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
x_JOyw5CLP	neurips	2,021	main	NeurIPS.cc/2021/Conference	10,633	Evaluation of Human-AI Teams for Learned and Rule-Based Agents in Hanabi	Deep reinforcement learning has generated superhuman AI in competitive games such as Go and StarCraft. Can similar learning techniques create a superior AI teammate for human-machine collaborative games? Will humans prefer AI teammates that improve objective team performance or those that improve subjective metrics of ...	[ "Ho Chit Siu", "Jaime Daniel Pena", "Edenna Chen", "Yutai Zhou", "Victor Lopez", "Kyle Palko", "Kimberlee Chestnut Chang", "Ross Emerson Allen" ]	[ "Human-AI Teaming", "Deep Learning", "Reinforcement Learning" ]	NeurIPS 2021 Poster	Accept (Poster)	Reviewers were unanimous that the paper addresses an increasingly important area of research (human-AI collaborations), focuses on an understudied but crucial aspect of the research (i.e., how do humans perceive and assess their AI teammates?), and offers insightful results. While they found the paper well-written over...	4	[{"review_id": "OQhwV-6wL4", "reviewer": "Reviewer_PtgZ", "summary": "This paper conducts a user study in which humans play alongside a rule-based AI bot (SmartBot) and a learning-based AI bot (Other Play) in the cooperative card game Hanabi, and then are asked to rate the bots along several axes. The key finding of in...	Evaluation of Human-AI Teams for Learned and Rule-Based Agents in Hanabi Ho Chit Siu Jaime D. Peña Yutai Zhou Edenna Chen Victor J. Lopez Kyle Palko Kimberlee C. Chang Ross E. Allen Abstract Deep reinforcement learning has generated superhuman Al in competitive games such as Go and StarCraft. Can similar learning techn...	47,241
xV6ZDMwRspN	neurips	2,021	main	NeurIPS.cc/2021/Conference	8,690	Unlabeled Principal Component Analysis	We introduce robust principal component analysis from a data matrix in which the entries of its columns have been corrupted by permutations, termed Unlabeled Principal Component Analysis (UPCA). Using algebraic geometry, we establish that UPCA is a well-defined algebraic problem in the sense that the only matrices of m...	[ "Yunzhen Yao", "Liangzu Peng", "Manolis C. Tsakiris" ]	[ "unlabeled sensing", "linear regression without correspondences", "robust principal component analysis", "algebraic geometry" ]	NeurIPS 2021 Poster	Accept (Poster)	The authors identified an interesting problem about principal component analysis with unknown permutations of the feature in each data sample. They then proposed an algorithm and provided performance guarantee. All 4 reviewers agreed to accept this paper.	4	[{"review_id": "v9HKO2zsBUQ", "reviewer": "Reviewer_azUW", "summary": "The paper introduced a variant of PCA in the presence of noise, i.e., a variant of robust PCA. The \"noise\" in this case is permutations of the column entries of the data matrix. Using concept from Algebraic geometry, the authors showed that they c...	Unlabeled Principal Component Analysis Yunzhen Yao, Liangzu Peng, Manolis C. Tsakiris School of Information Science and Technology ShanghaiTech University Abstract We introduce robust principal component analysis from data matrix which the entries of its columns have been corrupted by permutations, termed Unlabeled Pri...	41,178
xVLzpMOexqo	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,287	Dr Jekyll & Mr Hyde: the strange case of off-policy policy updates	The policy gradient theorem states that the policy should only be updated in states that are visited by the current policy, which leads to insufficient planning in the off-policy states, and thus to convergence to suboptimal policies. We tackle this planning issue by extending the policy gradient theory to policy updat...	[ "Romain Laroche", "Remi Tachet des Combes" ]	[ "policy gradient", "reinforcement learning", "actor-critic", "theory", "algorithm", "deep reinforcement learning", "exploration", "policy updates" ]	NeurIPS 2021 Poster	Accept (Poster)	The reviewers discussed this paper and came to a consensus that it is a useful contribution and should be accepted. There were concerns about the fact that this work is primarily in the tabular setting and novelty relative to existing work. The restriction to the tabular setting is quite limited, in that convergence ...	4	[{"review_id": "mivovYp8rN", "reviewer": "Reviewer_622C", "summary": "This paper studies the effects of off-policy updates for policy gradient methods in RL. The authors show that under certain conditions on the state update distribution and learning rate, exact PG methods will converge to the optimal solution. Based o...	Dr Jekyll and Mr Hyde: The Strange Case of Off-Policy Policy Updates Romain Laroche" Microsoft Research Montréal, Canada Rémi Tachet des Combes" Microsoft Research Montréal, Canada Abstract The policy gradient theorem states that the policy should only be updated in states that are visited by the current policy, which ...	49,607
xfDXF0I_bt	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,941	Locally Valid and Discriminative Prediction Intervals for Deep Learning Models	Crucial for building trust in deep learning models for critical real-world applications is efficient and theoretically sound uncertainty quantification, a task that continues to be challenging. Useful uncertainty information is expected to have two key properties: It should be valid (guaranteeing coverage) and discrimi...	[ "Zhen Lin", "Shubhendu Trivedi", "Jimeng Sun" ]	[ "Deep Learning", "Local Validity", "Conformal Prediction", "Uncertainty Quantification", "Prediction Interval" ]	NeurIPS 2021 Poster	Accept (Poster)	Three reviewers indicated acceptance, two of them with clearly positive scores. Their main arguments in favor of this paper were good methodological contributions, convincing experiments, clarity of motivation and arguments, and a convincing rebuttal that added many details and addressed most points of criticism. On th...	4	[{"review_id": "TSgO2umePU6", "reviewer": "Reviewer_nv6c", "summary": "The authors propose a new method for locally valid (e.g. marginal coverage guarantees) and discriminative prediction* intervals by connecting two ideas: kernel regression and conformal inference. They evaluate their method on a suite of small to med...	Locally Valid and Diseriminative Prediction Intervals for Deep Learning Models Zhen Lin Shubhendu Trivedi MIT University of Illinois at Urbana-Champale Urbana, IL 61801 Cambridge, MA 02139 shubhendu@csail miitttai........................ edu edu Jimeng Sun University of Illinois at Trbana-Champplr Urbana, IL 61801 j jm...	47,884
xfskdMFkuTS	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,473	Meta Internal Learning	Internal learning for single-image generation is a framework, where a generator is trained to produce novel images based on a single image. Since these models are trained on a single image, they are limited in their scale and application. To overcome these issues, we propose a meta-learning approach that enables traini...	[ "Raphael Bensadoun", "Shir Gur", "Tomer Galanti", "Lior Wolf" ]	[ "hypernetworks", "meta learning", "single image generation", "gan", "computer vision" ]	NeurIPS 2021 Poster	Accept (Poster)	Meta-review of "Meta Internal Learning" This paper proposes a framework for single image generation from the perspective of meta-learning on larger datasets. The method uses hypernetworks as the generator and discriminator so that the weights of these networks can be conditionally adapted from the single image from a ...	4	[{"review_id": "WJ7Grt9kv6", "reviewer": "Reviewer_FnNF", "summary": "This paper introduces an advanced version of SinGAN by incorporating the hypernetwork into the framework so that the proposed model can perform all the tasks that the original SinGAN can do, but with much faster test inference. ", "questions": "", "l...	Meta Internal Learning Raphael Bensadoun" The School of Computer Science Tel Aviv University Shir Gur The School of Computer Science Tel Aviv University Tomer Galanti Lior Wolf The School of Computer Science Tel Aviv University The School of Computer Science Tel Aviv University Abstract Internal learning for single-ima...	41,041
xyoFSmocONi	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,136	Structure-Aware Random Fourier Kernel for Graphs	Gaussian Processes (GPs) define distributions over functions and their generalization capabilities depend heavily on the choice of kernels. In this paper, we propose a novel structure-aware random Fourier (SRF) kernel for GPs that brings several benefits when modeling graph-structured data. First, SRF kernel is defined...	[ "Jinyuan Fang", "Qiang Zhang", "Zaiqiao Meng", "Shangsong Liang" ]	[ "Structure-Aware Random Fourier Kernel", "Gaussian Process", "Graph Learning", "Random Fourier Feature" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper received positive reviews and the rebuttal phase clarified some potential issues. Overall, this is a solid contribution, which should be accepted. The final submission should, however, incorporate suggestions made during the rebuttal phase.	4	[{"review_id": "jhVuGEeH5zs", "reviewer": "Reviewer_hdfy", "summary": "This paper proposed a novel SRF kernel defined with spectral distribution based on the Fourier duality given by Bochner's theorem on the vector embedding of subgraphs using GCNN. Based on the SRF kernel, the paper proposes a GP model, GPSRF, for ...	Structure-Aware Random Fourier Kernel for Graphs Jinyuan Fang!?? 1.2, Qiang Zhang2:4.5 Zaiqiao Meng"., Shangsong 1,2,7 Liangl.2,7+ School of Computer Science and Engineering, Sun Yat-sen University, China Guangdong Key Laboratory of Big Data Analysis and Processing, Guangzhou, China Hangzhou Innovation Center, Zhejiang...	54,440
xlNpxfGMTTu	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,598	Adversarial Attack Generation Empowered by Min-Max Optimization	The worst-case training principle that minimizes the maximal adversarial loss, also known as adversarial training (AT), has shown to be a state-of-the-art approach for enhancing adversarial robustness. Nevertheless, min-max optimization beyond the purpose of AT has not been rigorously explored in the adversarial contex...	[ "Jingkang Wang", "Tianyun Zhang", "Sijia Liu", "Pin-Yu Chen", "Jiacen Xu", "Makan Fardad", "Bo Li" ]	[ "Min-max optimization", "Ensemble attack", "Universal perturbation" ]	NeurIPS 2021 Poster	Accept (Poster)	This review results for this paper are borderline. While the reviewers appreciate some of the positive aspects of the paper (such as the formulation and the comprehensive numerical experiments), they have reservations about the novelty and the presentation of parts of the paper. However, regarding the contribution of t...	6	[{"review_id": "xsKu-yiV5tv", "reviewer": "Reviewer_cLBV", "summary": "The paper studies a min-max formulation for adversarial attacks over multiple domains, which covers attacking ensembling attacks, multiple threat models, and universal perturbations. They propose a regularized formulation and show improvements perfo...	Adversarial Attack Generation Empowered by Min-Max Optimization Jingkang Wang .2- Tianyun Zhang Sijia Liu Pin-Ye Yu Chen Jiacen Xu Makan Fardad Li University of Toronto', Vector Institute², Cleveland State University Michigan State University*. MIT-IBM Watson AI Lab, IBM Research University of California, Irvine Syracu...	52,502
zdNEp82a-_q	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,314	Morié Attack (MA): A New Potential Risk of Screen Photos	Images, captured by a camera, play a critical role in training Deep Neural Networks (DNNs). Usually, we assume the images acquired by cameras are consistent with the ones perceived by human eyes. However, due to the different physical mechanisms between human-vision and computer-vision systems, the final perceived imag...	[ "Dantong Niu", "Ruohao Guo", "Yisen Wang" ]	[ "Moiré effect", "Adversarial attack" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper presents a novel kind of attack: Moire attack. It is inspired by that there will be Moire effect when shooting images on the LCD monitors. Although the Moire effect is perceptible by human eyes, it is very hard to distinguish between images with different Moire effects. Moire attack therefore is hard to recog...	4	[{"review_id": "rdflB8zwd8p", "reviewer": "Reviewer_zVAp", "summary": "This paper presents an interesting new deep attack method, which utilizes the morie effect caused by digital image processing. It mimics the shooting process of digital devices and generates the physical-world morie pattern, which is then added to t...	Moiré Attack (MA): A New Potential Risk of Screen Photos Dantong N Ruohao Guo2 Yisen Wang 'Department of EECS, University of California, Berkeley "oolleee of Information and Electrical Engineering. China Agricultural University KKey Lab. Machine Perception, School of Artificial Intelligence. Peking University Innttitte...	38,543
xQGYquca0gB	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,145	Neural Production Systems	Visual environments are structured, consisting of distinct objects or entities. These entities have properties---visible or latent---that determine the manner in which they interact with one another. To partition images into entities, deep-learning researchers have proposed structural inductive biases such as slot-bas...	[ "Anirudh Goyal", "Aniket Rajiv Didolkar", "Nan Rosemary Ke", "Charles Blundell", "Philippe Beaudoin", "Nicolas Heess", "Michael Curtis Mozer", "Yoshua Bengio" ]	[ "production systems", "attention-based sparse entity interactions", "knowledge factorization" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper introduces a differentiable and learnable version of the classic production system architecture. This is a worthwhile and interesting attempt at enabling more flexible reasoning in deep learning systems. The reviewers found the initial evaluations of the system uncompelling, however the substantial additions...	4	[{"review_id": "ubmurL82qW", "reviewer": "Reviewer_CUnV", "summary": "The authors present a new neural network system that can reason over entities. The solution should be applicable to any tasks that require visual reasoning.\n\nSpecifically, the NPS algorithm is:\n```\nfor every step in a sequence:\n update the sl...	Neural Production Systems Aniket Didolkar' Anirudh Goyal ",, Nan Rosemary Ke ? Charles Blundell 2, Philippe Beaudoin 3 Nicolas Heess" 2, Michael Mozer 4. Yoshua Bengio 1, Abstract Visual environments are structured, consisting of distinct objects or entities. These entities have properties-vin they interact with one an...	59,226
xRJ_Xqmb6d	neurips	2,021	main	NeurIPS.cc/2021/Conference	8,020	Perturb-and-max-product: Sampling and learning in discrete energy-based models	Perturb-and-MAP offers an elegant approach to approximately sample from a energy-based model (EBM) by computing the maximum-a-posteriori (MAP) configuration of a perturbed version of the model. Sampling in turn enables learning. However, this line of research has been hindered by the general intractability of the MAP c...	[ "Miguel Lazaro-Gredilla", "Antoine Dedieu", "Dileep George" ]	[ "max-product", "perturb-and-map", "discrete graphical models", "energy based models", "belief revision", "belief propagation" ]	NeurIPS 2021 Poster	Accept (Poster)	The authors present an interesting framework in which they combine the "perturb-and-MAP" strategy for generating samples from a Gibbs distribution with the perspective that incorrect inference (such as substituting the simple and efficient max-product method) can be compensated by learning the model parameters using a ...	4	[{"review_id": "xudvT7lFbZ", "reviewer": "Reviewer_MkBQ", "summary": "This paper proposes to use max-product belief propagation to provide an approximate solution to the MAP problem encountered in the perturb-and-MAP sampling method. The paper argues that, despite its lack of convergence guarantees, max-product is fast...	Perturb-and-max--rroduct Sampling and learning in discrete energy-based models Miguel Láraro-Gredilla, Antoine Dedieu, Dileep George Vicarious Al SF Bay Area, CA {miguel, antoine, d1llee))@ricarioussccmm Abstract Perturb-and-MAP offers an elegant approach to approximately sample from an energy-based model (EBM) by comp...	44,236
