Expert Demonstrations	KL-regularized reinforcement learning from expert demonstrations has proved successful in improving the sample efficiency of deep reinforcement learning algorithms, allowing them to be applied to challenging physical real-world tasks. However, we show that KL-regularized reinforcement learning with behavioral reference...	[ "Tim G. J. Rudner", "Cong Lu", "Michael Osborne", "Yarin Gal", "Yee Whye Teh" ]	[ "Reinforcement Learning", "Expert Demonstrations", "Behavioral Cloning", "Gaussian processes", "Uncertainty Quantification" ]	NeurIPS 2021 Poster	Accept (Poster)	There has been a split amongst the reviewers for this paper. Although each expressed interest in the pathology of KL-regularization stated by the author, there has not been a consensus about whether the author's approach to this is impactful. I would recommend the position of the majority, 3 reviewers have found the se...	4	[{"review_id": "iTCNnGFDKlK", "reviewer": "Reviewer_Vbhn", "summary": "This paper presents a previously unrecognized pathology in KL-regularized RL with expert demonstrations. Specifically the commonly used parametric behavioral policies suffer from collapse of predictive variance at states far from the demonstrations....	On Pathologies in KL-Regularized Reinforcement Learning from Expert Demonstrations Tim G. J. Rudner" University of Oxford Cong Lu" University of Oxford Michael A. Osborne Yarin Gal Yee Whye Teh University of Oxford University of Oxford University of Oxford Abstract KL-regularized reinforcement learning from expert demo...	50,426
sNw3VBPL7rg	neurips	2,021	main	NeurIPS.cc/2021/Conference	10,283	Calibration and Consistency of Adversarial Surrogate Losses	Adversarial robustness is an increasingly critical property of classifiers in applications. The design of robust algorithms relies on surrogate losses since the optimization of the adversarial loss with most hypothesis sets is NP-hard. But, which surrogate losses should be used and when do they benefit from theoretical...	[ "Pranjal Awasthi", "Natalie Frank", "Anqi Mao", "Mehryar Mohri", "Yutao Zhong" ]	[ "Adversarial Robustness", "Learning Theory", "Consistency", "Calibration", "Statistical learning", "Classification." ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	I agree with the reviewers that this paper makes substantive contributions to understanding the role of surrogate losses in adversarial learning. To make the paper more accessible, I encourage the authors to make the exposition (at least the first few sections) more friendly to readers who don’t have an extensive backg...	4	[{"review_id": "m4X_TIODJpK", "reviewer": "Reviewer_vHnZ", "summary": "This paper provides an extensive analysis of the satisfiability of H-calibration and H-consistency for surrogate losses in binary classification, where the evaluation metric is not the standard 0-1 loss but the adversarial 0-1 loss. The adversarial ...	Calibration and Consistency of Adversarial Surrogate Losses Pranjal Awasthi Natalie S. Frank Courant Institute New York, NY 10012 nf 1 1211660@yu.eed Anqi Mao Courant Institute New York, NY 10012 Google Research New York, NY 10011 pranjalasasthi@@ooglee.co aqmao@cima nyu edu Mehryar Mohri Yutao Zhong Google Research Co...	44,433
sHu8-ux9VH	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,626	Testing Probabilistic Circuits	Probabilistic circuits (PCs) are a powerful modeling framework for representing tractable probability distributions over combinatorial spaces. In machine learning and probabilistic programming, one is often interested in understanding whether the distributions learned using PCs are close to the desired distribution. Th...	[ "Yash Pote", "Kuldeep S. Meel" ]	[ "Probabilistic Circuits", "Knowledge Compilation", "Distribution Testing" ]	NeurIPS 2021 Poster	Accept (Poster)	All reviewers recognized the value of having an approximate testing scheme to compute the total variation distance between two weighted d-DNNF formulas. During the discussion, a number of suggestions to strengthen the work before being published has been highlighted. Namely, these include - fixing the presentation...	4	[{"review_id": "jx1r5r8Qqsf", "reviewer": "Reviewer_LJEd", "summary": "This paper presents a statistical closeness test for evaluating whether the total variation distance between the probability distributions encoded by two circuits is below a lower threshold or above an upper threshold with a given confidence in poly...	Testing Probabilistic Circuits *f Yash Pote Kuldeep S. Meel School of Computing, National University of Singapore Abstract Probabilistic circuits (PCs) are powerful modeling framework for representing tractable probability distributions over combinatorial spaces. In machine learning is often interested in understanding...	34,736
sfzseGUqFrd	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,904	POODLE: Improving Few-shot Learning via Penalizing Out-of-Distribution Samples	In this work, we propose to use out-of-distribution samples, i.e., unlabeled samples coming from outside the target classes, to improve few-shot learning. Specifically, we exploit the easily available out-of-distribution samples to drive the classifier to avoid irrelevant features by maximizing the distance from protot...	[ "Duong Hoang Le", "Khoi Duc Nguyen", "Khoi Nguyen", "Quoc-Huy Tran", "Rang Nguyen", "Binh-Son Hua" ]	[ "few-shot learning" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper proposes using unlabeled samples outside of target classes to improve few-shot learning. The main idea is that discriminating these samples from in-distribution samples would allow the model to improve feature learning. Reviewers recognize that the proposed method improves over SOTA but they point to limite...	4	[{"review_id": "v_bB9cr2xpI", "reviewer": "Reviewer_FMo7", "summary": "The proposed approach leverage the out-of-distribution (OOD) samples to improve the few-shot learning (FSL). The OOD samples were exploited from easily available base classes, although random uniform sampling shows better results. Maximizing the dis...	POODLE: Improving Few-shot Learning via Penalizing Out-of-Distribution Samples Duong H. VinAI Research v. wwuunngnh@@vinai..il 11 Khoi D. Nguyen VinAI Research Khoi Nguyen VinAI Research ducminhkhoi1gmmm wathhho@@mmmii... com choinggyenucd@gmal..comm Quoc-Huy Tran Retrocausal, Inc. Rang Nguyen Binh-Son Hua VinAI Resear...	50,452
sBBnfOFtPc	neurips	2,021	main	NeurIPS.cc/2021/Conference	9,253	Locality defeats the curse of dimensionality in convolutional teacher-student scenarios	Convolutional neural networks perform a local and translationally-invariant treatment of the data: quantifying which of these two aspects is central to their success remains a challenge. We study this problem within a teacher-student framework for kernel regression, using 'convolutional' kernels inspired by the neural ...	[ "Alessandro Favero", "Francesco Cagnetta", "Matthieu Wyart" ]	[ "Deep learning", "Convolutional Neural Networks", "Kernel Methods", "Learning Curves" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper addresses a topic which has drawn a fair amount of attention recently, namely: if and how do locality and translation-invariance aid convnets in easing the curse of dimensionality. The paper does so by studying a teacher-student framework for kernel regression (see [4]). The reviewers found the implications o...	4	[{"review_id": "l1ld1ljU0AX", "reviewer": "Reviewer_qJrg", "summary": "The authors proved that the generalization error for a CNN can be made independent of the data dimension under certain assumptions. They study the problem in a teacher-student framework and the test error proportional to $P^{-\\beta}$ with P to be t...	Locality defeats the curse of dimensionality in convolutional teacher-student scenarios Alessandro Favero Francesco Cagnetta Institute of Physics Institute of Physics École Polytechnique Fédérale de Lausanne alessandro favero@epfl.ch École Polytechnique Fédérale de Lausanne francesco wwgnetta@epfl.comm ch Matthieu Wyar...	40,831
t5-Mszu1UkO	neurips	2,021	main	NeurIPS.cc/2021/Conference	9,052	Information Directed Reward Learning for Reinforcement Learning	For many reinforcement learning (RL) applications, specifying a reward is difficult. In this paper, we consider an RL setting where the agent can obtain information about the reward only by querying an expert that can, for example, evaluate individual states or provide binary preferences over trajectories. From such ex...	[ "David Lindner", "Matteo Turchetta", "Sebastian Tschiatschek", "Kamil Ciosek", "Andreas Krause" ]	[ "reward learning", "reinforcement learning", "active learning", "preference learning", "human feedback" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper proposes a new active reward learning algorithm, where where the agent is required to reason about unknown reward function by querying an expert. The main idea is to select queries that maximize the information gain about the difference in return between policies with high uncertainties of their return diffe...	4	[{"review_id": "zKLxJ4aoCCR", "reviewer": "Reviewer_2bor", "summary": "The authors propose a Bayesian method for learning the reward function in an RL setting. They study the setting in which the algorithm can query experts. The form these queries can take is quite general. They analyse their algorithm theoretically...	Information Directed Reward Learning for Reinforcement Learning David Lindner Matteo Turchetta Department of Computer Science ETH Zurich Department of Computer Science ETH Zurich david. wiw.eer@inf.etieee...ee.e matteo turchetta@inf wwweettttaiffeeh....... ch Sebastian Tschiatschek Kamil Ciosek Department of Computer S...	48,094
rxAS126OC-A	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,545	PLUGIn: A simple algorithm for inverting generative models with recovery guarantees	We consider the problem of recovering an unknown latent code vector under a known generative model. For a $d$-layer deep generative network $\mathcal{G}:\mathbb{R}^{n_0}\rightarrow \mathbb{R}^{n_d}$ with ReLU activation functions, let the observation be $\mathcal{G}(x)+\epsilon$ where $\epsilon$ is noise. We introduce ...	[ "Babhru Joshi", "Xiaowei Li", "Yaniv Plan", "Ozgur Yilmaz" ]	[ "generative neural network", "inverse problem", "denoise", "gradient type method", "non-convex optimization" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	This paper studies the problem of inverting a generative model. More precisely, we are given a ReLU network $G$ and we observe $y = G(x^) + \epsilon$ and the goal is to recovery $x^$. The paper gives strong provable guarantees under much milder assumptions than in earlier work. Earlier work of Hand and Voroninski giv...	3	[{"review_id": "ovlxSVfYVem", "reviewer": "Reviewer_eL82", "summary": "This paper introduces an algorithm for recovering an unknown latent code vector under a specific generative model: a deep generative network with ReLU activation functions. Innovation of paper compared with [10] is to show that one drops the $D_j$ a...	PLUGIn: A simple algorithm for inverting generative models with recovery guarantees Babhru Joshi Yaniv Plan Department of Mathematics The University of British Columbia Xiaowei Li Özgür Yılmaz {b. jb.jo xli, yaniv, oyilmaz)@math.. wal.maaa/mattuuu...... ubc.uuc.... ca Abstract We consider the problem of recovering unkn...	34,370
rvKD3iqtBdk	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,462	Deep Jump Learning for Off-Policy Evaluation in Continuous Treatment Settings	We consider off-policy evaluation (OPE) in continuous treatment settings, such as personalized dose-finding. In OPE, one aims to estimate the mean outcome under a new treatment decision rule using historical data generated by a different decision rule. Most existing works on OPE focus on discrete treatment settings. To...	[ "Hengrui Cai", "Chengchun Shi", "Rui Song", "Wenbin Lu" ]	[ "Statistical Learning", "Off-Policy Evaluation", "Deep Learning", "Continuous Treatments", "Multi-Scale Change Point Detection", "Precision medicine" ]	NeurIPS 2021 Poster	Accept (Poster)	The reviewers appreciated the paper and agree it provides a useful new method for off-policy evaluation supported by appropriate theory and that the paper should be accepted. The authors are expected to address the points raised by reviewers in a final version as they outlined in their response, including additional di...	4	[{"review_id": "OQT0Xa-GbAU", "reviewer": "Reviewer_8Ltc", "summary": "This paper considers off-policy evaluation with continuous treatment. The authors propose a new method using deep jump learning that adaptively discretizes the treatment space, estimates a deep model in each discretization, and combines them using a...	Deep Jump Learning for Off-Policy Evaluation in Continuous Treatment Settings Hengrui Cai Chengchun Shi North Carolina State University Raleigh, USA hcai5fncsu. edu London School of Economics and Political Science London, UK C. S.Shi7@sse.c... ac.uukk.k.kk Rui Song Wenbin Lu North Carolina State University Raleigh, USA...	57,309
sFyrGPCKQJC	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,616	Optimizing Reusable Knowledge for Continual Learning via Metalearning	When learning tasks over time, artificial neural networks suffer from a problem known as Catastrophic Forgetting (CF). This happens when the weights of a network are overwritten during the training of a new task causing forgetting of old information. To address this issue, we propose MetA Reusable Knowledge or MARK, a ...	[ "Julio Hurtado", "Alain Raymond", "Alvaro Soto" ]	[ "Continual Learning", "Meta-learning", "Knowledge transfer", "Catastrophic forgetting" ]	NeurIPS 2021 Poster	Accept (Poster)	After viewing the review and rebuttal and skimming through the paper I agree that the paper proposes a few components that are at least quite interesting, and that I feel the community can build on easily moving forward. One of the main arguments against the work is that it assumes similarity between tasks. I think p...	4	[{"review_id": "lWoQDywX8bL", "reviewer": "Reviewer_GtkW", "summary": "This paper mainly solves the catastrophic forgetting problem in task-based continual learning. In particular, this paper proposes a new meta-learning based method, MARK, which uses knowledge base to store some useful knowledge and uses meta-learning...	Optimizing Reusable Knowledge for Continual Learning via Metalearning Julio Hurtado Alain Raymond-Saer, Department of Computer Science Department of Computer Science Pontificia Universidad Católica de Chile j jahurtadofuc.cl Pontificia Universidad Católica de Chile afraymon@uc.co cl Alvaro Soto Department of Computer S...	44,415
sMYnHv5I14V	neurips	2,021	main	NeurIPS.cc/2021/Conference	10,823	XDO: A Double Oracle Algorithm for Extensive-Form Games	Policy Space Response Oracles (PSRO) is a reinforcement learning (RL) algorithm for two-player zero-sum games that has been empirically shown to find approximate Nash equilibria in large games. Although PSRO is guaranteed to converge to an approximate Nash equilibrium and can handle continuous actions, it may take an e...	[ "Stephen Marcus McAleer", "John Banister Lanier", "Kevin Wang", "Pierre Baldi", "Roy Fox" ]	[ "Double Oracle", "Two-player zero-sum games", "Reinforcement Learning" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper presents two algorithms extensive-form double oracle (XDO) and a a neural version NXDO, for solving two-player zero-sum games. The technique is clearly inspired by Policy-Space Response Oracles (PSRO): the main algorithmic novelty is casting the meta-game as an iteratively-built extensive-form game rather tha...	4	[{"review_id": "bnU8jeyfp-c", "reviewer": "Reviewer_7BdA", "summary": "This paper introduces two variants of the double oracle algorithm for extensive form games, making use of the problem structure to avoid a possibly exponential blowup in problem size in the standard algorithm within the matrix-game setting. The firs...	XDO: A Double Oracle Algorithm for Extensive-Form Games Stephen McAleer John Lanier Department of Computer Science University of California, Irvine smcaleer@uci edu Department of Computer Science University of California, Irvine jblanierQuci edu Kevin A. Wang Department of Computer Science University of California, Irv...	44,068
sAaymAJB_OW	neurips	2,021	main	NeurIPS.cc/2021/Conference	5,848	On Contrastive Representations of Stochastic Processes	Learning representations of stochastic processes is an emerging problem in machine learning with applications from meta-learning to physical object models to time series. Typical methods rely on exact reconstruction of observations, but this approach breaks down as observations become high-dimensional or noise distribu...	[ "Emile Mathieu", "Adam Foster", "Yee Whye Teh" ]	[ "contrastive learning", "neural processes", "representation learning" ]	NeurIPS 2021 Poster	Accept (Poster)	One of the reviews lacks thoroughness - I'm happy to disregard this. The work proposes meta learning through neural processes and contrastive learning. There is some disagreement about the level of contribution here, with several reviewers pointing out that the FCLR method also contributes contrastive learning, and a...	4	[{"review_id": "arQ6DU3tvWh", "reviewer": "Reviewer_jJb1", "summary": "The authors propose a method for representation of an observational context by optimization of a contrastive learning objective. In contrast to (conditional) neural process modeling, this avoids specification of a likelihood, which may be difficult ...	On Contrastive Representations of Stochastic Processes Emile Mathieu Adam Foster Yee Whye Tehi-l {emile mathieu, adam. foster, y t.w.tetssstats. ox. ac. uk, Department of Statistics, University of Oxford, United Kingdom DeepMind, United Kingdom Abstract Learning representations of stochastic processes is an emerging pr...	49,544
s95BePNvykX	neurips	2,021	main	NeurIPS.cc/2021/Conference	10,449	Video Instance Segmentation using Inter-Frame Communication Transformers	We propose a novel end-to-end solution for video instance segmentation (VIS) based on transformers. Recently, the per-clip pipeline shows superior performance over per-frame methods leveraging richer information from multiple frames. However, previous per-clip models require heavy computation and memory usage to achi...	[ "Sukjun Hwang", "Miran Heo", "Seoung Wug Oh", "Seon Joo Kim" ]	[ "video", "instance segmentation", "video instance segmentation", "tracking", "transformers" ]	NeurIPS 2021 Poster	Accept (Poster)	The submission discusses video instance segmentation, specifically an inter-frame communication transformer which significantly reduces the overhead of information that is shared between frames. This helps to attain SOTA results while running at an impressive 89FPS. All reviewers appreciated the submission in their ini...	4	[{"review_id": "w7HzabiT6e6", "reviewer": "Reviewer_G3h2", "summary": "This paper introduces the memory tokens into VisTR. They modified the box-based matching to the mask-based matching. They also employ instance matching and clip-level tracking. The experiment results on Youtube-VIS2019 and Youtube-VIS2021 demonstr...	Video Instance Segmentation using Inter-Frame Communication Transformers Sukjun Hwang' Miran Heo' Yonsei University Seoung Wug Oh Adobe Research Seon Joo Kim {sj hwang, miran, ponjoookim@@yoneeee ac. kr seoh@adobe.com com Abstract We propose novel end-to-end solution for video instance segmentation (VIS) based on trans...	38,280