xWq1MVj7YrE	neurips	2,021	main	NeurIPS.cc/2021/Conference	442	Novel Visual Category Discovery with Dual Ranking Statistics and Mutual Knowledge Distillation	In this paper, we tackle the problem of novel visual category discovery, i.e., grouping unlabelled images from new classes into different semantic partitions by leveraging a labelled dataset that contains images from other different but relevant categories. This is a more realistic and challenging setting than convent...	[ "Bingchen Zhao", "Kai Han" ]	[ "novel category discovery", "transfer learning", "clustering", "deep learning" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper presents a novel class discovery technique based on dual ranking statistics and mutual knowledge distillation. The main idea is reasonable and shows superior performance compared to existing methods. All reviewers are positive about this paper and there are no particularly negative comments. However, the pro...	4	[{"review_id": "z36c1SmYQB0", "reviewer": "Reviewer_uqsh", "summary": "The paper tackles novel class discovery of unlabeled data via transfer learning-based two-brach learning with pairwise pseudo-labels by two different criteria: similarity on global descriptor and local features. The distinct criteria generate more r...	Novel Visual Category Discovery with Dual Ranking Statistics and Mutual Knowledge Distillation Bingchen Zhao Kai Han2.3,4. TToggi University The University of Hong Kong zhaobc, wwweeeigmmmmm.mmmmmmmm Google Research UUniversity of Bristol com com kaihanx0hku, hk Abstract In this paper, we tackle the problem of novel vi...	47,829
xVZx1SXb_IU	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,289	Robust Allocations with Diversity Constraints	We consider the problem of allocating divisible items among multiple agents, and consider the setting where any agent is allowed to introduce {\emph diversity constraints} on the items they are allocated. We motivate this via settings where the items themselves correspond to user ad slots or task workers with attribute...	[ "Zeyu Shen", "Lodewijk L. Gelauff", "Ashish Goel", "Aleksandra Korolova", "Kamesh Munagala" ]	[ "Resource allocation", "algorithmic fairness", "online advertising", "Nash Welfare", "social welfare" ]	NeurIPS 2021 Poster	Accept (Poster)	Overall the reviewers are quite positive about this paper: every reviewer thought the paper studies an interesting topic, and the results are interesting. A weakness of the paper is that it does not quite fully explore the space. This makes it hard to really compare the different allocation rules, since various sub...	4	[{"review_id": "x9a3Sc5ovz-", "reviewer": "Reviewer_gMgx", "summary": "This paper proposes a framework for and and analysis of algorithms that allocate divisible resources to a set of agents when the agents can specify both utilities as well as constraints over the items as well. This is motivated by, e.g., computation...	Robust Allocations with Diversity Constraints Zeyu Shen Duke University Durham NC 27708-0129 zeyu. shen.ukkee edu Lodewijk Gelauff Ashish Goel Management Science and Engineering Stanford University, Stanford CA 94305 [lodewi jk, ashish ord. edu Aleksandra Korolova Department of Computer Science University of Southern C...	47,586
xVs5d5ZSWaa	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,200	Learning Diverse Policies in MOBA Games via Macro-Goals	Recently, many researchers have made successful progress in building the AI systems for MOBA-game-playing with deep reinforcement learning, such as on Dota 2 and Honor of Kings. Even though these AI systems have achieved or even exceeded human-level performance, they still suffer from the lack of policy diversity. In t...	[ "Yiming Gao", "Bei Shi", "Xueying Du", "Liang Wang", "Guangwei Chen", "Zhenjie Lian", "Fuhao Qiu", "GUOAN HAN", "Weixuan Wang", "Deheng Ye", "QIANG FU", "Yang Wei", "Lanxiao Huang" ]	[ "Deep Reinforcement Learning", "Diverse Policies", "Game Playing", "Goal-based Learning" ]	NeurIPS 2021 Poster	Accept (Poster)	After reading the reviews, authors response and discussions, I suggest to accept the paper. The ethical concerns have been answered and the authors took action to conform to what was required by the ethical review. Questions and some concerns on the method have been answered by the authors during the rebuttal and one r...	4	[{"review_id": "fbu_qADg5Y7", "reviewer": "Reviewer_y4xZ", "summary": "The paper addresses RL agent training in MOBA games, specifically the ability to acquire a diverse set of policies when playing a variety of strategic scenarios (team compositions, aka lineups). Human example data is processes to extract future stat...	Learning Diverse Policies in MOBA Games via Macro-Goals Yiming Gao' Bei Shi! Xueying Du' Liang Wang' Guangwei Chen' Zhenjie Lian' Fuhao Qiu Guoan Han' Weixuan Wang' Deheng Ye! Qiang Fu² Wei Yang' Lanxiao Huang? Tencent AI Lab, Shenzhen, China Tencent TiMi LI Studio, Chengdu, China (yatminggao, beishi, sheri enginewang ...	42,020
xRLT28nnlFV	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,359	On Robust Optimal Transport: Computational Complexity and Barycenter Computation	We consider robust variants of the standard optimal transport, named robust optimal transport, where marginal constraints are relaxed via Kullback-Leibler divergence. We show that Sinkhorn-based algorithms can approximate the optimal cost of robust optimal transport in $\widetilde{\mathcal{O}}(\frac{n^2}{\varepsilon})$...	[ "Khang Le", "Huy Nguyen", "Quang Minh Nguyen", "Tung Pham", "Hung Bui", "Nhat Ho" ]	[ "optimal transport", "optimization", "complexity", "robustness" ]	NeurIPS 2021 Poster	Accept (Poster)	Most of the reviewers and the AC agree that the submission makes a worthwhile theoretical contribution in designing and analyzing new algorithms for RSOT and RIBP. The reviewers highlight the following main concerns: - The algorithm appears to be a simple normalized version of the Sinkhorn algorithm for UOT. - the t...	5	[{"review_id": "yJySXhXg1xe", "reviewer": "Reviewer_yb1x", "summary": "This paper presents the complexity analysis of two problems: robust optimal transport and robust barycenter problem. For both these problems, the paper analyzes a sinkhorn-based algorithm and obtains a complexity of O(n^2 / \\epsilon). Finally, the ...	On Robust Optimal Transport: Computational Complexity and Barycenter Computation Khang Le* Huy Nguyen* Quang Minh Nguyen University of Texas, Austin VinAI Research Massachusetts Institute of Technology khanglnt@utexas.cem edu Muyym12@vinee iio nmquang@mit.cc Tung Pham Hung Bui Nhat Ho VinAI Research VinAI Research Univ...	38,946
x6tV8QhHjs1	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,676	Online Learning and Control of Complex Dynamical Systems from Sensory Input	Identifying an effective model of a dynamical system from sensory data and using it for future state prediction and control is challenging. Recent data-driven algorithms based on Koopman theory are a promising approach to this problem, but they typically never update the model once it has been identified from a relativ...	[ "Oumayma Bounou", "Jean Ponce", "Justin Carpentier" ]	[ "Robotics and Control", "Computer Vision" ]	NeurIPS 2021 Poster	Accept (Poster)	From the SAC. This is an instance where the rebuttal and the discussion worked. While the original decision for this paper was to not accept, it is being raised to a recommended accept. The primary reason is the quality of the rebuttal, and the useful technical discussion between authors and reviewers that ensued that ...	3	[{"review_id": "lKiMJ7bu_YO", "reviewer": "Reviewer_9oAb", "summary": "This work proposed an approach of online learning and control of nonlinear systems relying on the insights from Koopman operator theory, and included empirical examples.\n", "questions": "", "limitations": "", "rating": 6, "confidence": 5, "soundnes...	Online Learning and Control of Dynamical Systems from Sensory Input Oumayma Bounou', Jean Ponce!:2, and Justin Carpentier' Inria and Département d' Informatiiuu de I'Ecole Normale Supérieure, PSL Research University CCnter for Data Science, New York University /oumayma.bounou, jean pennc, custin.carpentier) @ @inria..r...	45,707
x8qirBbT9xp	neurips	2,021	main	NeurIPS.cc/2021/Conference	8,104	Causal Inference for Event Pairs in Multivariate Point Processes	Causal inference and discovery from observational data has been extensively studied across multiple fields. However, most prior work has focused on independent and identically distributed (i.i.d.) data. In this paper, we propose a formalization for causal inference between pairs of event variables in multivariate recur...	[ "Tian Gao", "Dharmashankar Subramanian", "Debarun Bhattacharjya", "Xiao Shou", "Nicholas Mattei", "Kristin Bennett" ]	[ "causal inference", "point process" ]	NeurIPS 2021 Poster	Accept (Poster)	Causal inference in point processes is a novel and open area, and the reviewers agreed that this is a useful contribution both methodologically and from an applied perspective. There are a number of changes that you should definitely make to improve the manuscript.	4	[{"review_id": "k8lRBMdMIU-", "reviewer": "Reviewer_CcPK", "summary": "This paper proposes a model for estimating the average causal effect in point processes where the estimand of interest is the effect of an event occurring within a window on the rate parameter of a point process. The authors use the formulation comm...	Causal Inference for Event Pairs in Multivariate Point Processes Tian Gao IBM Research tgaouus. ibm. com Dharmashankar Subramanian IBM Research dharmash@us, ibm. com Debarun Bhattacharjya IBM Research debarunb@us .iaiass.bbme imm. com Xíao Shou RPI shoux@rpi.ede edu Nicholas Mattei Tulane University nsmattei@tul ane. e...	51,636
xRrdX_wV1JI	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,682	Generalized DataWeighting via Class-Level Gradient Manipulation	Label noise and class imbalance are two major issues coexisting in real-world datasets. To alleviate the two issues, state-of-the-art methods reweight each instance by leveraging a small amount of clean and unbiased data. Yet, these methods overlook class-level information within each instance, which can be further uti...	[ "Can Chen", "Shuhao Zheng", "Xi Chen", "Erqun Dong", "Xue Liu", "Hao Liu", "Dejing Dou" ]	[ "instance weighting", "meta-learning", "bi-level optimization", "classification", "chain rule" ]	NeurIPS 2021 Poster	Accept (Poster)	There is a strong disagreement with this paper, with scores ranging as 8,7,5, and 4. The reviewer who provided a 4 said she increased her score to 5 but this has not propagated. In this case, the rating would be 6.25 rather than 6. While the accept reviews are stronger score wise, the reasons to reject (marginal impro...	4	[{"review_id": "PJ0LroomLN", "reviewer": "Reviewer_WWma", "summary": "The paper presents a method called Generalized Data Weighting (GDW) to mitigate label noise and class imbalance problems simultaneously. GDW works by re-weighting instances at the class level and it leverages meta-learning to learn the optimal weight...	Generalized Data Weighting via Class-level Gradient Manipulation Can Chen'; Shuhao Zheng'; Xi Chen', Erqun Dong', Xue Liu', Hao Liu', Dejing Dou 'McGill University, 'The Hong Kong University of Science and Technology, 'Baidu Research (can.chen, shuhao zzeng, erqun (rqqundoong)eee dong) @mail ciciiilmmmllll. worgill.co ...	46,007
x4t0fxWPNdi	neurips	2,021	main	NeurIPS.cc/2021/Conference	5,290	Implicit Transformer Network for Screen Content Image Continuous Super-Resolution	Nowadays, there is an explosive growth of screen contents due to the wide application of screen sharing, remote cooperation, and online education. To match the limited terminal bandwidth, high-resolution (HR) screen contents may be downsampled and compressed. At the receiver side, the super-resolution (SR)of low-reso...	[ "Jingyu Yang", "Sheng Shen", "Huanjing Yue", "Kun Li" ]	[ "Implicit fuction", "Transformer", "Continuous super-resolution", "Screen content images" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper proposes a new model for image super-resolution, which builds on an implicit transformer, to address the resolution enhancement problem in the specific settings of screen-content images. The performance of the model has been extensively studied by the other and improved through the active discussion with the ...	5	[{"review_id": "qVEStyDxcio", "reviewer": "Reviewer_2ePa", "summary": "This work addresses the SCISR problem by combining transformers to the network architecture. The backbone of the architecture utilizes a CNN to extract feature maps, to enlarge the receptive field, an implicit transformer is used. After that the imp...	Implicit Transformer Network for Screen Content Image Continuous Super-Resolution Jingyu Yang' Sheng Shen' Huanjing Yue!" Kun Li School of Electrical and Information Engineering, Tianjin University College of Intelligence and Computing, Tianjin University {yjy, codyshen, , muanjingg yue, yue, 1ik]@tju. edu. https://git...	41,456
xN3XX6pKSD5	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,658	Deep Marching Tetrahedra: a Hybrid Representation for High-Resolution 3D Shape Synthesis	We introduce DMTet, a deep 3D conditional generative model that can synthesize high-resolution 3D shapes using simple user guides such as coarse voxels. It marries the merits of implicit and explicit 3D representations by leveraging a novel hybrid 3D representation. Compared to the current implicit approaches, which ar...	[ "Tianchang Shen", "Jun Gao", "Kangxue Yin", "Ming-Yu Liu", "Sanja Fidler" ]	[ "3D Deep Learning", "3D Super-Resolution", "3D Content Creation", "3D Shape Representation" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper received mostly positive reviews, with the exception of reviewer f9sZ. Taking into account feedback from the other reviewers as well as the message to the area chair, the AC tends to agree that f9sZ's concerns are not sufficient to hold back publishing this work. The final revision of this paper should cla...	4	[{"review_id": "y2VAU-64mC", "reviewer": "Reviewer_f9sZ", "summary": "This paper introduces a shape refinement approach that first uses an implicit representation to obtain an initial shape. The initial shape is then converted into a mesh representation refined by loop subdivision. ", "questions": "", "limitations": ""...	Deep Marching Tetrahedra: a Hybrid Representation for High-Resolution 3D Shape Synthesis Tianchang Shen 1,2,3 Jun Gao 1,2,3 Kangxue Yin Ming- Yu Liu Sanja Fid NVIDIA University of Toronto Vector Institute" {frshon, jung, kangxoey, ningyul, sfidler)dnvidia.com Abstract We introduce DMTET, deep 3D conditional generative ...	54,662