s-NI4H4e3Rf	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,293	H-NeRF: Neural Radiance Fields for Rendering and Temporal Reconstruction of Humans in Motion	We present neural radiance fields for rendering and temporal (4D) reconstruction of humans in motion (H-NeRF), as captured by a sparse set of cameras or even from a monocular video. Our approach combines ideas from neural scene representation, novel-view synthesis, and implicit statistical geometric human representatio...	[ "Hongyi Xu", "Thiemo Alldieck", "Cristian Sminchisescu" ]	[ "Neural Radiance Field", "SDF", "Human Reconstruction", "Dynamics" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	The submission has received 4 positive final ratings: 6, 7, 7, 8. The reviewers overall were excited about the method and the idea of combining implicit representations with explicit priors, and also acknowledged strong empirical results and solid presentation. The remaining questions and concerns were mostly addressed...	4	[{"review_id": "lPBRVjWYMCb", "reviewer": "Reviewer_VaSk", "summary": "This paper presents a new NeRF based method for rendering and reconstruction of humans observed from sparse cameras. The main contribution is combining volumetric radiance fields and an implicit SDF for the tasks of novel view synthesis and geometri...	H-NeRF: Neural Radiance Fields for Rendering and Temporal Reconstruction of Humans in Motion Hongyi Xu Google Research hongyixu.@google com Thiemo Alldieck Google Research al slldieck@googge.coom Cristian Sminchisescu Google Research smi wesinccissscuugooglloo Abstract We present neural radiance fields for rendering an...	47,998
sGUet7yVVgn	neurips	2,021	main	NeurIPS.cc/2021/Conference	8,490	Meta-Adaptive Nonlinear Control: Theory and Algorithms	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...	[ "Guanya Shi", "Kamyar Azizzadenesheli", "Michael O'Connell", "Soon-Jo Chung", "Yisong Yue" ]	[ "Online Learning", "Representation Learning", "Adaptive Control", "Nonlinear Control", "Meta-learning" ]	NeurIPS 2021 Poster	Accept (Poster)	Overall there is not enough support here from reviewers for me to recommend acceptance, even though one reviewer raised their score from 5 to 6. Reviewers agreed on some real strengths to the paper, including that it is well-written and addresses an important topic: * Reviewer 72B6: "The manuscript is well-organized a...	4	[{"review_id": "xO0TfQsmgEB", "reviewer": "Reviewer_7eB6", "summary": "This manuscript investigates the online multi-task adaptive control of nonlinear systems subject to unknown uncertainty and disturbances. Several instantiations of the proposed approach are investigated with theoretical guarantees on the resulting a...	Meta-Adaptive Nonlinear Control: Theory and Algorithms Guanya Shit, Kamyar Azizzadenesheli, Michael O'Connell', Soon-Jo Chung', Yisong Yue Michael 'Caltech Purruee University {gshi moc, s jjchung.yyue)occalteed edu. kamyar edu Abstract We present an online multi-task learning approach for adaptive nonlinear control, wh...	42,999
rxud5HYKX55	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,760	Optimal Sketching for Trace Estimation	Matrix trace estimation is ubiquitous in machine learning applications and has traditionally relied on Hutchinson's method, which requires $O(\log(1/\delta)/\epsilon^2)$ matrix-vector product queries to achieve a $(1 \pm \epsilon)$-multiplicative approximation to $\text{trace}(A)$ with failure probability $\delta$ on p...	[ "Shuli Jiang", "Hai Pham", "David Woodruff", "Qiuyi Zhang" ]	[ "Trace estimation", "Query complexity", "Sketching algorithms" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	This paper considers the fundamental problem of estimating the trace of a PSD matrix from matrix vector product queries. Prior results had a gap between the adaptive and non-adaptive versions of the problem. In this paper, the authors propose a non-adaptive algorithm with nearly the same performance as the best adaptiv...	4	[{"review_id": "jkxm3_Jflp", "reviewer": "Reviewer_5vUg", "summary": "This paper contains two main contributions. First, they provided an improved analysis on the NA-Hutch++ algorithm to bridge the gap between adaptive and non-adaptive algorithm. Secondly, they also provided a nearly matching lower-bound on the complex...	Optimal Sketching for Trace Estimation Shuli Jiang Robotics Institute Hai Pham Language Technologies Institute Carnegie Mellon University shulij@andred edu Carnegie Mellon University htpham@cs. cmu edu David P. Woodruff Qiuyi (Richard) Zhang Computer Science Department Cameeiie Mellon University dwoodruf@ca cmu. edu Go...	50,270
rqfq0CYIekd	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,924	Reliable Post hoc Explanations: Modeling Uncertainty in Explainability	As black box explanations are increasingly being employed to establish model credibility in high stakes settings, it is important to ensure that these explanations are accurate and reliable. However, prior work demonstrates that explanations generated by state-of-the-art techniques are inconsistent, unstable, and provi...	[ "Dylan Z Slack", "Sophie Hilgard", "Sameer Singh", "Himabindu Lakkaraju" ]	[ "explanations", "interpretability", "robustness" ]	NeurIPS 2021 Poster	Accept (Poster)	There are three extremely favorable reviews and one very strong dissenting review. The dissenting opinion (after good discussion) is: LIME, SHAP , Anchors etc.. - why are these appropriate notions of post hoc explanations. Something that is akin to probability of sufficiency/ necessity etc (causal notions) are truly...	4	[{"review_id": "XCC5CcIwUgg", "reviewer": "Reviewer_TaGg", "summary": "This paper proposes an approach to computing local explanations for machine learning (ML) predictions that are argued to be more reliable and accurate than the explanations provided by the prior approaches of LIME, Anchors, MAPLE, and SHAP. The deve...	Reliable Post hoc Explanations: Modeling Uncertainty in Explainability Dylan Slack UC Irvine dslack@uci. edu Sophie Hilgard Harvard University ash7980g. harvard. edu Sameer Singh Himabindu Lakkaraju Harvard University UC Irvine sameerQuci.edi hlakkaraju@hbs. edu Abstract As black box explanations are increasingly being...	53,356
rsd-9hCIit3	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,682	On Joint Learning for Solving Placement and Routing in Chip Design	For its advantage in GPU acceleration and less dependency on human experts, machine learning has been an emerging tool for solving the placement and routing problems, as two critical steps in modern chip design flow. Being still in its early stage, there are several fundamental issues unresolved: scalability, reward de...	[ "Ruoyu Cheng", "Junchi Yan" ]	[ "Reinforcement Learning", "Combinatorial Optimization" ]	NeurIPS 2021 Poster	Accept (Poster)	There was a consensus among reviewers that this paper should be accepted. It is the first to propose an RL agent for the combined task of placement and routing in chip design. The main critique after initial reviews were insufficient experiments. However, this was addressed to the reviewers' satisfaction by extensive a...	4	[{"review_id": "viV-tx6nq9W", "reviewer": "Reviewer_5u46", "summary": "This paper proposes a joint learning method to solve the placement and routing problems together. Such problems used to be studied separately. For the placement, this paper is also solving the placement of macro and standard cells together. Moreove...	On Joint Learning for Solving Placement and Routing in Chip Design Ruoyu Cheng Junchi Yan' Department of Computer Science and Engineering MoË Key Lab of Artificial Intelligence, Al Institute Shanghai Jiao Tong University, Shanghai, China, 200240 (roy account, yan j@sjtu.edu. cn Abstract For its advantage in GPU acceler...	42,953
s6MWPKgL5XB	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,495	Necessary and sufficient graphical conditions for optimal adjustment sets in causal graphical models with hidden variables	The problem of selecting optimal backdoor adjustment sets to estimate causal effects in graphical models with hidden and conditioned variables is addressed. Previous work has defined optimality as achieving the smallest asymptotic estimation variance and derived an optimal set for the case without hidden variables. For...	[ "Jakob Runge" ]	[ "causal inference", "information theory", "statistics" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	The paper aims to get around a known negative result in the theory of covariate adjustment due to Rotnitzky and Smucler: in hidden variable causal models, an adjustment set that minimizes the variance of the target parameter does not depend only on the model, but also on the element within the model (in other words, no...	4	[{"review_id": "zglqRAiywRj", "reviewer": "Reviewer_9L1v", "summary": "The paper defines the optimality of a (valid) adjustment set in terms of a notion called 'adjustment information.' Then, it focuses on estimators for which maximizing adjustment information equates to minimizing asymptotic estimation variance. The ...	Necessary and sufficient graphical conditions for optimal adjustment sets in causal graphical models with hidden variables Jakob Runge German Aerospace Center Institute of Data Science 07745 Jena, Germany and Technische Universität Berlin 10623 Berlin, Germany j jakob. rungeddlr.... de Abstract The problem of selecting...	44,459
rkA36z2plsI	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,538	Beyond the Signs: Nonparametric Tensor Completion via Sign Series	We consider the problem of tensor estimation from noisy observations with possibly missing entries. A nonparametric approach to tensor completion is developed based on a new model which we coin as sign representable tensors. The model represents the signal tensor of interest using a series of structured sign tensors. U...	[ "Chanwoo Lee", "Miaoyan Wang" ]	[ "Tensor completion", "high dimension", "nonparametric learning", "classification." ]	NeurIPS 2021 Poster	Accept (Poster)	This paper presents an interesting and innovative new notion of tensor rank that can model high rank tensors with low complexity, and show how to use this idea to solve tensor completion problems. While the reviewers differ in their beliefs about the impact this method will have, none dispute that the paper is clear, ...	4	[{"review_id": "iz2Cs-mPjvQ", "reviewer": "Reviewer_UX9r", "summary": "This paper presents a nonparametric tensor completion technique that recovers a tensor that is not necessarily low-rank. At each target level to compare the entries to, it finds a low rank tensor that can approximate the sign tensor of difference be...	Beyond the Signs: Nonparametric Tensor Completion via Sign Series Chanwoo Lee Department of Statistics University of Wisconsin-Madise chanwoo lee@wisc. edu Miaoyan Wang Department of Statistics University Wisconsin-Madison miaoyan. wang@wisc. edu Abstract We consider the problem of tensor estimation from noisy observat...	45,202
t8HduwpoQQv	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,885	Robust Inverse Reinforcement Learning under Transition Dynamics Mismatch	We study the inverse reinforcement learning (IRL) problem under a transition dynamics mismatch between the expert and the learner. Specifically, we consider the Maximum Causal Entropy (MCE) IRL learner model and provide a tight upper bound on the learner's performance degradation based on the $\ell_1$-distance between ...	[ "Luca Viano", "Yu-Ting Huang", "Parameswaran Kamalaruban", "Adrian Weller", "Volkan Cevher" ]	[ "Inverse Reinforcement Learning", "Imitation Learning", "Robust Reinforcement Learning" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper considers the inverse reinforcement learning setting with a mismatch between the estimated transition dynamics and the true dynamics (due to the environment being only known from state-only demonstrations and not possible to directly query). This is an important practical problem for IRL methods that the aut...	4	[{"review_id": "j_p9cD2M1aU", "reviewer": "Reviewer_Zhym", "summary": "This paper studies the performance degradation of a Maximum Causal Entropy learner under a transition dynamics mismatch between the expert and the learner in the imitation learning setting. The authors show that the performance of the learner is bou...	Robust Inverse Reinforcement Learning under Transition Dynamics Mismatch Luca Viano LIONS, EPFL Yu-Ting Huang EPFL Parameswaran Kamalaruban' The Alan Turing Institute Adrian Weller Volkan Cevher LIONS, EPFL University of Cambridge & The Alan Turing Institute Abstract We study the inverse reinforcement learning (IRL.) p...	49,411
rsRq--gsiE	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,798	Small random initialization is akin to spectral learning: Optimization and generalization guarantees for overparameterized low-rank matrix reconstruction	Recently there has been significant theoretical progress on understanding the convergence and generalization of gradient-based methods on nonconvex losses with overparameterized models. Nevertheless, many aspects of optimization and generalization and in particular the critical role of small random initialization are n...	[ "Dominik Stöger", "Mahdi Soltanolkotabi" ]	[ "low-rank matrix recovery", "overparameterized learning", "non-convex optimization" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper provides and analyzes a setting where small random initialization has a certain implicit spectral bias. Reviewers are positive and I recommend acceptance. That said, the reviewers had many detailed concerns, which led to extensive feedback discussions as below; I request the authors address these points ca...	4	[{"review_id": "x-ofHxgU88r", "reviewer": "Reviewer_pNVA", "summary": "This paper takes an initial step to explain why small random initialization followed by a few iterations gradient descent on non-convex optimization can often converge to a global minimum point with also good generalization ability. Specifically, th...	Small random initialization is akin to spectral learning: Optimization and generalization guarantees for overparameterized low-rank matrix reconstruction Dominik Stöger Mahdi Soltanolkotabi University of Southern California Los Angeles, CA 90089 toltanoluusc. edu Katholische Universität Eichstät-Ingolstadd 85072 Eichst...	44,652
rrf6XgIS_Ek	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,167	Multi-Person 3D Motion Prediction with Multi-Range Transformers	We propose a novel framework for multi-person 3D motion trajectory prediction. Our key observation is that a human's action and behaviors may highly depend on the other persons around. Thus, instead of predicting each human pose trajectory in isolation, we introduce a Multi-Range Transformers model which contains of a ...	[ "Jiashun Wang", "Huazhe Xu", "Medhini Narasimhan", "Xiaolong Wang" ]	[ "Motion Prediction", "Multi-Person Interaction", "Transformer" ]	NeurIPS 2021 Poster	Accept (Poster)	The final scores for this paper are just below the usual threshold for acceptance. While this paper explores an interesting approach to modelling multiple 3D agent motions, the experimental work was assessed by half of the reviewers as needing more work in a number of ways to fully evaluate the method. The number of u...	4	[{"review_id": "Xm5GogtGnZ", "reviewer": "Reviewer_yyef", "summary": "They propose a novel framework for multi-person 3D motion trajectory prediction. The multi-Range Transformer model contains a local-range encoder for individual motion and a global-range encoder for social interactions. ", "questions": "", "limitatio...	Multi-Person 3D Motion Prediction with Multi-Range Transformers Jiashun Wang' Huazhe Xu² Medhini Narasimhan Xiaolong Wang' UC San Diego UC Berkeley ji#0770ucsd. edu (huazhe_xu_ _x, Binazzee_xumeedhnnee medhini) -berkeleebberreyee edu xis0120eng ucsd. edu Abstract We propose a novel framework for multi- pertiipeesonn 3D...	52,690
rjIjkiyAJao	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,404	A Faster Decentralized Algorithm for Nonconvex Minimax Problems	In this paper, we study the nonconvex-strongly-concave minimax optimization problem on decentralized setting. The minimax problems are attracting increasing attentions because of their popular practical applications such as policy evaluation and adversarial training. As training data become larger, distributed training...	[ "Wenhan Xian", "Feihu Huang", "Yanfu Zhang", "Heng Huang" ]	[ "minimax optimization", "decentralized optimization" ]	NeurIPS 2021 Poster	Accept (Poster)	All reviewers agree that this paper makes an important contribution by extending recent faster convergence results for nonconvex-strongly concave minimax problems to the decentralized setting and hence I recommend the paper for acceptance. However, I ask the authors to incorporate reviewer suggestions e.g., adding disc...	4	[{"review_id": "mdeiHyYS2OZ", "reviewer": "Reviewer_paF7", "summary": "This paper studied the nonconvex-strongly-concave minimax optimization problem on decentralized setting. Existing literature investigateing decentralized minimax problems suffers from high gradient complexity. This paper proposed a faster decentrali...	A Faster Decentralized Algorithm for Nonconvex Minimax Problems Wenhan Xian, Feihu Huang, Yanfu Zhang, Heng Huang Electrical and Computer Engincering, University of Pittsburgh, Pittsburgh, PA 15213 wex370pitt, edu, com, paz91@pitt... edu, neng.huang@pitt.c Abstract In this paper, we study the sonconve--ttrongly-com pro...	41,344
rqjfa49ODLE	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,782	Evidential Softmax for Sparse Multimodal Distributions in Deep Generative Models	Many applications of generative models rely on the marginalization of their high-dimensional output probability distributions. Normalization functions that yield sparse probability distributions can make exact marginalization more computationally tractable. However, sparse normalization functions usually require altern...	[ "Phil Chen", "Masha Itkina", "Ransalu Senanayake", "Mykel Kochenderfer" ]	[ "Deep Learning or Neural Networks", "Sparsity and Feature Selection", "Variational Inference", "(Application) Natural Language and Text Processing" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper introduces a normalizing function called evidential softmax (ev-softmax). Based on principles of evidential theory, ev-softmax is able to assigns zero probability to some classes. It extends Itkina et al. (2020) by recovering their transformation and allowing its usage at training time. The experiment section...	4	[{"review_id": "ZmGpTi5h7sZ", "reviewer": "Reviewer_G99K", "summary": "The paper introduces a normalizing function called evidential softmax (ev-softmax). Based on principles of evidential theory, ev-softmax is able to assign zero probability to classes that lack evidence of the data. It can be seen as a generalization...	Evidential Softmax for Sparse Multimodal Distributions in Deep Generative Models Phil Chen, Masha Itkina, Ransalu Senanayake, Mykel J. Kochenderfer Stanford University (philhe, mitkina, ransalu, mykel)@stanford. edu Abstract Many applications of generative models rely on the marginalization of their high- dimensional o...	47,788
rsNBA9gtDf4	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,284	Aligning Silhouette Topology for Self-Adaptive 3D Human Pose Recovery	Articulation-centric 2D/3D pose supervision forms the core training objective in most existing 3D human pose estimation techniques. Except for synthetic source environments, acquiring such rich supervision for each real target domain at deployment is highly inconvenient. However, we realize that standard foreground sil...	[ "Mugalodi Rakesh", "Jogendra Nath Kundu", "Varun Jampani", "Venkatesh Babu Radhakrishnan" ]	[ "3D Human pose estimation", "Human mesh recovery", "3D from Single Images", "Domain adaptation", "Silhouette topology extraction" ]	NeurIPS 2021 Poster	Accept (Poster)	This submission has received 4 positive final ratings: 6, 7, 6, 7. The reviewers appreciated overall novelty, extensive experiments, strong empirical performance and clear presentation. The remaining minor questions and concerns were addressed by the authors in the rebuttal, as acknowledged by the reviewers. As a resul...	4	[{"review_id": "rMfCj-N5SwC", "reviewer": "Reviewer_oYg4", "summary": "The paper proposes an online adaptation method to adapt pretrained models to new datasets through the silhouette and their proposed topological skeleton via inward distance map and outward distance map. They also propose a Chamfer-inspired topology-...	Aligning Silhouette Topology for Self-Adaptive 3D Human Pose Recovery Mugalodi Rakesh Jogendra Nath Kundu Varun Jampani Venkatesh Babu Indian Institute of Science, Bangalore 2 Google Research Abstract Articulation-centtia 2D/3D pose supervision forms the core training objective most existing 3D human posc estimation te...	47,866