zAuDbrHC6fq	neurips	2,021	main	NeurIPS.cc/2021/Conference	555	Aligned Structured Sparsity Learning for Efficient Image Super-Resolution	Lightweight image super-resolution (SR) networks have obtained promising results with moderate model size. Many SR methods have focused on designing lightweight architectures, which neglect to further reduce the redundancy of network parameters. On the other hand, model compression techniques, like neural architecture ...	[ "Yulun Zhang", "Huan Wang", "Can Qin", "Yun Fu" ]	[ "Neural Network Pruning", "Lightweight Image Super-Resolution", "Aligned Structured Sparsity Learning" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	This paper studies the issue of pruning for super-resolution networks. The reviewers agree that the approach to pruning architecture-specific components, like resnet blocks, offers some interesting advances. However the authors seem to agree that the paper could do a better job communicating the novelty of the propos...	3	[{"review_id": "ukXbw0A67jM", "reviewer": "Reviewer_k8sL", "summary": "The paper proposes a novel aligned structured sparsity learning (ASSL) method to prune the SR network. To tackle the pruned filter location mismatch issue in SR networks, a sparsity structure alignment penalty term is introduced to align the pruned ...	Aligned Structured Sparsity Learning for Efficient Image Super-Resolution Yulun Zhanglt Huan Wang 1t+ Can Qin' Yun Fu DDpaarmentt of ECE, Northeastern University Khoury College of Computer Science, Northeastern University Abstract Lightweight image super-resolution (SR) networks have obtained promising sults with moder...	48,422
yGKi6deX8bX	neurips	2,021	main	NeurIPS.cc/2021/Conference	10,672	Towards understanding retrosynthesis by energy-based models	Retrosynthesis is the process of identifying a set of reactants to synthesize a target molecule. It is of vital importance to material design and drug discovery. Existing machine learning approaches based on language models and graph neural networks have achieved encouraging results. However, the inner connections of t...	[ "Ruoxi Sun", "Hanjun Dai", "Li Li", "Steven Kearnes", "Bo Dai" ]	[ "chemical application", "energy-based model" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper tries to include all the existing retrosynthesis algorithms into the framework of Energy-Based Model (EBM). Based the this, it derives a dual model to measure the energy of a pair (Reactants, Product), which depends on both the backward, forward energy function and their consistency. Experiments demonstrate ...	4	[{"review_id": "BqgxR-fC0XX", "reviewer": "Reviewer_r7dk", "summary": "This paper presents a unified framework with energy-based models for retrosynthesis prediction. Different methods (sequence and graph methods) can be untied with different energy functions within the framework. A dual variant model is presented and ...	Towards understanding retrosynthesis by energy-based models Ruoxi Sun', Hanjun Dai2, Li Li Steven Kearnes', and Bo Dai² Google Cloud AI Google Brain 'Google Research (ruoxis, hadai, leeley, kearnes, bodai) @google.com Abstract Retrosynthesis is the process of identifying set of reactants to synthesize a target molecule...	30,364
xAFm5knU7Nc	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,031	Accelerating Quadratic Optimization with Reinforcement Learning	First-order methods for quadratic optimization such as OSQP are widely used for large-scale machine learning and embedded optimal control, where many related problems must be rapidly solved. These methods face two persistent challenges: manual hyperparameter tuning and convergence time to high-accuracy solutions. To ad...	[ "Jeffrey Ichnowski", "Paras Jain", "Bartolomeo Stellato", "Goran Banjac", "Michael Luo", "Francesco Borrelli", "Joseph E. Gonzalez", "Ion Stoica", "Ken Goldberg" ]	[ "quadratic optimization", "convex optimization", "first-order methods", "reinforcement learning for optimization" ]	NeurIPS 2021 Poster	Accept (Poster)	This is a difficult one to assess. There were some strong opinions among the reviewers, and one reviewer that did seem a bit excessively harsh - perhaps expecting the paper to have the same sort of rigor one might find in a pure optimization paper, rather than an RL application paper. Evaluated as an RL application pa...	4	[{"review_id": "qHo52ZDsc17", "reviewer": "Reviewer_riVf", "summary": "This paper aims to speed up solving of QPs by using RL to learn adaptive parameters of the OSQP solver. The authors develop and implement an RL algorithm based on TD3 to obtain optimal parameters for OSQP that significantly speed up solution times f...	Accelerating Quadratic Optimization with Reinforcement Learning Jeffrey Ichnowski) Paras Jain 11 Bartolomeo Stellato', Goran Banjac', Michael Luo', Francesco Borrelli' Joseph E. Gonzalez', Ion Stoica', and Ken Goldberg' University of Califormia, Berkeley, Princetoon University, 'ETH Zürich Correspondence to: {jeffi, ...	44,076
x4zs7eC-BsI	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,201	Improving Deep Learning Interpretability by Saliency Guided Training	Saliency methods have been widely used to highlight important input features in model predictions. Most existing methods use backpropagation on a modified gradient function to generate saliency maps. Thus, noisy gradients can result in unfaithful feature attributions. In this paper, we tackle this issue and introduce a...	[ "Aya Abdelsalam Ismail", "Hector Corrada Bravo", "Soheil Feizi" ]	[ "Interpretability" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper proposes a new procedure for training neural networks that achieves improved interpretability by masking the bottom k input gradients, helping address the noise in saliency map techniques. The reviewers found the paper well motivated, liked the experiments (especially the fact that they’re conducted on many d...	4	[{"review_id": "hsG2oSfBDmA", "reviewer": "Reviewer_jiTu", "summary": "The authors propose an interpretable training procedure, in which the input features with the bottom k input gradients are masked. The loss term is comprised of the original task loss, plus a KL divergence between the model outputs on the original a...	Improving Deep Learning Interpretability by Saliency Guided Training Aya Abdelsalam Ismail, Soheil Feizi Héctor Corrada Bravo a aasalam, sfeizi miiii]p @css, umd. edu, corradaht @gne. com Department of Computer Science, University of Maryland Data Science and Statistical Computing, Genentech, Inc. Abstract Saliency met...	45,476
x_n34KpwAvI	neurips	2,021	main	NeurIPS.cc/2021/Conference	11,322	SIMILAR: Submodular Information Measures Based Active Learning In Realistic Scenarios	Active learning has proven to be useful for minimizing labeling costs by selecting the most informative samples. However, existing active learning methods do not work well in realistic scenarios such as imbalance or rare classes,out-of-distribution data in the unlabeled set, and redundancy. In this w...	[ "Suraj Nandkishor Kothawade", "Nathan Alexander Beck", "Krishnateja Killamsetty", "Rishabh K Iyer" ]	[ "Submodularity", "Active Learning", "Information Measures", "Realistic", "Redundancy", "Rare Classes", "Out of distribution", "Robust" ]	NeurIPS 2021 Poster	Accept (Poster)	The work proposes a novel active learning framework based on submodular information gain maeasures. The framework is well motivated and explained, and then empirically evaluated. This submission was well appreciated by the reviewers. The authors are encouraged to take the reviewer's comments and their responses into a...	4	[{"review_id": "qWfktrqzTJB", "reviewer": "Reviewer_hRHU", "summary": "The paper introduces a novel active learning framework (called SIMILAR) that leverages Submodular Information Measures to deal with both \"mainstream\" active learning scenarios and situations in which there are rare classes, redundancy, or out-of-d...	SIMILAR: Submodular Information Measures Based Active Learning In Realistic Scenarios Suraj Kothawade University of Texas at Dallas suraj kothawade@utdallas edu Nathan Beck University of Texas at Dallas nathan. beckkudaalaa edu Krishnateja Killamsetty University Texas at Dallas krishnate killamsttty@uttallllla edu Rish...	51,849
xB4lGVLvXDz	neurips	2,021	main	NeurIPS.cc/2021/Conference	985	Reliable Estimation of KL Divergence using a Discriminator in Reproducing Kernel Hilbert Space	Estimating Kullback–Leibler (KL) divergence from samples of two distributions is essential in many machine learning problems. Variational methods using neural network discriminator have been proposed to achieve this task in a scalable manner. However, we noticed that most of these methods using neural network discrimin...	[ "Sandesh Ghimire", "Aria Masoomi", "Jennifer Dy" ]	[ "RKHS", "KL Divergence", "Sample Complexity", "Statistical Learning Theory", "Kernel Methods", "Bayesian inference", "Mutual Information", "Reliable estimation" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	Converging reviews regarding the acceptance of the paper after a fruitful discussion between reviewers and authors, where the latter took time to clarify all the questions raised. The recommendation for this paper is to accept it, counting on the authors to take into account the outcomes of the discussions with all the...	3	[{"review_id": "V8CGbQ_zAdd", "reviewer": "Reviewer_kqvB", "summary": "This papers studies the problem of estimating KL divergence from samples via variational methods with neural network discriminators. They propose a novel way to construct the discriminator in some RKHS and prove the consistency of the resulting esti...	Reliable Estimation of KL Divergence using a Discriminator in Reproducing Kernel Hilbert Space Sandesh Ghimire, , Aria Masoomi, Jennifer Dy Department of Electrical and Computer Engineering Northeastern University sandesh@ece.c neu. edu, masoomi.. adnoorchaastter..edd, jdyβece. neu. edu Abstract Estimating Kullback-Lei...	44,891
x6z8J_17LP3	neurips	2,021	main	NeurIPS.cc/2021/Conference	8,518	Lip to Speech Synthesis with Visual Context Attentional GAN	In this paper, we propose a novel lip-to-speech generative adversarial network, Visual Context Attentional GAN (VCA-GAN), which can jointly model local and global lip movements during speech synthesis. Specifically, the proposed VCA-GAN synthesizes the speech from local lip visual features by finding a mapping function...	[ "Minsu Kim", "Joanna Hong", "Yong Man Ro" ]	[ "video driven speech synthesis", "speech reconstruction from silent video", "lip reading", "audio-visual attention", "context attentional gan" ]	NeurIPS 2021 Poster	Accept (Poster)	The authors investigate speech generation out of silent videos based on the so-called visual context attentional GAN. The problem is interesting and challenging. The authors propose to use global visual context to reduce the ambiguity when mapping visemes to phonemes and introduce a synchronization mechanism via cont...	4	[{"review_id": "s0amb7oyOxU", "reviewer": "Reviewer_Nvn1", "summary": "This paper describes a synthesis approach that uses visual lip movement as input and generates speech as output. They use a GAN style synthesizer. The most novel contribution is the synchronization technique to align visual and speech signals.", "q...	Lip to Speech Synthesis with Visual Context Attentional GAN Minsu Kim, Joanna Hong, Yong Man Ro Image and Video Systems Lab KAIST {ms. k, joanna2587, ymro) Ckaist.aaat..r ac. kr Abstract In this paper, we propose a novel lip-to-speech generative adversarial network, Visual Context Attentional GAN (VCA-GAN), which can j...	47,205
x4oe1W8Hpl3	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,597	MOMA: Multi-Object Multi-Actor Activity Parsing	Complex activities often involve multiple humans utilizing different objects to complete actions (e.g., in healthcare settings, physicians, nurses, and patients interact with each other and various medical devices). Recognizing activities poses a challenge that requires a detailed understanding of actors' roles, object...	[ "Zelun Luo", "Wanze Xie", "Siddharth Kapoor", "Yiyun Liang", "Michael Cooper", "Juan Carlos Niebles", "Ehsan Adeli", "L. Fei-Fei" ]	[ "activity recognition", "video understanding", "human-object interaction", "temporal action detection" ]	NeurIPS 2021 Poster	Accept (Poster)	The authors introduce action hypergraph and an associated task (activity parsing), as well as a new dataset (MOMA) for complex activity recognition. All of the reviewers are positive, and find the representation has certain novelty and the dataset will be valuable to the community. The AC concurs that this paper is a...	4	[{"review_id": "V5LRGWCKLnp", "reviewer": "Reviewer_PT1W", "summary": "This paper proposes that video understanding be approached by trying to recognize at several different levels of granularity, simultaneously. At the highest temporal granularity is the \"activity\", which consists of a sequence of temporally-localiz...	MOMA: Multi-Objeet Multi-Actor Activity Parsing Zelun Luo; Wanze Xie; Siddharth Kapoor, Yiyun Liang, Michael Cooper, Juan Carlos Niebles, Ehsan Adeli, Fei-Fei Stanford University (alenziuo, wanzexie, siddkap, isaliang, cocpesm), jniebles, eadeli, https:/ http::/mmmaa.aatttt.oommm Dining service Activity Lidiy Take the ...	61,148
x8k1nAoGu1U	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,296	Fast Doubly-Adaptive MCMC to Estimate the Gibbs Partition Function with Weak Mixing Time Bounds	We present a novel method for reducing the computational complexity of rigorously estimating the partition functions of Gibbs (or Boltzmann) distributions, which arise ubiquitously in probabilistic graphical models. A major obstacle to applying the Gibbs distribution in practice is the need to estimate their partition ...	[ "Shahrzad Haddadan", "Yue Zhuang", "Cyrus Cousins", "Eli Upfal" ]	[ "MCMC", "Gibbs distribution", "partition functions" ]	NeurIPS 2021 Poster	Accept (Poster)	The goal of this paper is a better algorithm to estimate the partition function / normalization constant in a Gibbs distribution. These are often of interest in many different forms in the NeurIPS community. The best current methods are based on annealing, essentially starting at a normalized distribution and moving th...	4	[{"review_id": "zc85nLVmZMC", "reviewer": "Reviewer_YZ7S", "summary": "The authors propose a method for estimating the partition function of Gibbs distributions. The method primarily relies on an adaptive mean estimator, and is shown to have theoretical and practical advantages over previous methods.", "questions": "",...	Fast Doubly-Adaptive MCMC to Estimate the Gibbs Partition Function with Weak Mixing Time Bounds Shahrzad Haddadan Brown University The Data Science Initiative shahrzad haddadan@@maal......... com Yue Zhuang Cyrus Cousins r Brown University Brown University The Data Science Initiative yue_ zhuangl @brown. edu Department...	49,359