rg8gNkvs3u	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,542	Probabilistic Margins for Instance Reweighting in Adversarial Training	Reweighting adversarial data during training has been recently shown to improve adversarial robustness, where data closer to the current decision boundaries are regarded as more critical and given larger weights. However, existing methods measuring the closeness are not very reliable: they are discrete and can take onl...	[ "Qizhou Wang", "Feng Liu", "Bo Han", "Tongliang Liu", "Chen Gong", "Gang Niu", "Mingyuan Zhou", "Masashi Sugiyama" ]	[ "Adversarial robustness" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper studied the reweighting strategy in adversarial training and got 8776 scores. Before rebuttal, only one reviewer had some concerns on the experiments, while the authors provided a successful rebuttal with additional experiments and addressed the reviewer's concerns. Thus, I recommend accepting the paper.	4	[{"review_id": "tCb2qcFU8-", "reviewer": "Reviewer_SNCE", "summary": "This paper focuses on reweighting adversarial data in adversarial training problems. Actually, how to determine the sensible weights for each adversarial example matters for the adversarial robustness of the trained model. In this way, this paper pro...	Probabilistic Margins for Instance Reweighting in Adversarial Training Qizhou Wang 1. Feng Liu Bo Han Tongliang Liu?, Chen Gong Gang Niu Mingyuan Zhou?, Masashi Sugiyama Department of Computer Science, Hong Kong Baptist University DeST Lab, Australian Artificial Intelligence Institute, University of Technology Sydney S...	40,394
reOnED4N_P-	neurips	2,021	main	NeurIPS.cc/2021/Conference	9,550	ParK: Sound and Efficient Kernel Ridge Regression by Feature Space Partitions	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...	[ "Luigi Carratino", "Stefano Vigogna", "Daniele Calandriello", "Lorenzo Rosasco" ]	[ "Large-scale kernel methods", "kernel ridge regression", "random projections", "partitions" ]	NeurIPS 2021 Poster	Accept (Poster)	The focus of the submission is the solution of kernel ridge regression (KRR) in the large-scale setting. Particularly, the authors propose to use a combination of partitioning (a Voronoi one in the feature space), preconditioned iterative optimization and sketching. The resulting KRR solver is shown to be able to achie...	4	[{"review_id": "oGAyxqXRAJf", "reviewer": "Reviewer_J64c", "summary": "The paper proposes a computationally efficient algorithm for KRR, called \"ParK\". ParK combines three ideas in the literature: partitioning input samples, the FALKON solver for KRR, and Nystrom subsampling. After partitioning, FALKON and Nystrom ar...	ParK: Sound and Efficient Kernel Ridge Regression by Feature Space Partitions Luigi Carratino" MaLGa DIBRIS, University of Genova carraaiirrati unige. bagge.1 it Stefano Vigogna" MaLGa DIBRIS, University of Genova vigognaédibris. unige.it it Daniele Calandriello Lorenzo Rosasco DeepMind Paris MaLGa DIBRIS, University o...	41,989
sMRdrUIrZbT	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,007	Biological learning in key-value memory networks	In neuroscience, classical Hopfield networks are the standard biologically plausible model of long-term memory, relying on Hebbian plasticity for storage and attractor dynamics for recall. In contrast, memory-augmented neural networks in machine learning commonly use a key-value mechanism to store and read out memories...	[ "Danil Tyulmankov", "Ching Fang", "Annapurna Vadaparty", "Guangyu Robert Yang" ]	[ "neuroscience", "computational neuroscience", "memory", "recall", "biological learning", "learning rule", "synaptic plasticity", "key-value network" ]	NeurIPS 2021 Poster	Accept (Poster)	The significance of the proposal was not satisfactorily established for acceptance. While the committee accepts that the authors' burden is to prove their proposal is an interesting and valid model for biological memory (rather than being competitive with state-of-the-art non-biological memory networks), the experiment...	4	[{"review_id": "xSBnifyLMwt", "reviewer": "Reviewer_dV7b", "summary": "The paper proposes a new ML-inspired model for biological associative memory. Auto-associative memory networks are a popular model for biological memory, with one-shot learning implemented by hebbian learning and attractor dynamics / gradient descen...	Biological learning in key-value memory networks Danil Tyulmankov' Ching Fang" Columbia University ching. fawg@coolmmiia..ee.m udu Columbia University dt:256@cclumbia. edu Annapurna Vadaparty Guangyu Robert Yang Columbia University Columbia University Stanford University Massachusetts Institute of Technology yanggremit...	41,722
s6JD_xBS31	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,847	Directed Graph Contrastive Learning	Graph Contrastive Learning (GCL) has emerged to learn generalizable representations from contrastive views. However, it is still in its infancy with two concerns: 1) changing the graph structure through data augmentation to generate contrastive views may mislead the message passing scheme, as such graph changing action...	[ "Zekun Tong", "Yuxuan Liang", "Henghui Ding", "Yongxing Dai", "Xinke Li", "Changhu Wang" ]	[ "Contrastive Learning", "Directed Graph", "Graph Neural Network", "Curriculum Learning" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper proposes a novel idea of self-supervised learning with digraphs. The scores of the paper are somewhat borderline and the reviewers have had adequate reviews and discussions with the authors. The pros and cons of the paper are well discussed by the reviewers in the reviews. The AC finds the concerns on the nov...	4	[{"review_id": "DtiKjT58Au_", "reviewer": "Reviewer_haDS", "summary": "The paper proposes to adopt graph contrastive learning (GCL) on directed graphs by addressing two problems: (i) retaining structure information (specific for digraph) in augmentations; (ii) increasing views of contrasting in the CL framework.\nTo re...	Directed Graph Contrastive Learning Zekun Tong' Yuxuan Liang' Henghui Ding 2,3,* Yongxing Dai Xinke Li' Changhu Wang² National University Singapore 'ByteDance "ETH Zürich "Peking University {zekuntong, 1 Liangyuxuan, xinke. 1i] @u nus. .du edu henghui www.ddiiiiiiooooo ee ethz. ch, yongxingdai edu. cn changhu. wang@byt...	53,945
rq_UD6IiBpX	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,274	Simple steps are all you need: Frank-Wolfe and generalized self-concordant functions	Generalized self-concordance is a key property present in the objective function of many important learning problems. We establish the convergence rate of a simple Frank-Wolfe variant that uses the open-loop step size strategy $\gamma_t = 2/(t+2)$, obtaining a $\mathcal{O}(1/t)$ convergence rate for this class of func...	[ "Alejandro Carderera", "Mathieu Besançon", "Sebastian Pokutta" ]	[ "Convex Optimization", "Machine Learning" ]	NeurIPS 2021 Poster	Accept (Poster)	Overall all reviewers, and also me, liked the paper, which very elegantly simplifies a recent result of interest regarding convergence of Frank-Wolfe for self-concordent functions, which required a more involved appraoch.	4	[{"review_id": "_7JiYZMB0qz", "reviewer": "Reviewer_62NX", "summary": "This work studies simple step size, i.e., 2/(t+2), in FW for generalized self-concordant functions. The main idea is to wait the step size to shrink until f(x_{t+1}) <= f(x_t). By doing so, one can avoid relying on second order information or line s...	Simple steps are all you need: Frank-Wolfe and generalized self-concordant functions Alejandro Carderera Mathieu Besançon Sebastian Pokutta Abstract Generalized self-concordance is key property present in the objective function many important learning problems. We establish the convergence rate of simple Frank-Wolfe va...	40,930
rl2FreDHTb0	neurips	2,021	main	NeurIPS.cc/2021/Conference	5,484	Progressive Feature Interaction Search for Deep Sparse Network	Deep sparse networks (DSNs), of which the crux is exploring the high-order feature interactions, have become the state-of-the-art on the prediction task with high-sparsity features. However, these models suffer from low computation efficiency, including large model size and slow model inference, which largely limits th...	[ "Chen Gao", "Yinfeng Li", "quanming yao", "Depeng Jin", "Yong Li" ]	[ "Deep Sparse Network", "Neural Architecture Search", "Automated Machine Learning" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper proposes using NAS for learning the feature interaction space for deep sparse networks that are used for high dimensional sparse networks commonly encountered in recommendation systems. The discovered architectures have higher efficiency compared to the baselines. During discussion period some points of co...	5	[{"review_id": "p6AzvSoDgLc", "reviewer": "Reviewer_5u8P", "summary": "This paper proposes a neural architecture search approach, PROFIT, for deep sparse networks. PROFIT introduces a distilled low-rank search space and a progressive search algorithm from the lower orders. The proposed method is evaluated on three be...	Progressive Feature Interaction Search for Deep Sparse Network Chen Gao', Yinfeng Li", Quanming Yao Depeng Jin', and Yong Li Beijing National Research Center for Information Science and Technology, Department of Electronic Engineering, Tsinghua University 24Paradigm Inc. 1iyong070tsinghuune edu. cn Abstract Deep sparse...	40,389
rndqBJsGoKh	neurips	2,021	main	NeurIPS.cc/2021/Conference	209	SOFT: Softmax-free Transformer with Linear Complexity	Vision transformers (ViTs) have pushed the state-of-the-art for various visual recognition tasks by patch-wise image tokenization followed by self-attention. However, the employment of self-attention modules results in a quadratic complexity in both computation and memory usage. Various attempts on approximating the se...	[ "Jiachen Lu", "Jinghan Yao", "Junge Zhang", "Xiatian Zhu", "Hang Xu", "Weiguo Gao", "Chunjing Xu", "Tao Xiang", "Li Zhang" ]	[ "Transformer" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	The reviewers unanimously agree that the presented softmax-free transformer and the kernel approximation techniques are novel and interesting. The paper also demonstrates good performance with reduced computation complexity on ImageNet. After the rebuttal, most of concerns from the reviewers are addressed. The authors ...	4	[{"review_id": "T7c3miaThhW", "reviewer": "Reviewer_M8ep", "summary": "This paper proposes a novel method named SOFT to calculate the token similarity for self-attention without softmax operation. The authors state that the computational complexity could be reduced to O(n) with SOFT. A family of backbones are designed ...	SOFT: Softmax-free Transformer with Linear Complexity Jiachen Lu Jinghan Yao Junge Zhang' Xiatian Zhu? Hang Xu Weiguo Gao Chunjing Xu Tao Xiang" Li Zhang' Fudan University Universitty of Surrey Huawei Noah's Ark Lab https: http///uaan//////ee/ea zvg github..b......................b 10/SOFT Abstract Vision transformers ...	42,288
sIDvIyR5I1R	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,202	Learning Transferable Adversarial Perturbations	While effective, deep neural networks (DNNs) are vulnerable to adversarial attacks. In particular, recent work has shown that such attacks could be generated by another deep network, leading to significant speedups over optimization-based perturbations. However, the ability of such generative methods to generalize to ...	[ "Krishna kanth Nakka", "Mathieu Salzmann" ]	[ "Adversarial attacks" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper studied the generalization of adversarial attacks induced by generative methods. The authors found that, by maximizing the difference of mid-level features of neural networks between the clean and perturbed images, the generated perturbation can be better transferred to another task setting such as a differen...	4	[{"review_id": "fzpILLwbnlj", "reviewer": "Reviewer_WvGo", "summary": "This paper shows that mid-level features show a common pattern across different domains/tasks/architectures. Motivated by this, this paper introduced to train a generative model which leveraging the mid-level features to produce adversarial examples...	Learning Transferable Adversarial Perturbations Krishna Kanth Nakka', Mathieu Salzmann 1.2 CCVLab, EPFL. Switzerland "CacarSpace. Switzerland (krishna mrrish mathieu. salzmann)@epfl. .ch Abstract While effective, deep neural networks (DNNs) are vulnerable to adversarial at- tacks. In particular, recent work has shown t...	44,684
rXppDp76U9	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,599	Making a (Counterfactual) Difference One Rationale at a Time	Rationales, snippets of extracted text that explain an inference, have emerged as a popular framework for interpretable natural language processing (NLP). Rationale models typically consist of two cooperating modules: a selector and a classifier with the goal of maximizing the mutual information (MMI) between the "sele...	[ "Mitchell Plyler", "Michael A Green", "Min Chi" ]	[ "interpretability", "rationales", "counterfactual", "data augmentation", "natural language processing", "information theory" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper initially received diverging scores, but after the author response, two reviewers updated their reviews and increased their scores. The reviewers are now unanimous in their assessment that the paper tackles an important problem using an interesting and novel approach, and draws a unique connection between ex...	4	[{"review_id": "dKBUFurw7z", "reviewer": "Reviewer_Sfbi", "summary": "This paper aims to provide a novel rationalization scheme for text classification models in NLP. The authors use the standard setup of a selector and classifier model wherein the selectors selects specific words from the input text and classifier mak...	Making a (Counterfactual) Difference One Rationale at a Time Mitchell Plyler Michael Green Department of Computer Science North Carolina State University mlplyler@ncsu.ee edu Laboratory for Analytic Sciences magree228ncsu. edu Min Chi Department of Computer Science North Carolina State University mchi@ncsu. edu Abstrac...	52,851
rbdKZJxDWWx	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,453	$\alpha$-IoU: A Family of Power Intersection over Union Losses for Bounding Box Regression	Bounding box (bbox) regression is a fundamental task in computer vision. So far, the most commonly used loss functions for bbox regression are the Intersection over Union (IoU) loss and its variants. In this paper, we generalize existing IoU-based losses to a new family of power IoU losses that have a power IoU term an...	[ "Jiabo He", "Sarah Monazam Erfani", "Xingjun Ma", "James Bailey", "Ying Chi", "Xian-Sheng Hua" ]	[ "bounding box regression", "localization loss", "object detection", "intersection over union" ]	NeurIPS 2021 Poster	Accept (Poster)	Reviewers agreed that this is a solid paper that deserves acceptance. Authors are highly encouraged to address the key comments reported by reviewers as well as to implement all the improvements (as indicated by authors in the rebuttal) in the final camera-ready version.	7	[{"review_id": "nOaCuAQUn2b", "reviewer": "Reviewer_Zi6N", "summary": "This paper proposes a novel bounding box regression loss, $\\alpha$-IoU loss, which is a family of power IoU loss. Essentially, the proposed loss is a power transformation of IoU loss. This paper demonstrates that applying such weighting improves de...	Alpha-loU: A Family of Power Intersection over Union Losses for Bounding Box Regression Jiabo Hel. Sarah Erfani', Xingjun Ma² James Bailey', Ying Chi Xian-Sheng Hua 'School of Computing and Information Systems, The University of Melbourne School of Computer Science, Fudan University DAMO Academy, Alibaba Group (jiaboh@...	45,987
r_KsP_YjX3O	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,779	Optimal Order Simple Regret for Gaussian Process Bandits	Consider the sequential optimization of a continuous, possibly non-convex, and expensive to evaluate objective function $f$. The problem can be cast as a Gaussian Process (GP) bandit where $f$ lives in a reproducing kernel Hilbert space (RKHS). The state of the art analysis of several learning algorithms shows a signif...	[ "Sattar Vakili", "Nacime Bouziani", "Sepehr Jalali", "Alberto Bernacchia", "Da-shan Shiu" ]	[ "Gaussian Process Bandit", "Confidence Intervals", "RKHS", "Optimal Order Simple Regret" ]	NeurIPS 2021 Poster	Accept (Poster)	Based on the reviews and discussion, the consensus is that the contributions of the paper are valuable and worth publication in NeurIPS, particularly the new confidence bounds leading to a simple proof of order-optimal simple regret. The initial concerns were largely resolved following the author response. One revie...	4	[{"review_id": "lt8bRpT0ejq", "reviewer": "Reviewer_yRtV", "summary": "The paper derives improved upper bounds for the best-arm identification problem in GP bandits. Specifically, they improve the previously best-known upper bound on simple regret from O(\\gamma_N/\\sqrt{N}) to O(\\sqrt{\\gamma_N/N}), where \\gamma_N i...	Optimal Order Simple Regret for Gaussian Process Bandits Sattar Vakili", Nacime Bouziani Sepehr Jalali", Alberto Bernacchia', Da-shan Shiu" MediaTek Research (sattar, vakili, sepehr. bopehr.jalle jalalii alberto. bernacchia, ds. www.sh/attrreeerrcccccoomm +mperial College London n. bouzianii@@imperriiii ac.uk Abstract ...	47,948
rqEoV-bub4E-	neurips	2,021	main	NeurIPS.cc/2021/Conference	11,621	Charting and Navigating the Space of Solutions for Recurrent Neural Networks	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...	[ "Elia Turner", "Kabir Vinay Dabholkar", "Omri Barak" ]	[ "RNN", "underspecification", "variability", "space of solutions", "neuroscience", "reverse engineering", "task optimized networks" ]	NeurIPS 2021 Poster	Accept (Poster)	The authors present in this work an analysis of the multiple solutions that can be identified by a recurrent neural network trained on a task. in particular, the authors discuss the implications for computational neuroscience studies that leverage such tools. The reviewers all agree that the work was well presented, an...	4	[{"review_id": "wLNkmxB4jhN", "reviewer": "Reviewer_teJ3", "summary": "The authors study the solutions RNNs find and characterize them based on extrapolation and recurrent dynamics. They primarily consider two cases -- a simple two neuron network performing a memory task, and larger networks performing a Ready-Set-Go ...	CHARTING AND NAVIGATING THE SPACE OF SOLUTIONS FOR RECURRENT NEURAL NETWORKS Elia Turner Kabir Dabholkar Department of Mathematics Technion, Israel Institute of Technology el iliaturnerocamuu.. technion onniiin.oee ac. Department of Mathematics Technion, Israel Institute of Technology kabir@campus technion. ac il Omri ...	41,669