xXYjxli-2i	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,640	Optimization-Based Algebraic Multigrid Coarsening Using Reinforcement Learning	Large sparse linear systems of equations are ubiquitous in science and engineering, such as those arising from discretizations of partial differential equations. Algebraic multigrid (AMG) methods are one of the most common methods of solving such linear systems, with an extensive body of underlying mathematical theory....	[ "Ali Taghibakhshi", "Scott MacLachlan", "Luke Olson", "Matthew West" ]	[ "Algebraic Multigrid", "Reinforcement Learning", "Graph Partitioning" ]	NeurIPS 2021 Poster	Accept (Poster)	## Original AC meta-review This paper proposes a systematic way of computing optimal coarsening for algebraic multigrid (AMG) algorithms by using a) connection of this problem to graph algorithm b) formulation of the optimal selection as a game c) solution of the RL problem using graph neural networks. The resulting a...	4	[{"review_id": "v3odeNuFb0J", "reviewer": "Reviewer_hVW8", "summary": "The authors propose using RL agent via a TAGCOnv network in DQN to select coarse points in the procedure of coarsening algebraic grids. Theoretical results are established for the integration of the RL into a convergent scheme, and the complexity of...	Optimization-Based Algebraic Multigrid Coarsening Using Reinforcement Learning Ali Taghibakhshi Scott MacLachlan Mechanical Science and Engineering University of Illinois at Arbana-Champaal Urbana, IL 61801, USA Mathematics and Statistics Memorial University of Newfoundland and Labrador alit2@illinois edu John's 's NI,...	40,485
x8gM-4nFq9b	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,182	Fighting Gradients with Gradients: Dynamic Defenses against Adversarial Attacks	Adversarial attacks optimize against models to defeat defenses. Existing defenses are static, and stay the same once trained, even while attacks change. We argue that models should ﬁght back, and optimize their defenses against attacks at test time. We propose dynamic defenses, to adapt the model and input during testi...	[ "Dequan Wang", "An Ju", "Evan Shelhamer", "David Wagner", "Trevor Darrell" ]	[ "adversarial", "robustness", "defense", "dynamic" ]	NeurIPS 2021 Submitted	Reject	The paper proposed to dynamically modify the model and the input images during the test time (named "defensive entropy minimization"). While the idea is interesting, the true novelty needs to be carefully clarified given many missing related papers, because the test-time defense is in fact not new but one of the two ma...	5	[{"review_id": "oB0W57Yr9KK", "reviewer": "Reviewer_jWma", "summary": "The paper proposes a novel test-time dynamic defense method, which adapts the model and input during testing by defensive entropy minimization (named dent). Given an input, the proposed method minimizes the entropy of the model prediction to adapt t...	Fighting Gradients with Gradients: Dynamic Defenses against Adversarial Attacks Anonymous Author(s) Affiliation Address email Abstract Adversarial attacks optimize against models to defeat defenses. Existing defenses are static, and stay the same once trained, even while attacks change. We argue that models should figh...	45,202
xJYek6zantM	neurips	2,021	main	NeurIPS.cc/2021/Conference	9,621	Tighter Expected Generalization Error Bounds via Wasserstein Distance	This work presents several expected generalization error bounds based on the Wasserstein distance. More specifically, it introduces full-dataset, single-letter, and random-subset bounds, and their analogous in the randomized subsample setting from Steinke and Zakynthinou [1]. Moreover, when the loss function is bounded...	[ "Borja Rodríguez Gálvez", "German Bassi", "Ragnar Thobaben", "Mikael Skoglund" ]	[ "generalization error", "wasserstein distance" ]	NeurIPS 2021 Poster	Accept (Poster)	Overall, the reviewers were positive about the paper. My main concern is regarding its novelty: it seems that this paper combines Wang et al. 2019 with the individual sample approach proposed by Bu et al., 2020 and with the conditional mutual information approach proposed by Steinke and Zakynthinou, 2020, respectively....	4	[{"review_id": "tC63ISQa_zS", "reviewer": "Reviewer_SCNA", "summary": "In this paper new generalization error bounds based on Wasserstein distance are derived. As shown in the paper, these bounds are tighter than existing bounds based on mutual information. Also, the authors briefly discuss how their bounds could be ex...	Tighter Expected Generalization Error Bounds via Wasserstein Distance Borja Rodriguez-Gälver Germán Bassi Ericsson Research KTH Royal Institute of Technology Stockholm, Sweden Stockholm, Sweden german. wws.baasssssssee weweeeeeeeiiccceeeeeeeeeeeeeeeee com Ragnar Thobaben Mikael Skoglund KTH Royal Institute Technology S...	42,951
x5hh6N9bUUb	neurips	2,021	main	NeurIPS.cc/2021/Conference	10,532	Stochastic Solutions for Linear Inverse Problems using the Prior Implicit in a Denoiser	Deep neural networks have provided state-of-the-art solutions for problems such as image denoising, which implicitly rely on a prior probability model of natural images. Two recent lines of work – Denoising Score Matching and Plug-and-Play – propose methodologies for drawing samples from this implicit prior and using i...	[ "Zahra Kadkhodaie", "Eero P Simoncelli" ]	[ "inverse problems", "image priors", "unsupervised learning", "semi-supervised learning", "deblurring", "superresolution", "compressive sensing" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper provides an original way to solve linear inverse problems when given a denoiser. The method uses forward calls to the denoiser in a way that leverages the signal prior implicit to it. The paper establishes strong quantitative and qualitative recovery performance. In the camera ready version of the paper, ...	4	[{"review_id": "q07hpzwJ_id", "reviewer": "Reviewer_9hAm", "summary": "The paper derives an algorithm for solving linear inverse problems using the prior that is implicit in an end-to-end denoiser in the form of a CNN.\n\nConsider a denoising algorithm in form of a CNN that takes a noisy observation $y = x+z$ as input ...	Stochastic Solutions for Linear Inverse Problems using the Prior Implicit in a Denoiser Zahra Kadkhodaie Center for Data Science New York University zk388@nyu. edu Eero Simoncelli Center for Neural Science, and Courant Inst. Mathematical Sciences, New York University Flatiron Institute, Simons Foundation eero. wwm.ocel...	46,244
x00mCNwbH8Q	neurips	2,021	main	NeurIPS.cc/2021/Conference	5,403	Twice regularized MDPs and the equivalence between robustness and regularization	Robust Markov decision processes (MDPs) aim to handle changing or partially known system dynamics. To solve them, one typically resorts to robust optimization methods. However, this significantly increases computational complexity and limits scalability in both learning and planning. On the other hand, regularized MDPs...	[ "Esther Derman", "Matthieu Geist", "Shie Mannor" ]	[ "reinforcement learning", "robust MDPs" ]	NeurIPS 2021 Poster	Accept (Poster)	All but one reviewer recommend publication. The one who does not is concerned about the lack of empirical results, but I think this is a topic of interest to the community and its reasonable to have a purely theoretical paper on this.	4	[{"review_id": "uZC2TpeS_u1", "reviewer": "Reviewer_NyMe", "summary": "The paper highlights two limitations of current approaches for Robust RL: \n\n1. Robust optimization methods are computationally demanding.\n2. They do not account for uncertainty in the model dynamics.\n\nThe paper proposes to work around these lim...	Twice regularized MDPs and the equivalence between robustness and regularization Esther Derman' Technion Matthieu Geist Google Research, Brain Team Shie Mannor Technion, NVIDIA Research Abstract Robust Markov decision processes (MDPs) aim to handle changing or partially known system dynamics. To solve them, one typical...	50,710
x9jS8pX3dkx	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,298	Gradient Inversion with Generative Image Prior	Federated Learning (FL) is a distributed learning framework, in which the local data never leaves clients’ devices to preserve privacy, and the server trains models on the data via accessing only the gradients of those local data. Without further privacy mechanisms such as differential privacy, this leaves the system v...	[ "Jinwoo Jeon", "Jaechang Kim", "Kangwook Lee", "Sewoong Oh", "Jungseul Ok" ]	[ "Federated Learning", "Privacy Leakage", "Gradient Inversion", "Generative Model" ]	NeurIPS 2021 Poster	Accept (Poster)	The reviewers all agreed that this is an interesting paper and a worthy improvement over existing gradient inversion attacks against federated learning. I agree with the reviewers: the paper is clearly written and the many experimental results are worthy of publication. The paper will, at the very least, serve as an im...	4	[{"review_id": "H0kGzOMDsYS", "reviewer": "Reviewer_ofcZ", "summary": "The paper considers the interesting Federated Learning problem of recovering latent code data using gradient information, in the presence of additional prior knowledge and without. For an unknown dataset $\\{(x_i,y_i)\\}_{i=1}^B$, where $y_i$ is the...	Gradient Inversion with Generative Image Prior Jinwoo Jeon',, Jaechang Kim?*, Kangwook Lee", Sewoong Oh', Jungseul Ok!.2 Department of Computer Science Engineering, Pohang University of Science and Technology Graduate School of Artificial Intelligence. Pohang University of Science and Technology 3 Department of Electri...	37,429
wxBGz3ScBBo	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,959	Analytic Study of Families of Spurious Minima in Two-Layer ReLU Neural Networks: A Tale of Symmetry II	We study the optimization problem associated with fitting two-layer ReLU neural networks with respect to the squared loss, where labels are generated by a target network. We make use of the rich symmetry structure to develop a novel set of tools for studying families of spurious minima. In contrast to existing approach...	[ "Yossi Arjevani", "Michael Field" ]	[ "Neural networks", "optimization", "symmetry", "symmetry breaking", "spurious minima", "bad local minima", "nonconvex", "deep learning" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper gives a detailed characterization of spurious local minima for 2-layer ReLU networks, which was a model that received substantial theoretical interest. The paper was able to do this using a powerful technique of symmetry-breaking, and the results explained some of the observations in previous works. Overall ...	4	[{"review_id": "lbJ-v-S_yMS", "reviewer": "Reviewer_7HVf", "summary": "The paper theoretically study a one hidden layer neural network with ReLU activation trained using Gaussian data, and analytically derive the values of the different classes of local minima of this objective, and the spectrum of the Hessian at these...	Analytic Study of Families of Spurious Minima in Two-Layer ReLU Neural Networks: A Tale of Symmetry II Yossi Arjevani Michael Field UC Santa Barbara mikefield0gmail.coomm The Hebrew University yossi. evani @gmail.com com Abstract We study the optimization problem associated with fitting two-layer ReLU neural networks w...	42,066
x3RPoH3bCQ-	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,053	Inverse Optimal Control Adapted to the Noise Characteristics of the Human Sensorimotor System	Computational level explanations based on optimal feedback control with signal-dependent noise have been able to account for a vast array of phenomena in human sensorimotor behavior. However, commonly a cost function needs to be assumed for a task and the optimality of human behavior is evaluated by comparing observed ...	[ "Matthias Schultheis", "Dominik Straub", "Constantin A. Rothkopf" ]	[ "motor control", "optimal feedback control", "inverse optimal control", "signal-dependent noise" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper introduces a theoretical framework for estimating cost functions in linear, quadratic, Gaussian systems with action- and state-dependent noises. Algorithms for inference are derived for the complete and partial observation cases and distinguish between the agent’s and the experimenter's inference tasks. The a...	4	[{"review_id": "u0lCHoZndKI", "reviewer": "Reviewer_VUPJ", "summary": "The authors present a method for recovering cost functions from observed state trajectories through inverse optimal control with signal-dependent noise for linear systems and quadratic costs. \nThey formulate a POMDP and distinguish between the agen...	Inverse Optimal Control Adapted to the Noise Characteristics of the Human Sensorimotor System Matthias Schultheis' Dominík Straub' Centre for Cognitive Science Centre for Cognitive Science Technical University of Darmstadt Technical University Darmstadt Darmstadt, Germany Darmstadt, Germany matthias.schultheis duddarms...	48,380
wo_0R04TSrF	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,717	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)	The paper proposes a new approach to detect coordinated accounts in social media via neural TPPs. Reviewers highlighted that the paper proposes an interesting approach to this important problem and, furthermore, shows promising results. However, based on the current manuscript, reviewers raised concerns with regard to ...	4	[{"review_id": "ilZQi1y_Ju", "reviewer": "Reviewer_q2wM", "summary": "This paper proposes extensions to Temporal Point Processes and apply them to Coordination Detection on Social Media. The main contributions are the inclusion of a \"prior knowledge based graph\" and the corresponding Gibbs-sampling based EM Algorithm...	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...	48,125
wvylaMP20_b	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,306	Improved Coresets and Sublinear Algorithms for Power Means in Euclidean Spaces	In this paper, we consider the problem of finding high dimensional power means: given a set $A$ of $n$ points in $\R^d$, find the point $m$ that minimizes the sum of Euclidean distance, raised to the power $z$, over all input points. Special cases of problem include the well-known Fermat-Weber problem -- or geometric m...	[ "Vincent Cohen-Addad", "David Saulpic", "Chris Schwiegelshohn" ]	[ "Clustering", "sublinear algorithms", "coresets", "sampling" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	This is a strong paper making contributions to sublinear time algorithms and coresets for power means clustering. All reviewers agree it should be accepted, and I think it is a candidate spotlight paper.	4	[{"review_id": "x3J5RZz7U4", "reviewer": "Reviewer_isuV", "summary": "Given n points in R^d, the goal is to compute a center x in R^d that minimizes the sum (dist(p,x))^z of distance to the power of z over every input point p. The main (sub-linear time) algorithm uniformly sample ~1/eps^z points and use smart processin...	Improved Coresets and Sublinear Algorithms for Power Means in Euclidean Spaces Vincent Cohen-Addad" Google Research, Zurich. David Saulpic" Sorbonne Université, Paris Chris Schwiegelshohn' Aarhus University Abstract In this paper, consider the problem of finding high dimensional power means: given a set A of n points i...	46,909