rdT5GV-LnZU	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,870	Not All Attention Is All You Need	Beyond the success story of pre-trained language models (PrLMs) in recent natural language processing, they are susceptible to over-fitting due to unusual large model size. To this end, dropout serves as a therapy. However, existing methods like random-based, knowledge-based and search-based dropout are more general bu...	[ "Hongqiu Wu", "hai zhao", "Min Zhang" ]	[ "dropout", "pre-trained language model", "self-attention" ]	NeurIPS 2021 Submitted	Reject	The submission introduces an approach to preventing overfitting when fine-tuning large pre-trained NLP models. A network is trained to predict a dropout mask to improve validation performance, optimizing the network with policy gradients. Reviewers agree that the approach is interesting, and appears to be effective. Ho...	4	[{"review_id": "kDnz1WBzgSy", "reviewer": "Reviewer_xozH", "summary": "This work proposes a new adaptive dropout technique for large self-attention pre-trained language models. Their method, called AttendOut has three main elements, an Attacker, a Defender and a Generator. The Generator is trained through policy gradie...	Not All Attention Is All You Need Anonymous Author(s) Affiliation Address email Abstract Beyond the success story of pre-trained language models (PrLMs) in recent natu- ral language processing, they are susceptible to over-fitting due to unusual large model size. To this end, dropout serves therapy. However, existing m...	40,171
rdMQrE-loT5	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,227	Parallelizing Thompson Sampling	How can we make use of information parallelism in online decision-making problems while efficiently balancing the exploration-exploitation trade-off? In this paper, we introduce a batch Thompson Sampling framework for two canonical online decision-making problems with partial feedback, namely, stochastic multi-arm ban...	[ "amin karbasi", "Vahab Mirrokni", "Mohammad Shadravan" ]	[ "Thompson Sampling", "Bandit", "Batch", "Exploration", "Exploitation" ]	NeurIPS 2021 Poster	Accept (Poster)	While the problems studied in this paper look to be interesting, this paper has two weaknesses: (1) the bound $O(N\log T)$ batches is too large, compared with $O(\log T)$ or $O(\log T/\log\log T)$ using UCB or successive-elimination. The author's feedback appears to be attempting to minimize the importance of $N$ but o...	4	[{"review_id": "kPFhYpeKI0i", "reviewer": "Reviewer_2bFv", "summary": "This paper introduces a batched version of the Thompson Sampling algorithm. For the multi-arm bandit problem, within O(NlogT) batches, it is able to achieve a regret (both problem-dependent and problem-independent) that is the same order as the full...	Parallelizing Thompson Sampling Amin Karbasi Vahab Mirrokni Mohammad Shadravan Yale University Google Research mirr cirrokniien mirrokni@google... com Yale University mohammad wwaaaaaaaaaaaaaaaae.a.aaaaaaaaae edu amin.karbas @eale.aalllele.. edu Abstract How can we make use of information parallelism in online decision...	47,623
rTxCRLXRtk9	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,626	Least Square Calibration for Peer Reviews	Peer review systems such as conference paper review often suffer from the issue of miscalibration. Previous works on peer review calibration usually only use the ordinal information or assume simplistic reviewer scoring functions such as linear functions. In practice, applications like academic conferences often rely o...	[ "Sijun Tan", "Jibang Wu", "Xiaohui Bei", "Haifeng Xu" ]	[ "peer review", "calibration" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper addresses miscalibration in peer review. Miscalibration can cause problems with the fairness of the peer review process, but so far not much has been achieved in terms of theoretical algorithms that also have practical appeal. Some previous papers study simple linear models and another study considers arbitr...	4	[{"review_id": "ztRcxWUJNAg", "reviewer": "Reviewer_Emh1", "summary": "The paper addresses the problem of review calibration, i.e., to\nestimate true paper qualities from reviewer ratings, if they\nare monotonic functions in the true quantities. The authors model\nthe problem as an optimization problem that aims to min...	Least Square Calibration for Peer Reviews Sijun Tan" Jibang Wu" Department Computer Science University of Virginia Charlottesvilll, VA 22903 jw7jbevirginia. edu Department of Computer Science University of Virginia Charlottesville, VA 22903 stäeuśvirginia.c edu Xiaohui Bei Haifeng Xu Department of Computer Science Scho...	44,851
uJGObgFU0lU	neurips	2,021	main	NeurIPS.cc/2021/Conference	8,222	Differentiable Quality Diversity	Quality diversity (QD) is a growing branch of stochastic optimization research that studies the problem of generating an archive of solutions that maximize a given objective function but are also diverse with respect to a set of specified measure functions. However, even when these functions are differentiable, QD algo...	[ "Matthew Christopher Fontaine", "Stefanos Nikolaidis" ]	[ "quality diversity optimization", "generative adversarial network", "latent space exploration" ]	NeurIPS 2021 Oral	Accept (Oral)	Meta-review of Differentiable Quality Diversity This paper proposes the first differentiable version of “Quality Diversity” optimization. QD, along with other multi-objective optimization methods, look at generating a large collection of diverse solutions, and are well-explored in the evolutionary computation communit...	4	[{"review_id": "nXPkVp3EaD", "reviewer": "Reviewer_fNzE", "summary": "The paper makes three primary contributions:\n\n(1) The paper proposes a new problem setup, called Differentiable Quality Diversity (DQD), where the goal is to optimize for a set of diverse solutions that maximize a target objective. Diversity is for...	Differentiable Quality Diversity Matthew C. Fontaine University of Southern California Los Angeles, CA mfontain @ edu Stefanos Nikolaidis University of Southern California Los Angeles, CA nikolaid@usc.ed Abstract Quality diversity (QD) is growing branch of stochastic optimization research that studies the problem of ge...	48,917
rWPxhfz2_S	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,265	Contrastively Disentangled Sequential Variational Autoencoder	Self-supervised disentangled representation learning is a critical task in sequence modeling. The learnt representations contribute to better model interpretability as well as the data generation, and improve the sample efficiency for downstream tasks. We propose a novel sequence representation learning method, named C...	[ "Junwen Bai", "Weiran Wang", "Carla P Gomes" ]	[ "contrastive learning", "sequential VAE", "disentanglement", "mutual information", "self-supervised learning", "generative models" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper applies contrastive learning to improve disentanglement in sequential latent variables. The idea is based on mutual information as used in many disentangled latent variable models, and then apply contrastive learning techniques as to estimate the mutual information. Reviewers find the proposed approach nove...	4	[{"review_id": "PxJsj5xfFEW", "reviewer": "Reviewer_wLW8", "summary": "The focus of this work is disentangled representation in sequential data. It merges existing approaches/views of contrastive learning, disentanglement, and dynamical data modeling, into a unified framework with the goal to extract better representat...	Contrastively Disentangled Sequential Variational Autoencoder Junwen Bai Cornell University jb2467@corne11.c edu Weiran Wang Carla Gomes Cornell University gomesêcs cornell. Google weiranvangggoogle.ccom com Abstract Self-supervised disentangled representation learning is a critical task in sequence modeling. The learn...	48,008
rMKTq-ca0qu	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,245	Early-stopped neural networks are consistent	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...	[ "Ziwei Ji", "Justin D. Li", "Matus Telgarsky" ]	[ "Neural Networks", "Deep Networks", "calibration", "consistency", "nonseparable", "gradient descent" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	This paper analyzes shallow neural networks trained with gradient descent on logistic loss and shows that early stopping achieves optimal population risk. The reviewers and I are all in agreement that the results are strong and technically interesting, and will be valuable for the NeurIPS community. I encourage the au...	4	[{"review_id": "uBNESB9F6dQ", "reviewer": "Reviewer_Q5eH", "summary": "This paper presents a mathematical analysis of the behavior of neural networks trained with the logistic loss via gradient descent on binary classification data. ", "questions": "", "limitations": "", "rating": 6, "confidence": 2, "soundness": null,...	Early-stopped neural networks are consistent Ziwei Ji Justin D. Li Matus Telgarsky $i mjttmmiiioiiiiii University of Illinois, Urbana-Champaige Abstract 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 dis...	46,429
rMm9d_aDtOa	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,512	Online Knapsack with Frequency Predictions	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 ...	[ "Sungjin Im", "Ravi Kumar", "Mahshid Montazer Qaem", "Manish Purohit" ]	[ "online knapsack", "learning-augmented algorithms", "semi-online algorithms" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper proposes a new algorithm for the online knapsack problem. The algorithm utilizes the knowledge of upper and lower bounds for the number of items of each value, and its competitive ratio depends on the gap between the bounds. The reviewers found the problem formulation and the algorithm to be interesting and...	3	[{"review_id": "plAAP0hU4tn", "reviewer": "Reviewer_K1Av", "summary": "The paper considers the following variation of the online knapsack problem: Items of profit $p_i$ and size $s_i$ arrive online and upon arrival of an item it has to irrevocably be either packed to our knapsack (of unit capacity) or rejected. Let $v_...	Online Knapsack with Frequency Predictions Sungjin Im Electrical Engineering and Computer Science University of California, Merced sim3@ucmerced.ed. Ravi Kumar Google Research Mountain View, CA ravi wwwkkk3ammll...mmm com Mahshid Montazer Qaem Electrical Engineering and Computer Science University of California, Merced...	39,146
rHNF8Kq3u2P	neurips	2,021	main	NeurIPS.cc/2021/Conference	9,449	Neural Routing by Memory	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...	[ "Kaipeng Zhang", "Zhenqiang Li", "Zhifeng Li", "Wei Liu", "Yoichi Sato" ]	[ "convolutional neural networks", "routing", "memory" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper presents a new dynamic routing-by-memory mechanism for MoE which is simple and elegant. The mechanism effectiveness is validated by the experimental results on three public benchmarks. All the reviewers agreed the proposed memory routing is of interest to community thanks to its simplicity and good empir...	4	[{"review_id": "ovHRsZ7xSRA", "reviewer": "Reviewer_vPSw", "summary": "Routing strategy is used to select process units in convolutional neural networks. Memory is one interesting idea, which is used to record the latent information of inputs. Because different samples have different memory footprints, routing-by-memor...	Neural Routing by Memory Kaipeng Zhang!? 1.2 Zhenqiang Li Zhifeng Li Wei Liu Yoichi Sato 'Intiitute of Industrial Science, The University of Tokyo Tennent Data Platform (kpzhang, liq] taa.suu-tton u-tokyo.ac. ac. michaelefli@tencent com v122238columbia, edu yeato@iis.u-tokee ac. Abstract Recent Convolutional Neural Net...	44,900
rDdb26AQ0SO	neurips	2,021	main	NeurIPS.cc/2021/Conference	11,021	Stochastic Online Linear Regression: the Forward Algorithm to Replace Ridge	We consider the problem of online linear regression in the stochastic setting. We derive high probability regret bounds for online $\textit{ridge}$ regression and the $\textit{forward}$ algorithm. This enables us to compare online regression algorithms more accurately and eliminate assumptions of bounded observations a...	[ "Reda Ouhamma", "Odalric-Ambrym Maillard", "Vianney Perchet" ]	[ "Online linear regression", "Multi-armed bandit" ]	NeurIPS 2021 Poster	Accept (Poster)	The authors consider ridge regression and the "forward" algorithm for l2-penalized regression in the stochastic setting. High-probability regret bounds were given. A significant feature of this setting over the adversarial setting was the possibility to remove the usual boundless assumption on the inputs. The revi...	4	[{"review_id": "iz0iXqLpHy9", "reviewer": "Reviewer_WQsZ", "summary": "The paper studies the well-known ridge and forward regression in online linear regression problems with stochastic feedback. A high probability bound for both algorithms is shown for Gaussian noise variables, respectively. Furthermore, the new insi...	Stochastic Online Linear Regression: the Forward Algorithm to Replace Ridge Reda Ouhamma Univ. Lille, CNRS, Inria, Centrale Lille, UMR 9189 CRIStAL, F-59000 reda. ouharma@univ-iill.. Odalric. Maillard Univ. Lille, CNRS, Inria, Centrale Lille, UMR 9189 CRISIAL, F-59000 Vianney. Perchet Criteo, ENSAE, ENS PARIS-SACLAY Ab...	38,638
rJhCP_vC6T	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,404	Spatio-Temporal Variational Gaussian Processes	We introduce a scalable approach to Gaussian process inference that combines spatio-temporal filtering with natural gradient variational inference, resulting in a non-conjugate GP method for multivariate data that scales linearly with respect to time. Our natural gradient approach enables application of parallel filt...	[ "Oliver Hamelijnck", "William J. Wilkinson", "Niki Andreas Loppi", "Arno Solin", "Theo Damoulas" ]	[ "Gaussian processes", "Variational Inference", "Spatio-Temporal Analysis" ]	NeurIPS 2021 Poster	Accept (Poster)	The authors of this paper improve on the computational complexity of inferring spatiotemporal GPs. Specifically, via a combination of sparse priors, inducing points, and Markov separability they reduce a cubic cost to a linear cost. The reviewers all agree that the work is clear and stands to benefit the community. Mor...	4	[{"review_id": "mw_Nm9Y7ih5", "reviewer": "Reviewer_HABe", "summary": "The authors propose a novel combination of existing strategies for scalable inference for spatio-temporal GPs. The techniques utilized were Kernel separability for convenient Kronecker structure, inference by filter-smoother in the time domain, ind...	Spatio-Temporal Variational Gaussian Processes Oliver Hamelijnck' William J. Wilkinson" Aalto University william. wi wilkiinson@alllo.... Niki A. Loppi NVIDIA Aloppi@nvidia.com The Alan Turing Institute University of Warwick phameli.jcckkturing ac.uk uk Arno Solin Theodoros Damoulas Aalto University The Alan Turing Ins...	47,555
rJwDMui8DI	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,514	Neural Regression, Representational Similarity, Model Zoology & Neural Taskonomy at Scale in Rodent Visual Cortex	How well do deep neural networks fare as models of mouse visual cortex? A majority of research to date suggests results far more mixed than those produced in the modeling of primate visual cortex. Here, we perform a large-scale benchmarking of dozens of deep neural network models in mouse visual cortex with both repres...	[ "Colin Conwell", "David Mayo", "Andrei Barbu", "Michael A Buice", "George A. Alvarez", "Boris Katz" ]	[ "neuro_ai", "deep neural networks", "rodent visual cortex", "mouse brains", "optical physiology" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper presents a large-scale / systematic study fitting many modern deep nets to mouse calcium data from the Allen institute. This paper received two accepts, one borderline accept and one borderline reject and was discussed on the forum. There was a bit of a discussion regarding the actual contributions of this p...	4	[{"review_id": "xSPFij5RmsQ", "reviewer": "Reviewer_omuc", "summary": "Extracts intermediate features (across all layers) for models pretrained on ImageNet classification (and other computer vision tasks) and tests its modeling performance of mice two-photon neural responses (from 6 cortical areas) using regression and...	Neural Regression, Representational Similarity, Model Zoology & Neural Taskonomy at Scale in Rodent Visual Cortex Colin Conwell" Department Psychology David Mayo CSAIL CBMM MIT Harvard University Michael A. Buice Boris Katz CSAIL & CBMM MIT Modeling, Analysis & Theory Allen Institute MindScope Program George A. Alvarez...	68,549
rEBScZF6G70	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,989	Residual Relaxation for Multi-view Representation Learning	Multi-view methods learn representations by aligning multiple views of the same image and their performance largely depends on the choice of data augmentation. In this paper, we notice that some other useful augmentations, such as image rotation, are harmful for multi-view methods because they cause a semantic shift th...	[ "Yifei Wang", "Zhengyang Geng", "Feng Jiang", "Chuming Li", "Yisen Wang", "Jiansheng Yang", "Zhouchen Lin" ]	[ "Self-supervised Learning", "Representation Learning", "Multi-view Learning" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper relaxes the alignment objective in multiview self-supervised learning when the data augmentation causes semantic shifts in different views. The proposed pretext-aware residual relaxation method allows an adaptive residual vector between different views. The proposed method can benefit from stronger image aug...	4	[{"review_id": "utrEO-BJ2YX", "reviewer": "Reviewer_TbTy", "summary": "The paper focus on the representation learning using a relaxed alignment based on the fact that some strong data augmentations may hurt the performance. The authors introduced an adaptive residual vector and designed several novel objectives to make...	Residual Relaxation for Multi-view Representation Learning Yifei Wang's Zhengyang Geng² Feng Jiang" Chuming Li Yisen Wang2.4 Jiansheng Yang' Zhouchen Lin School of Mathematical Sciences, Peking University, China 2 Key Lab. of Machine Perception, School Artificial Intelligence, Peking University. Beijing, China School o...	39,563
rJq1SdaNPX4	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,338	Disentangled Contrastive Learning on Graphs	Recently, self-supervised learning for graph neural networks (GNNs) has attracted considerable attention because of their notable successes in learning the representation of graph-structure data. However, the formation of a real-world graph typically arises from the highly complex interaction of many latent factors. Th...	[ "Haoyang Li", "Xin Wang", "Ziwei Zhang", "Zehuan Yuan", "Hang Li", "Wenwu Zhu" ]	[ "Graph Neural Network", "Contrastive Learning", "Self-supervised Learning", "Disentangled Representation Learning" ]	NeurIPS 2021 Poster	Accept (Poster)	The manuscript has been reviewed by four experienced reviewers, among whom three voted for acceptance and one reviewer (vaM6) voted for borderline reject. The borderline reviewer mainly complained about novelty, some missing experiments, as well as the discussion on scalability and efficiency. Regarding the novelty, in...	4	[{"review_id": "kUsf14C9GjH", "reviewer": "Reviewer_mcrT", "summary": "This manuscript proposes a disentangled graph contrastive learning method, which can generate the graph-level disentangled representation. The proposed method contains two main parts: the graph encoder to generate the factors and the factor-wise con...	Disentangled Contrastive Learning on Graphs Haoyang Li', Xin Wang': Ziwei Zhang', Zehuan Yuan', Hang Li,, Wenwu Zhu 'Tsinghua University, *Bytedance lihy18@mails. www.:mmmmils..tinnuuua.aa. edu. en, wang, zazhang)@tainghua. edu. (yuanzehuan, lihang. @bytedanee com, wuzhußtsinghua. edu..edddde...... cn Abstract Recently...	46,210

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.