wn79e85F42W	neurips	2,021	main	NeurIPS.cc/2021/Conference	11,100	Differentially Private Stochastic Optimization: New Results in Convex and Non-Convex Settings	We study differentially private stochastic optimization in convex and non-convex settings. For the convex case, we focus on the family of non-smooth generalized linear losses (GLLs). Our algorithm for the $\ell_2$ setting achieves optimal excess population risk in near-linear time, while the best known differentially...	[ "Raef Bassily", "Cristóbal A Guzmán", "Michael Menart" ]	[ "Differential Privacy", "Stochastic Optimization" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper generated a lot of discussion among reviewers. The authors would be advised to improve the clarity of the discussion in order to better integrate their diverse-seeming results into one story. Please also include the improvements suggested in discussions between the reviewers and the authors.	4	[{"review_id": "yEIKjD5WS7Q", "reviewer": "Reviewer_c6Mx", "summary": "The paper studies three different (slightly related) problems in private optimization: 1. convex private optimization for non-smooth functions, 2. non-convex smooth private optimization, and 3. non-convex non-smooth private optimization. The authors...	Differentially Private Stochastic Optimization: New Results in Convex and Non-Convex Settings Raef Bassily Cristóbal Guzmán Department of Computer Science & Engineering Translational Data Analytics Institute (TDAI) Department of Applied Mathematics University of Twente Inst. for Mathematical and Comput. Eng. Pontificia...	41,760
x2pF7Tt_S5u	neurips	2,021	main	NeurIPS.cc/2021/Conference	977	Memory Efficient Meta-Learning with Large Images	Meta learning approaches to few-shot classification are computationally efficient at test time, requiring just a few optimization steps or single forward pass to learn a new task, but they remain highly memory-intensive to train. This limitation arises because a task's entire support set, which can contain up to 1000 i...	[ "John F Bronskill", "Daniela Massiceti", "Massimiliano Patacchiola", "Katja Hofmann", "Sebastian Nowozin", "Richard E Turner" ]	[ "meta-learning", "few-shot learning", "few-shot image classification" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper is addressing the problem of meta-learning with large-scale images. In this setting, many of the existing algorithms require an intractable amount of memory. The proposed method addresses this challenge by simply sub-sampling the support set in the backward pass. Although the method is sensible, the empirical...	4	[{"review_id": "XsSWeD80ZE4", "reviewer": "Reviewer_ijTZ", "summary": "This paper considers the problem of training image classification for meta-learning/few-shot learning models with large images. The authors point out that many categories of meta-learning algorithms have difficulty using large images because of memo...	Memory Efficient Meta-Learning with Large Images John Bronskill University of Cambridge jfb540cam.ad ac. uk Daniela Massiceti' Microsoft Research Massimiliano Patacchiola" University of Cambridge mp2008@cam.as ac. uk dmassicet @mi crosoft.com com Katja Hofmann Sebastian Nowozin Richard E. Turner University of Cambridge...	48,307
wfGbrrWgXDm	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,365	Safe Pontryagin Differentiable Programming	We propose a Safe Pontryagin Differentiable Programming (Safe PDP) methodology, which establishes a theoretical and algorithmic framework to solve a broad class of safety-critical learning and control tasks---problems that require the guarantee of safety constraint satisfaction at any stage of the learning and control...	[ "Wanxin Jin", "Shaoshuai Mou", "George J. Pappas" ]	[ "safe reinforcement learning", "safe learning and control", "constrained learning and control", "differentiable programming", "bi-level optimization", "safety-critical tasks" ]	NeurIPS 2021 Poster	Accept (Poster)	The authors response has address concerns from the reviewers, and all reviewers agree on an acceptance. I encourage the authors to revise the draft by including the discussion and the additional experiments from the rebuttal.	4	[{"review_id": "pPD2ln4FoX", "reviewer": "Reviewer_v3kP", "summary": "This paper provides an efficient method based on Pontryagin Differentiable Programming (PDP) for the control of safety-critical systems. The method entails formulating the control problem as a bi-level optimization problem, where\n* The lower-level p...	Safe Pontryagin Differentiable Programming Wanxin Jin Shaoshuai Mou Purdue University mous@purdue.co edu George J. Pappas University of Pennsylvania pappasgaseas. upenn. edu University of Pennsylvania wanxinjindgmail.cmm Abstract We propose a Safe Pontryagin Differentiable Programming (Safe PDP) methodol- ogy, which es...	62,799
wfJCeMS-jH	neurips	2,021	main	NeurIPS.cc/2021/Conference	5,910	Causal Identification with Matrix Equations	Causal effect identification is concerned with determining whether a causal effect is computable from a combination of qualitative assumptions about the underlying system (e.g., a causal graph) and distributions collected from this system. Many identification algorithms exclusively rely on graphical criteria made of a ...	[ "Sanghack Lee", "Elias Bareinboim" ]	[ "Causal Inference", "Causal Effect Identifiability", "Causal Effect", "Causality" ]	NeurIPS 2021 Oral	Accept (Oral)	The paper contains a solid and novel theoretical contribution. Authors should make sure that the remarks that convinced a reviewer to raise the score are sufficiently accounted for in the final version. Please take remarks on readability serious to help readers understand the challenging material. Abstract and introduc...	3	[{"review_id": "kk7m0Tc-hYE", "reviewer": "Reviewer_9UEy", "summary": "This submission presents a unified theoretical view of graphical and matrix-based approaches to nonparametric causal identification. The authors use this unified view to produce a new identification algorithm that is sound, and covers a number of ex...	Causal Identification with Matrix Equations Sanghack Lee Graduate School Data Science Seoul National University Elias Bareinboim Department of Computer Science Columbia University New York, USA eböcs. cowwmbia. edu Seoul, South Korea sanghackésmu. Abstract Causal effect identification is concerned with determining whet...	41,194
wxjtOI_8jO	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,117	Do Wider Neural Networks Really Help Adversarial Robustness?	Adversarial training is a powerful type of defense against adversarial examples. Previous empirical results suggest that adversarial training requires wider networks for better performances. However, it remains elusive how does neural network width affect model robustness. In this paper, we carefully examine the relati...	[ "Boxi Wu", "Jinghui Chen", "Deng Cai", "Xiaofei He", "Quanquan Gu" ]	[ "Adversarial Examples", "Network Width", "Neural Tangent Kernel" ]	NeurIPS 2021 Poster	Accept (Poster)	The reviewers had raised a few concerns which were mostly resolved after the authors' response. All the reviewer agreed in the discussions that the contributions of the paper are important and interesting. Also, the reviews contain all the points mentioned in the discussions, so there is nothing more to add here.	4	[{"review_id": "vilRlhisyc1", "reviewer": "Reviewer_ej3Q", "summary": "In this paper, the authors evaluate the impact of network width on the robustness of models trained through adversarial training. The authors (1) formulates robust accuracy as the intersection of natural accuracy and perturbation stability and show ...	Do Wider Neural Networks Really Help Adversarial Robustness? Boxi Wu Jinghui Chen State Key Lab of CAD&CG Pennsylvania State University State College, PA 16801 Zhejiang University boxiwudzju. edu. cn jzc5917dpsu. edu Deng Cai Xiaofei He Quanquan Gu Dept. of Computer Science UCLA qguêcs. ucla edu State Key Lab of CAD&CG...	53,287
wtLW-Amuds	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,651	The Sensory Neuron as a Transformer: Permutation-Invariant Neural Networks for Reinforcement Learning	In complex systems, we often observe complex global behavior emerge from a collection of agents interacting with each other in their environment, with each individual agent acting only on locally available information, without knowing the full picture. Such systems have inspired development of artificial intelligence a...	[ "Yujin Tang", "David Ha" ]	[ "self-organization", "attention", "reinforcement learning", "evolution strategies", "zero-shot generalization", "meta-learning", "permutation invariance", "multi-agent reinforcement learning" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	I thank the authors for their submission and active participation in the discussions. Reviewer dfwx has pledged that with the revisions by the authors, they would raise their score. Since reviewer dfwx has not done that yet and did not actively participate in the discussion with the other reviewers, I assume a higher r...	4	[{"review_id": "rX6mmPiA1G3", "reviewer": "Reviewer_Bz2K", "summary": "This paper presents a sensory neuron architecture as a layer in a neural network trained for reinforcement learning and claims that this sensory neuron makes a network generally permutation-invariant. There is an extensive related-work section cover...	The Sensory Neuron as a Transformer: Permutation-Invariant Neural Networks for Reinforcement Learning Yujin Tang David Ha² Google Brain Google Brain yujintangggoogle.com hadavid@google.cco com Abstract In complex systems, we often observe complex global behavior emerge from a collection of agents interacting with each ...	53,989
wnAN2ZU7br	neurips	2,021	main	NeurIPS.cc/2021/Conference	9,564	Pipeline Combinators for Gradual AutoML	Automated machine learning (AutoML) can make data scientists more productive. But if machine learning is totally automated, that leaves no room for data scientists to apply their intuition. Hence, data scientists often prefer not total but gradual automation, where they control certain choices and AutoML explores the...	[ "Guillaume Baudart", "Martin Hirzel", "Kiran Kate", "Parikshit Ram", "Avraham Shinnar", "Jason Tsay" ]	[ "AutoML", "scikit-learn", "programming models", "functional programming" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper elicited significant discussion. The reviewers agreed that the proposed framework could be a valuable tool and could potentially increase the uptake of AutoML systems. The user study, while small, also boosted confidence in the system's utility. The paper is also well-written. On the negative side, while t...	4	[{"review_id": "n7lwRqyQvKf", "reviewer": "Reviewer_6TXR", "summary": "This paper presents a system for gradual AutoML within sklearn. The system focuses on _gradual_ automation, as opposed to total automation, so that the data scientist can incorporate their human intuitions into the AutoML process, saving time and co...	Pipeline Combinators for Gradual AutoML Guillaume Baudart Martin Hirzel IBM Research, USA Miize@uusibmm.omm Kiran Kate IBM Research, USA kakate@us.ibm.com Inria, ENS - PSL University, France httplaume.baudrr@@iimn Parikshit Ram IBM Research, USA Avraham Shinnar IBM Research, USA shinnar@us.ibm.cd wtt.puus..b..oommmmm J...	55,485
waWmZSw0mn	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,450	Don’t Generate Me: Training Differentially Private Generative Models with Sinkhorn Divergence	Although machine learning models trained on massive data have led to breakthroughs in several areas, their deployment in privacy-sensitive domains remains limited due to restricted access to data. Generative models trained with privacy constraints on private data can sidestep this challenge, providing indirect access t...	[ "Tianshi Cao", "Alex Bie", "Arash Vahdat", "Sanja Fidler", "Karsten Kreis" ]	[ "Differential Privacy", "Generative Learning", "Optimal Transport", "Sinkhorn Divergence" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper proposes an DP algorithm for generative modeling, based on Sinkhorn divergence. This approach is non-adversarial, leading to a simpler and more stable optimization problem that is better suited for making differentially private. On the other hand, empirically, even non-private version of non-adversarial gene...	5	[{"review_id": "jPw-WlDQU6g", "reviewer": "Reviewer_U7YX", "summary": "The paper presents a DP model based on Sinkhorn GAN. Different from DP GAN models that are based on adversarial loss, they use an optimal transport (OT) distance similar to Wasserstein distance called Sinkhorn divergence. The idea is simple and incr...	Don't Generate Me: Training Differentially Private Generative Models with Sinkhorn Divergence Tianshi C,, Alex Sie Arash Vahdat Sanja Fidler Karsten Kreis* atXiv:submi//4074455 [cs.LG] 1 Nov 2021 'University of Toronto Vector Institute University of Waterloo "NVIDIA timmahtcnnvidaa.com yabieBuxaterloo.ca (avahdat, afid...	51,474
xLExSzfIDmo	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,939	Robust Contrastive Learning Using Negative Samples with Diminished Semantics	Unsupervised learning has recently made exceptional progress because of the development of more effective contrastive learning methods. However, CNNs are prone to depend on low-level features that humans deem non-semantic. This dependency has been conjectured to induce a lack of robustness to image perturbations or dom...	[ "Songwei Ge", "Shlok Kumar Mishra", "Chun-Liang Li", "Haohan Wang", "David Jacobs" ]	[ "contrastive learning", "out-of-the-domain", "self-supervised learning", "hard negative mining" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper copes with generating negative samples for contrastive learning, separating the signal from texture and shapes. The idea has been judged sufficiently novel and sound. The experiments are convincing, and the additional results presented in the rebuttal improved the empirical validation further. I recommend th...	4	[{"review_id": "teS0nWF2zGL", "reviewer": "Reviewer_ffpV", "summary": "This paper proposes new augmentation strategies for contrastive learning that produce negative samples designed to mimic the texture of points, but not have their shape data. The goal is to make models not rely on texture data. The authors report ex...	Robust Contrastive Learning Using Negative Samples with Diminished Semantics Songwei Ge Univeristy of Maryland songweigêcs. umd. edu Shiok Mishra Univeristy of Maryland shlokm@cs. umd. edu Haohan Wang Carnegie Mellon University haohanw@cs. cmu. edu Chun-Liang Li Google Cloud AI chunl maumliinnggoogle.coo David Jacobs U...	49,872
ws4BkjI1l-q	neurips	2,021	main	NeurIPS.cc/2021/Conference	571	GENESIS-V2: Inferring Unordered Object Representations without Iterative Refinement	Advances in unsupervised learning of object-representations have culminated in the development of a broad range of methods for unsupervised object segmentation and interpretable object-centric scene generation. These methods, however, are limited to simulated and real-world datasets with limited visual complexity. More...	[ "Martin Engelcke", "Oiwi Parker Jones", "Ingmar Posner" ]	[ "Generative models", "object-centric representations", "variational inference" ]	NeurIPS 2021 Poster	Accept (Poster)	- The proposed method is tackling an important problem. The reviewers found the approach is reasonable and has some novelty. Demonstration of the method on the real world dataset is a step forward in the line of research. - The major concerns from the reviewers are addressed well enough by the rebuttal. - The clarity...	4	[{"review_id": "ja6yvdptdsG", "reviewer": "Reviewer_jTRB", "summary": "Object-centric generative models aim to decompose a scene image $x$ into multiple object-wise (and background) latent $z_{1:K}$. To extract the set of latent vectors, prior work considered a fixed-order recurrent approach (e.g., GENESIS) or an ite...	GENESIS-V2: Inferring Unordered Object Representations without Iterative Refinement Martin Engelcke, Oiwi Parker Jones, and Ingmar Posner Applied AI Lab, University of Oxford, UK (martin, oiwi, ingmar)@robota ox ac. Abstract Advances in unsupervised learning of rereettteprsssntaaiinnn have culminated in the development...	35,363
x2rdRAx3QF	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,070	Self-Consistent Models and Values	Learned models of the environment provide reinforcement learning (RL) agents with flexible ways of making predictions about the environment. Models enable planning, i.e. using more computation to improve value functions or policies, without requiring additional environment interactions. In this work, we investigate a w...	[ "Gregory Farquhar", "Kate Baumli", "Zita Marinho", "Angelos Filos", "Matteo Hessel", "Hado van Hasselt", "David Silver" ]	[ "reinforcement learning", "model-based reinforcement learning", "planning", "value equivalence" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper proposes a self-consistency approach to model-based RL. The self-consistency refers to jointly optimizing the value function and the model such that (a variant) of the Bellman residual is minimized. The reviewers are all positive about this work. They believe that it is an original work and written (mostly) ...	4	[{"review_id": "xFrAQephl2u", "reviewer": "Reviewer_2igq", "summary": "This paper considers the problem of model-based reinforcement learning. The authors introduce a new loss, which drives the learned model and value function to be consistent with each other, rather than the model being consistent with the real enviro...	Self-Consistent Models and Values Gregory Farquhar DeepMind Kate Baumli DeepMind Zita Marinho DeepMind Angelos Filos University Oxford Matteo Hessel Hado van Hasselt David Silver DeepMind DeepMind DeepMind Abstract Learned models of the environment provide reinforcement learning (RL) agents with flexible ways of making...	50,879
wRXzOa2z5T	neurips	2,021	main	NeurIPS.cc/2021/Conference	5,595	Self-Attention Between Datapoints: Going Beyond Individual Input-Output Pairs in Deep Learning	We challenge a common assumption underlying most supervised deep learning: that a model makes a prediction depending only on its parameters and the features of a single input. To this end, we introduce a general-purpose deep learning architecture that takes as input the entire dataset instead of processing one datapoin...	[ "Jannik Kossen", "Neil Band", "Clare Lyle", "Aidan Gomez", "Tom Rainforth", "Yarin Gal" ]	[ "attention", "self-attention", "transformers", "multi-head self-attention", "dot-product attention", "equivariant", "equivariance", "invariant", "invariance", "interactions", "tabular", "supervised learning", "masking" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper explores alternative modeling paradigms for supervised deep learning. In this paper, the authors describe an approach wherein input to the model is the entire dataset along with the query instance. Predictions are made collectively via attention between (1) data points and (2) attributes. This model which at...	4	[{"review_id": "xKrH_ooJx-D", "reviewer": "Reviewer_m1rj", "summary": "This paper introduces a neural network architecture termed Non-Parametric Transformer (NPT) that uses self-attention (i) between data points and (ii) between features within each data point. The core conceptual contribution is to treat supervised de...	Self-Attention Between Datapoints: Going Beyond Individual Input-Output Pairs in Deep Learning Jannik Kossen Neil Band Clare Lyle" Aidan N. Gomez 1.3 Tom Rainforth Yarin Gal OATML, Department of Computer Science, University of Oxford 2 Department of Statistics, University of Oxford Cohere Abstract We challenge a common...	60,436
wQZWg82TWx	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,965	A Mathematical Framework for Quantifying Transferability in Multi-source Transfer Learning	Current transfer learning algorithm designs mainly focus on the similarities between source and target tasks, while the impacts of the sample sizes of these tasks are often not sufficiently addressed. This paper proposes a mathematical framework for quantifying the transferability in multi-source transfer learning prob...	[ "Xinyi Tong", "Xiangxiang Xu", "Shao-Lun Huang", "Lizhong Zheng" ]	[ "Transfer Learning", "Transferability", "Sample Complexity" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper proposes a new measure to quantify transferability in multi-source transfer learning problems, which takes into account the similarity between domains, sample size of each domain and model complexity. The authors have not been successful in responding to all the concerns raised by the reviewers. mainly, two ...	4	[{"review_id": "hW9G5KXZBv", "reviewer": "Reviewer_NvNX", "summary": "This paper proposes a new measure to quantify transferability in multi-source transfer learning problems, which takes into account three factors: 1). the similarity between domains, 2). the sample size of each domain, 3). the model complexity. Overal...	A Mathematical Framework for Quantifying Transferability in Multi-source Transfer Learning Xinyi Tong Xiangxiang Xu" Massachusetts Institute of Technology xuxx@mit.cm Tsinghua-Berkee Shenzhen Institute Tsinghua University txy 10yy18mmlls..mmmm tsinghua. edu. cn Shao-Lun Huang! Lizhong Zheng Massachusetts Institute of T...	41,377
wKf9iSu_TEm	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,347	Multi-Objective Meta Learning	Meta learning with multiple objectives has been attracted much attention recently since many applications need to consider multiple factors when designing learning models. Existing gradient-based works on meta learning with multiple objectives mainly combine multiple objectives into a single objective in a weighted sum...	[ "Feiyang Ye", "Baijiong Lin", "Zhixiong Yue", "Pengxin Guo", "Qiao Xiao", "Yu Zhang" ]	[ "Meta Learning", "Multi-objective Optimization", "Bi-level Optimization" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper proposes a method for multi-objective meta-learning which can be used for various settings such as semi-supervised and multi-task learning. The paper is well written. The authors provide the theoretical analysis on convergence, and the experimental results demonstrate the effectiveness of the proposed method...	4	[{"review_id": "x4LhfUl0IH", "reviewer": "Reviewer_MQP9", "summary": "The authors consider and propose a multiple objectives perspective for Meta-Learning. More specifically, they formulate the problem as a bilevel-optimization problem in which the lower-level problem is the within-task problem and the upper-level prob...	Multi-Objective Meta Learning Feiyang Ye Ye Baijiong Lin Zhixiong Yue!.?, Pengxin Guo', Qiao Xiao and Yu Zhang' 14.1 Department of Computer Science and Engineering, Southern University Science and Technology University of Technology Sydney Eindhoven University of Technology 4 Peng Cheng Laboratory (12060007, _1inbj,,,,...	55,138
we8d1FjibAc	neurips	2,021	main	NeurIPS.cc/2021/Conference	9,070	Robust and Fully-Dynamic Coreset for Continuous-and-Bounded Learning (With Outliers) Problems	In many machine learning tasks, a common approach for dealing with large-scale data is to build a small summary, {\em e.g.,} coreset, that can efficiently represent the original input. However, real-world datasets usually contain outliers and most existing coreset construction methods are not resilient against outlier...	[ "Zixiu Wang", "Yiwen Guo", "Hu Ding" ]	[ "coreset", "Continuous-and-Bounded learning", "outliers", "dynamic setting" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	This is a beautiful paper about coresets for handling outliers which will probably inspire many future related papers. While there are concerns regarding experimental results (that are hidden in the supp. material), the theoretical contribution with an algorithm that it is not hard to implement is strong enough for suc...	5	[{"review_id": "foN9XE2GAm", "reviewer": "Reviewer_vMDk", "summary": "This work provide coreset construction framework for functions that are Lipschitz continuous, smooth, and has Lipschitz continuous Hessian. The coreset in this paper is robust, i.e., it aims to handle outliers such that the coreset will be less biase...	Robust and Fully-Dynamic Coreset for Continuous-and-Boundee Learning (With Outliers) Problems Zixiu Wang Yiwen Guo Hu Ding- School Computer Science and Technology. University of Science and Technology of China "ByteDance Lab wzx20140mail. www.emmmaaa.......emmmmmmmmmeeet ustc. edu..eedduceco en, guoyiven880gmail.com co...	46,123
x2TMPhseWAW	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,874	Label Noise SGD Provably Prefers Flat Global Minimizers	In overparametrized models, the noise in stochastic gradient descent (SGD) implicitly regularizes the optimization trajectory and determines which local minimum SGD converges to. Motivated by empirical studies that demonstrate that training with noisy labels improves generalization, we study the implicit regularization...	[ "Alex Damian", "Tengyu Ma", "Jason D. Lee" ]	[ "stochastic gradient descent", "implicit regularization", "non-convex optimization" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper studies the convergence of SGD in the presence of label noise. The authors show that SGD with label noise converges to a stationary point of a certain regularized loss function, where the regularization depends on the amount of noise, batch size, and the stepsize. The assumptions on size of regularization an...	4	[{"review_id": "gZUyqQX32uI", "reviewer": "Reviewer_n6Kd", "summary": "## Post-rebuttal \nI thank the authors for their clarifications. We had a long discussion about the technical details, and some aspects became more clear to me, but it did not lead to a resolution of the concern that the theory is somewhat weak. ...	Label Noise SGD Provably Prefers Flat Global Minimizers Alex Damian Princeton University ad27@princeton.co edu Tengyu Ma Stanford University tengyuma@stannfrrd edu Jason Lee Princeton University jasomlee@princetoo.eed Abstract In overparametrized models, the noise in stochastic gradient descent (SGD) implic- itly regul...	40,622
wTutcPE7zOC	neurips	2,021	main	NeurIPS.cc/2021/Conference	8,813	Metropolis-CVAE: Bootstrapping Labels for Bayesian Inference via Semi-Supervised Conditional Variational Autoencoders	In Bayesian parameter estimation, models make simplifying assumptions to make parameter inference feasible. If learned inference methods are trained using data simulated by models, however, distributional differences between simulated and observed data may lead to biased inference results on the observed data. In this ...	[ "Jonathan Doucette", "Christian Kames", "Alexander Rauscher" ]	[ "Metropolis-Hastings", "CVAE", "Bayesian Inference", "MCMC", "Semi-Supervised" ]	NeurIPS 2021 Submitted	Reject	This paper describes an algorithm reminiscent of the dynamics of an Monte-Carlo expectation-maximization (MCEM) algorithm where the E step is replaced by MCMC, here the classical MH algorithm. During the training of a VAE (similar to M step), the authors progressively improve the psudolabels via a Metropolis-Hastings s...	4	[{"review_id": "utJ2Tv673C", "reviewer": "Reviewer_bwY8", "summary": "Variational inference and distribution sampling are two popular methods for statistical inference. The present paper combines these two approaches in a semi-supervised learning setting, where the distribution of unlabeled data may differ from that of...	Metropolis-CVAE: Bootsrrppiing Labels for Bayesian Inference via Semi-Supervised Conditional Variational Autoencoders Anonymous Author(s) Affiliation Address email Abstract In Bayesian parameter estimation, models make simplifying assumptions to make parameter inference feasible. If learned inference methods are traine...	37,336
wLsA3nurh9W	neurips	2,021	main	NeurIPS.cc/2021/Conference	8,350	Analysis of one-hidden-layer neural networks via the resolvent method	In this work, we investigate the asymptotic spectral density of the random feature matrix $M = Y Y^*$ with $Y = f(WX)$ generated by a single-hidden-layer neural network, where $W$ and $X$ are random rectangular matrices with i.i.d. centred entries and $f$ is a non-linear smooth function which is applied entry-wise. We ...	[ "Vanessa Piccolo", "Dominik Schröder" ]	[ "Neural network", "Resolvent method", "Cumulant expansion" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper considers the limited singular value distribution of a random feature model, which consists of an entrywise nonlinearity and a product of two i.i.d. matrices. In particular, the generalization performance of a single hidden layer neural network where the input data and hidden layer weights are i.i.d. random ...	4	[{"review_id": "bGkCU_1mkWH", "reviewer": "Reviewer_HGSJ", "summary": "This paper studies the limiting eigenvalue-distribution of the gram matrix $M$ associated with a random feature model, and extends current theory (Pennington and Worah 2017; Benigni and Péché 2019, etc.) to the case of multiple layers with additive ...	Analysis of one-hidden-layer Neural Networks via the Resolvent Method Vanessa Piccolo Dominik Schröder Institute for Theoretical Studies ETH Zurich dschroeder@ethhzcco ch ETH Zurich (current affiliation: ENS Lyon) vanessa, peceesseeeccoolees1y1ooooo Abstract In this work, we investigate the asymptotic spectral density ...	33,854
wdIDt--oLmV	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,769	Amortized Synthesis of Constrained Configurations Using a Differentiable Surrogate	In design, fabrication, and control problems, we are often faced with the task of synthesis, in which we must generate an object or configuration that satisfies a set of constraints while maximizing one or more objective functions. The synthesis problem is typically characterized by a physical process in which many dif...	[ "Xingyuan Sun", "Tianju Xue", "Szymon Rusinkiewicz", "Ryan P Adams" ]	[ "Amortization", "Inverse design", "Differentiable surrogate", "3D printing", "Soft robot" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	Thank you for your submission to NeurIPS. Even after discussion, there is some substantial disagreement on this paper, so I will need to express an opinion that is not a unanimous one. But with this caveat, I will state that I ultimately come down very much on the "positive" side of the disagreement, and I am recomme...	4	[{"review_id": "tPEXfaKVsVG", "reviewer": "Reviewer_P3xV", "summary": "The paper addresses the challenge of synthesis - where the goal is to find a configuration of a physical system that satisfies a set of constraints and optimizes one or more objective functions representing the desiderata. The synthesis challenge ha...	Amortized Synthesis of Constrained Configurations Using a Differentiable Surrogate Xingyuan Sun', Tianju Xue², Szymon Rusinkiewicz', Ryan P. Adams Department of Computer Science *Department of Civil and Environmental Engineering Princeton University (xs5, txue, smr, rpay@princeton.eer Abstract In design, fabrication, a...	61,094
weBSeGTv0i	neurips	2,021	main	NeurIPS.cc/2021/Conference	9,432	A Cramér Distance perspective on Non-crossing Quantile Regression in Distributional Reinforcement Learning	Distributional reinforcement learning (DRL) extends the value-based approach by estimating the full distribution over future returns instead of the mean only, providing a richer signal that leads to improved performances. Quantile-based methods like QR-DQN project arbitrary distributions into a parametric subset of sta...	[ "Alix Lhéritier", "Nicolas Bondoux" ]	[ "Distributional Reinforcement Learning", "Cramér distance", "Quantile regression", "Wasserstein distance", "Non-crossing quantiles", "Neural networks", "Atari benchmark" ]	NeurIPS 2021 Submitted	Reject	As the authors raised concerns with the reviews, I have thoroughly read the paper and even discussed the situation with the senior area chair. While the reviews are all against acceptance, my individual assessment is much more positive. Not to burry the lead, but, after discussions I will be suggesting that the paper n...	4	[{"review_id": "vmqUjG7ntbE", "reviewer": "Reviewer_fTr8", "summary": "The authors provide theoretical analysis of distributional RL under Cramer distance. The authors further propose a neural architecture that yields monotonic quantiles and achieved improvements over prior works.", "questions": "", "limitations": "", ...	A Cramér Distance perspective on Non-crossing Quantile Regression in Distributional Reinforcement Learning Anonymous Author(s) Affiliation Address email Abstract Distributional reinforcement learning (DRL) extends the value-based approach by using a deep convolutional network to approximate the full distribution over f...	35,562
wFuWSdCD7BN	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,062	A Constant Approximation Algorithm for Sequential Random-Order No-Substitution k-Median Clustering	We study k-median clustering under the sequential no-substitution setting. In this setting, a data stream is sequentially observed, and some of the points are selected by the algorithm as cluster centers. However, a point can be selected as a center only immediately after it is observed, before observing the next point...	[ "Tom Hess", "Michal Moshkovitz", "Sivan Sabato" ]	[ "k-median clustering", "constant approximation", "sequential algorithms" ]	NeurIPS 2021 Poster	Accept (Poster)	The authors present the first constant approximation factor on the optimal risk for k-median with no -substitution under a random arrival order model. The problem studied in the paper is an interesting problem and the authors present a simple and elegant algorithm that outperforms previous work. The only concern with t...	4	[{"review_id": "mTBENld9Rzg", "reviewer": "Reviewer_DQhU", "summary": "This is a theoretical paper, studying sequentially assigning points as k-median cluster centers. \nIt assumes a random arrival order, but has no other structural assumptions. It offers a constant factor approximation using O(k log^2 k) centers. T...	A Constant Approximation Algorithm for Sequential Random-Order No-Substitution k--Median Clustering Tom Hess Michal Moshkovitz Department of Computer Science Ben-Gurion University of the Negev Department of Computer Science Tel-Aviv University" Beer-Sheva, Israel tomhe@post. bgu.ac. Tel Aviv, Israel mmoshkovitz@eng ucs...	42,999
x1Lp2bOlVIo	neurips	2,021	main	NeurIPS.cc/2021/Conference	11,175	Diffusion Normalizing Flow	We present a novel generative modeling method called diffusion normalizing flow based on stochastic differential equations (SDEs). The algorithm consists of two neural SDEs: a forward SDE that gradually adds noise to the data to transform the data into Gaussian random noise, and a backward SDE that gradually removes th...	[ "Qinsheng Zhang", "Yongxin Chen" ]	[ "normalizing flow", "diffusion probabilistic models", "density estimation", "generative models" ]	NeurIPS 2021 Poster	Accept (Poster)	There was significant discussion of this paper, with one reviewer increasing their score. The score range was unusually wide, even after discussion. Reviewers felt that the idea presented was both novel and interesting. I agree with this assessment -- I think this paper draws a very nice and interesting link between n...	4	[{"review_id": "gwW4MrbPYcl", "reviewer": "Reviewer_H9qy", "summary": "This paper proposes to lean a more expressive forward process in a diffusion model by using a neural network drift function.", "questions": "", "limitations": "", "rating": 4, "confidence": 4, "soundness": null, "presentation": null, "contribution":...	Diffusion Normalizing Flow Qinsheng Zhang Georgia Institute of Technology gzhang4198gatech.. edu Yongxin Chen Georgia Institute of Technology yongchen@gatech. edu Abstract We present novel generative modeling method called diffusion normalizing flow based on stochastic differential equations (SDEs). The algorithm consi...	41,939
wZYWwJvkneF	neurips	2,021	main	NeurIPS.cc/2021/Conference	931	FACMAC: Factored Multi-Agent Centralised Policy Gradients	We propose FACtored Multi-Agent Centralised policy gradients (FACMAC), a new method for cooperative multi-agent reinforcement learning in both discrete and continuous action spaces. Like MADDPG, a popular multi-agent actor-critic method, our approach uses deep deterministic policy gradients to learn policies. However, ...	[ "Bei Peng", "Tabish Rashid", "Christian Schroeder de Witt", "Pierre-Alexandre Kamienny", "Philip Torr", "Wendelin Boehmer", "Shimon Whiteson" ]	[ "multi-agent reinforcement learning", "multi-agent policy gradients" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper generated an involved discussion between the reviewers and the authors, as well as between the reviewers themselves. The paper essentially combines two well-known baselines in MARL domain, and therefore was judged as a report of experiments for which reproducibility is of particular importance. The reviewers...	3	[{"review_id": "p5ITSEL_Gtz", "reviewer": "Reviewer_Byep", "summary": "This paper proposes to apply the idea of value function factorization from recent value-based multi-agent RL work (e.g., VDN, QTRAN, QMIX) and apply it in the multi-agent actor-critic setting. Specifically, the paper proposes to use a centralized fa...	FACMAC: Factored Multi-Agent Centralised Policy Gradients Bei Peng" University of Liverpool Tabish Rashid" University of Oxford Christian A. Schroeder de Witt" University of Oxford Pierre-Alexande Kamienny Philip H. S. Torr University Oxford Wendelin Böhmer Delft University of Technology Facebook Al Research Shimon Whi...	53,307
wP9twkexC3V	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,746	Learning to Synthesize Programs as Interpretable and Generalizable Policies	Recently, deep reinforcement learning (DRL) methods have achieved impressive performance on tasks in a variety of domains. However, neural network policies produced with DRL methods are not human-interpretable and often have difficulty generalizing to novel scenarios. To address these issues, prior works explore learni...	[ "Dweep Trivedi", "Jesse Zhang", "Shao-Hua Sun", "Joseph J Lim" ]	[ "Program Synthesis", "Neural Program Synthesis", "Programmatic Reinforcement Learning", "Reinforcement Learning", "Deep Reinforcement Learning" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper presents an interesting idea of learning policies for solving RL problems through program synthesis. The key is to learn a good program embedding space, through 3 different losses that take care of 3 different aspects of a program representation. The reviewers all appreciate the clarity of the paper and li...	4	[{"review_id": "sQhjVkaEI9A", "reviewer": "Reviewer_dyBV", "summary": "The paper approaches reinforcement learning from a program synthesis perspective and proposes a new method - LEAPS - that is capable of learning programmatic policies just from weak reward signals. This is achieved in two steps:\ni) learning a smoot...	Learning to Synthesize Programs as Interpretable and Generalizable Policies Dweep Trivedi* Jesse Zhang*! Shao-Hua Sun' Universstty of Southern California Idtrivedi, jessez, shaohuas, limjj @uscc edu Joseph J. Lim Abstract Recently, deep reinforcement learning (DRL) methods have achieved impressive performance on tasks ...	65,334
wPA_5Wsjt8i	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,713	Conditional Generation Using Polynomial Expansions	Generative modeling has evolved to a notable field of machine learning. Deep polynomial neural networks (PNNs) have demonstrated impressive results in unsupervised image generation, where the task is to map an input vector (i.e., noise) to a synthesized image. However, the success of PNNs has not been replicated in con...	[ "Grigorios Chrysos", "Markos Georgopoulos", "Yannis Panagakis" ]	[ "polynomial expansions", "polynomial neural networks", "GAN", "VAE", "unseen combinations", "conditional data generation" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper proposed an extension of polynomial neural networks (PNNs) for application to conditional generation tasks. Experiments show that the proposed conditional model can be applied to five different tasks (class-conditional generation, inverse problems, edges-to-image translation, image-to-image translation, attr...	4	[{"review_id": "V1_B56uvNKv", "reviewer": "Reviewer_Rugm", "summary": "This paper proposes a model for conditional generation of images, which relies on the creation of polynomials of the inputs. The paper evaluates this generation process in a multitude of tasks, showing improvement over similar techniques.", "questio...	Conditional Generation Using Polynomial Expansions Grigorios G Chrysos Markos Georgopoulos EPFL, Switzerland Imperial College London, UK grigorios. ctt.::ssss.crsssspp......oo. wr.gerggoooollsst ieme@rmpppriee www.oosssmmmemaaaa....mmaa ac.. Yannis Panagakis University of Athens, GR yannispēdi, uua. Abstract Generative...	53,912
wZrOOO9XBn	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,497	Lossy Compression for Lossless Prediction	Most data is automatically collected and only ever "seen" by algorithms. Yet, data compressors preserve perceptual fidelity rather than just the information needed by algorithms performing downstream tasks. In this paper, we characterize the bit-rate required to ensure high performance on all predictive tasks that are ...	[ "Yann Dubois", "Benjamin Bloem-Reddy", "Karen Ullrich", "Chris J. Maddison" ]	[ "Compression", "Invariances", "Information Theory", "Machine Learning", "Self-Supervised Learning" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	The paper analyzes "supervised" data compression for downstream predictive tasks, aiming to achieve higher compression rates with negligible or no performance loss. It theoretically characterizes the bitrate required to ensure high performance on all predictive tasks that are invariant under a set of transformations, s...	4	[{"review_id": "zo65jGcs0kU", "reviewer": "Reviewer_XSke", "summary": "This paper proposes methods to introduce a self-supervised approach to neural image compression for classification tasks.\nVIC is a modified neural compressor in which inputs are augmented but target reconstructions are not.\nBINCE is a modified neu...	Lossy Compression for Lossless Prediction Yann Dubois Benjamin Bloem-Reddy Vector Institute yamndubois99@gmiil.ccmmm The University of British Columbia benbrāstat.uke citiitat.bbb.c. ubc.ca ca Karen Ullrich Chris J. Maddison Facebook Al Research University of Toronto Vector Institute cmaddis@cs. toronto. karenuêfb. com...	56,400
wTLc2HcWLIM	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,352	Adaptive First-Order Methods Revisited: Convex Minimization without Lipschitz Requirements	We propose a new family of adaptive first-order methods for a class of convex minimization problems that may fail to be Lipschitz continuous or smooth in the standard sense. Specifically, motivated by a recent flurry of activity on non-Lipschitz (NoLips) optimization, we consider problems that are continuous or smooth ...	[ "Kimon Antonakopoulos", "Panayotis Mertikopoulos" ]	[ "convex optimization", "non-Lipschitz", "adaptive methods" ]	NeurIPS 2021 Poster	Accept (Poster)	Overall the reviewers liked the paper and found the contributions strong enough for NeurIPS. I have read the paper, and while I have an issue with the claim of the authors that their algorithm is adaptive, while it requires the specification of the "smoothing function" h(), I still appreciate the overall technical cont...	4	[{"review_id": "d3VPAjb__uT", "reviewer": "Reviewer_xpRJ", "summary": "This paper studies convex minimization problems lacking Lipschitz continuity of the objective and compactness of the feasible set. They assume access to a standard stochastic subgradient oracle with mean-zero noise and bounded variance. The main con...	Adaptive First-Order Methods Revisited: Convex Optimization without Lipschitz Requirements Kimon Antonakopoulos Panayotis Mertikopoulos Univ. Grenoble Alpes, CNRS, Inria, Grenoble INP, LIG 38000 Grenoble, France & Criteo Al Lab kimon, k ianoonannttonkkoouloeeeiria..... fr panayotis. mtr.oissoemetiiooooooseeeeeeee fr Ab...	46,921
wHxnK7Ucogy	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,886	Bridging Explicit and Implicit Deep Generative Models via Neural Stein Estimators	There are two types of deep generative models: explicit and implicit. The former defines an explicit density form that allows likelihood inference; while the latter targets a flexible transformation from random noise to generated samples. While the two classes of generative models have shown great power in many applic...	[ "Qitian Wu", "Rui Gao", "Hongyuan Zha" ]	[ "deep generative models", "generative adversarial networks", "energy models", "stein's methods" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper proposes Stein bridge as a generic framework to jointly train an explicit generative model as well as an implicit model. Some theoretical analyses are provided. Experiments are conducted on synthetic data as well as image generation with MNIST data. Reviewers agree that the proposed method is new. However, ...	4	[{"review_id": "QB9IlEESCNZ", "reviewer": "Reviewer_c3DX", "summary": "Explicit and implicit model training have their own advantages and limitations. The author proposes a joint training framework, which aims to combine the benefits from both worlds. Specifically, the author proposes a training framework called **Stei...	Bridging Explicit and Implicit Deep Generative Models via Neural Stein Estimators Qitian Wu' Rui Gao Hongyuan Zha 1:3 Department of Computer Science and Engineering. MoE Key Lab of Artificial Intelligence, AI Institute, Shanghai Jiao Tong University University of Texas Austin $ School of Data Science, Shenzhen Institut...	44,431
wWtk6GxJB2x	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,660	PDE-GCN: Novel Architectures for Graph Neural Networks Motivated by Partial Differential Equations	Graph neural networks are increasingly becoming the go-to approach in various fields such as computer vision, computational biology and chemistry, where data are naturally explained by graphs. However, unlike traditional convolutional neural networks, deep graph networks do not necessarily yield better performance than...	[ "Moshe Eliasof", "Eldad Haber", "Eran Treister" ]	[ "Graph Neural Networks", "Neural Network Architectures", "PDES in Deep Learning", "Partial Differential Equations" ]	NeurIPS 2021 Poster	Accept (Poster)	All ratings were (weak) "accept". Generally the work is novel and timely. A related paper appeared on arxiv in late June (GRAND), which one reviewer pointed to, but obviously this paper was submitted before that appeared, and doesn't factor into this paper's evaluation (as the reviewer appropriately recognized). To m...	3	[{"review_id": "lsWSWsCvNb8", "reviewer": "Reviewer_Buis", "summary": "This paper proposed using discretized PDE operator on the graph to generalize current graph convolutions. With diffusion and hyperbolic equations, the proposedd method has the flexibility to fit the unseen graph signals.", "questions": "", "limitati...	PDE-GCN: Novel Architectures for Graph Neural Networks Motivated by Partial Differential Equations Moshe Eliasof Eldad Haber Department of Computer Science Ben-Gurion University of the Negev Department of Earth, Ocean and Atmospheric Sciences University of British Columbia Beer-Sheva, Israel eliasof@post, bgu.ac. .c.il...	47,609
wHoIjrT6MMb	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,474	Robust Compressed Sensing MRI with Deep Generative Priors	The CSGM framework (Bora-Jalal-Price-Dimakis'17) has shown that deep generative priors can be powerful tools for solving inverse problems. However, to date this framework has been empirically successful only on certain datasets (for example, human faces and MNIST digits), and it is known to perform poorly on out-of-dis...	[ "Ajil Jalal", "Marius Arvinte", "Giannis Daras", "Eric Price", "Alex Dimakis", "Jonathan Tamir" ]	[ "Compressed sensing", "generative priors", "MRI reconstruction", "Langevin dynamics", "Inverse problems" ]	NeurIPS 2021 Poster	Accept (Poster)	The authors provide the first demonstration that Compressed Sensing with Generative Priors can be a competitive approach for MRI reconstruction relative to end-to-end, L1, and untrained neural network methods. The authors additionally demonstrate that the method is more robust than baselines in the context of certain ...	4	[{"review_id": "tyeRP-wFydI", "reviewer": "Reviewer_nt1J", "summary": "This paper proposed the robust MR image reconstruction framework based on CSGM (Compressed Sensing with Generative Modeling). Compared to the other MR reconstruction algorithms, the proposed method shows consistently superior reconstruction results ...	Robust Compressed Sensing MRI with Deep Generative Priors Ajil Jalal" ECE, UT Austin ajiljalal@utexen edu Marius Arvinte" ECE, UT Austin arvintelutexas edu Giannis Daras CS, UT Austin gi giamissdaaasuutexas edu Eric Price CS, UT Austin ecprice@cs ute.as. Alexandros G. Dimakis Jonathan Tamir ECE, UT Austin j jtamir@utex...	59,870
wFp6kmQELgu	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,133	MICo: Improved representations via sampling-based state similarity for Markov decision processes	We present a new behavioural distance over the state space of a Markov decision process, and demonstrate the use of this distance as an effective means of shaping the learnt representations of deep reinforcement learning agents. While existing notions of state similarity are typically difficult to learn at scale due to...	[ "Pablo Samuel Castro", "Tyler Kastner", "Prakash Panangaden", "Mark Rowland" ]	[ "Reinforcement Learning", "Metrics", "Deep Reinforcement Learning" ]	NeurIPS 2021 Poster	Accept (Poster)	This seems like a solid paper that develops new approach to the interesting problem of inducing bias in learning through a notion of state similarity. The notion overcomes some limitations of previous notions that were based on bisimulation concepts. The reviewers are aligned in their ratings of the paper.	4	[{"review_id": "wu2pF40ADzv", "reviewer": "Reviewer_vaui", "summary": "The paper presents MICo - a behavioural metric to learn representations for deep RL agents and shows its effectiveness - both theoretically and empirically (on the Arcade Learning Environment benchmark).", "questions": "", "limitations": "", "rating...	MICo: Improved representations via sampling-based state similarity for Markov decision processes Pablo Samuel Castro" Google Research, Brain Team Tyler Kastner" McGill University Prakash Panangaden McGill University Mark Rowland DeepMind Abstract We present a new behavioural distance over the state space of Markov deci...	50,514
wGRNAqVBQT2	neurips	2,021	main	NeurIPS.cc/2021/Conference	812	Leveraging SE(3) Equivariance for Self-supervised Category-Level Object Pose Estimation from Point Clouds	Category-level object pose estimation aims to find 6D object poses of previously unseen object instances from known categories without access to object CAD models. To reduce the huge amount of pose annotations needed for category-level learning, we propose for the first time a self-supervised learning framework to esti...	[ "Xiaolong Li", "Yijia Weng", "Li Yi", "Leonidas Guibas", "A. Lynn Abbott", "Shuran Song", "He Wang" ]	[ "6D Pose", "self-supervision", "SE(3) equivariance", "3D point cloud", "category-level", "real data", "shape reconstruction" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper proposes a self-supervised category-level 6D pose estimation method in which canonical shapes for the input objects are predicted without explicit supervision of orientation. It introduces an equivariant network architecture and shows its importance for canonical shape estimation. Reviewers point out miss...	3	[{"review_id": "uu1jPeNeN44", "reviewer": "Reviewer_ojV5", "summary": "This paper introduces a self-supervised category-level 6D pose estimation to minimize ideally canonical reconstructed point cloud and input point cloud based on rotation-equivariant features. Experiment results can verify the effectiveness on both c...	Leveraging SE(3) Equivariance for Self-Supervised Category-Level Object Pose Estimation Xiaolong Virginia Tech Yijia Weng Li Tsinghua University ericyi01248 gmail com Peking University 1xi 1xiaol9@vt.ed hal Asmmmemeer gmmam............ com Leonidas Guibas A. Lynn Abbott Virginia Tech Stanford University guibas@cs. stan...	48,206
wCrH0JBCFNm	neurips	2,021	main	NeurIPS.cc/2021/Conference	885	Post-Training Quantization for Vision Transformer	Recently, transformer has achieved remarkable performance on a variety of computer vision applications. Compared with mainstream convolutional neural networks, vision transformers are often of sophisticated architectures for extracting powerful feature representations, which are more difficult to be developed on mobile...	[ "Zhenhua Liu", "Yunhe Wang", "Kai Han", "Wei Zhang", "Siwei Ma", "Wen Gao" ]	[ "Post-training", "vision transformer", "ranking loss of self-attention", "mixed-precision" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper is solving an important problem for efficient vision transformer deployment in industrial environments. The goal of the paper is significant in practical applications. However, reviewers are concerned about the novelty is not enough and the method part is not clear; A large number of hyper-parameters are man...	4	[{"review_id": "iTEDOni2-n5", "reviewer": "Reviewer_Zj6U", "summary": "This paper proposed quantization of the output of a transformer network, to reduce memory and computation requirement.\n\nThe method proposes to quantize \n1. the MLP portion of the transformer using the Pearson correlation coefficient as a criteri...	Post- Post-Trainin Quantization for Vision Transformer Zhenhua Liu Yunhe Wang"; Kai Han', Wei Zhang', Siwei Ma 13, Wen Gao 'School of Electronic Engineering and Computer Science, Peking University Huawei Noah's Ark Lab 'Peng Cheng Laboratory 1iu Ziu-zh@pku.ede edu. cn, {yunhe. wang, kai. han,, wz. (e..zha)) com, (swma,...	39,830
wEOlVzVhMW_	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,701	Beware of the Simulated DAG! Causal Discovery Benchmarks May Be Easy to Game	Simulated DAG models may exhibit properties that, perhaps inadvertently, render their structure identifiable and unexpectedly affect structure learning algorithms. Here, we show that marginal variance tends to increase along the causal order for generically sampled additive noise models. We introduce varsortability as ...	[ "Alexander Gilbert Reisach", "Christof Seiler", "Sebastian Weichwald" ]	[ "Causality", "Directed Acyclic Graphs", "Structure Learning", "Continuous Optimization" ]	NeurIPS 2021 Poster	Accept (Poster)	A summary from one of the reviews: "This paper shows that one needs to be cautious when rescaling your data prior to using a causal structure learning algorithms. The authors introduce the concept of varsortability that measures the agreement in how much the marginal variance tends to increase along the causal order. ...	3	[{"review_id": "xWd2uHmIRu", "reviewer": "Reviewer_BrkD", "summary": "This submission is about causal structure learning and how some simulated directed acyclic graphs (that are used for benchmarking structure learning approaches, and that are generally used for causal representations) can have properties (in particula...	Beware of the Simulated DAG! Causal Discovery Benchmarks May Be Easy To Game Alexander G. Reisach Christof Seiler Sebastian Weichwald' 'Department of Mathematical Sciences, University of Copenhagen, Denmark DDepartment of Data Science and Knowledge Engineering, Maastricht University, The Netherlands 3 Mathematics Centr...	52,444
wDJUUcCTNI2	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,390	Weak-shot Fine-grained Classification via Similarity Transfer	Recognizing fine-grained categories remains a challenging task, due to the subtle distinctions among different subordinate categories, which results in the need of abundant annotated samples. To alleviate the data-hungry problem, we consider the problem of learning novel categories from web data with the support of a c...	[ "Junjie Chen", "Li Niu", "Liu Liu", "Liqing Zhang" ]	[ "transfer learning", "similarity transfer", "weak-shot learning" ]	NeurIPS 2021 Poster	Accept (Poster)	None of the reviewers recommend accepting this paper. There was substantive discussion during the rebuttal period, but the reviewers remained of the opinion that the work should not be accepted, with these opinions based on the quality of the presentation, their view of the novelty of the work and the degree to which t...	4	[{"review_id": "VDoIQcuhBQ9", "reviewer": "Reviewer_K9m3", "summary": "This paper proposes to transfer pairwise semantic similarity from clean image data to web image data for classifying novel fine-grained categories. The proposed method provides a way to alleviate the data-hungry problem in the fine-grained classific...	Weak-shot Fine-grained Classification via Similarity Transfer Junjie Chen, Li Niu, Liu Liu, Liqing Zhang" MoE Key Lab of Artificial Intelligence, Department of Computer Science and Engineering, Shanghai Jiao Tong University (chen.bys, ustcnewly, shirlley)asji edu. cn, chang-lq@css saaeelllossssstt... edu. cn Abstract R...	46,728
x2lBl0GRav5	neurips	2,021	main	NeurIPS.cc/2021/Conference	5,602	An Improved Analysis and Rates for Variance Reduction under Without-replacement Sampling Orders	When applying a stochastic algorithm, one must choose an order to draw samples. The practical choices are without-replacement sampling orders, which are empirically faster and more cache-friendly than uniform-iid-sampling but often have inferior theoretical guarantees. Without-replacement sampling is well understood on...	[ "Xinmeng Huang", "Kun Yuan", "Xianghui Mao", "Wotao Yin" ]	[ "convex optimization", "stochastic optimization", "SGD", "variance reduction" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper is favored by two of four expert reviewers and is acceptable to the other two. Reviewers agree that it is clearly written and organized. There is also general agreement that parts of the analysis are new and insightful, and possibly useful outside of this paper, in particular the "order-specific" norm that i...	4	[{"review_id": "zVPuYD4HAv", "reviewer": "Reviewer_dHKd", "summary": "This paper develops a proximal damped version of the Finito algorithm. The algorithm is proved to achieve the same convergence rate as proximal GD, for cyclic sampling, random reshuffling and shuffling-once versions of the algorithm. Further, the aut...	Improved Analysis and Rates for Variance Reduction under Without-replacmment Sampling Orders Xinmeng Huang" University of Pennsylvania Philadelphia, PA 19104 xinmengh@sas. upenn. edu Kun Yuan" DAMO Academy. Alibaba Group Bellevue, WA 98004 kun www.anallaaaaalaaaaaaaaaaa com Xianghui Mao Tsinghua University Beijing, Chi...	37,790
wEFC5PY0g_0	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,884	3D Siamese Voxel-to-BEV Tracker for Sparse Point Clouds	3D object tracking in point clouds is still a challenging problem due to the sparsity of LiDAR points in dynamic environments. In this work, we propose a Siamese voxel-to-BEV tracker, which can significantly improve the tracking performance in sparse 3D point clouds. Specifically, it consists of a Siamese shape-aware f...	[ "Le Hui", "Lingpeng Wang", "Mingmei Cheng", "Jin Xie", "Jian Yang" ]	[ "3D Point Cloud", "3D Single Object Tracking" ]	NeurIPS 2021 Poster	Accept (Poster)	Initially, one of the reviewers expressed concerns about the paper (lack of clarity and limited novelty) and ranked the paper below acceptance. Another area of concern was related to a number of claims that the authors made and were not fully justified experimentally. As the ensuing rebuttal managed to successfully add...	7	[{"review_id": "yZ-yVZ6S2hE", "reviewer": "Reviewer_QKmE", "summary": "This paper tackles the problem of single-object tracking from point clouds. Specifically, it adopts a tracking by detection framework to find the target in the search area, implemented with a siamese network that extracts both the template and searc...	3D Siamese Voxel-to-BEV Tracker for Sparse Point Clouds Le Hui Lingpeng Wang', Mingmei Cheng, Jin Xie", Jian Yang" PCA Lab, Nanjing University of Science and Technology, China hui, calpwang, chengmm, csjxie, csjyang)onjust... edu. en Abstract 3D object tracking in point clouds is still challenging problem due the spars...	52,764
wHkKTW2wrmm	neurips	2,021	main	NeurIPS.cc/2021/Conference	6,988	Neural Additive Models: Interpretable Machine Learning with Neural Nets	Deep neural networks (DNNs) are powerful black-box predictors that have achieved impressive performance on a wide variety of tasks. However, their accuracy comes at the cost of intelligibility: it is usually unclear how they make their decisions. This hinders their applicability to high stakes decision-making domains s...	[ "Rishabh Agarwal", "Levi Melnick", "Nicholas Frosst", "Xuezhou Zhang", "Ben Lengerich", "Rich Caruana", "Geoffrey Hinton" ]	[ "Additive Models", "Interpretability", "Multitask learning", "Explainable AI" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	The reviewers were overall very enthusiastic about this paper, highlighting its novelty, and clarity in terms of writing and visualizations. There are a few changes that I expect you will make based on your responses to the reviewers.	4	[{"review_id": "bEXNkDvjvZE", "reviewer": "Reviewer_Jr31", "summary": "In this paper, the authors introduce a new generalized additive model, called neural additive models, which use neural networks to learn a non-linear transformation of each input feature, independently. Along the way, they introduce a new activation...	Neural Additive Models: Interpretable Machine Learning with Neural Nets Rishabh Agarwal Levi Melnick Google Research, Brain Team Microsoft Research Nicholas Frosst Cohere Xuezhou Zhang Ben Lengerich MIT University of Wisconsin-Maddise Rich Caruana Geoffrey E. Hinton Google Research, Brain Team Microsoft Research Abstra...	47,813
wGmOLwb8ClT	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,009	Robust Counterfactual Explanations on Graph Neural Networks	Massive deployment of Graph Neural Networks (GNNs) in high-stake applications generates a strong demand for explanations that are robust to noise and align well with human intuition. Most existing methods generate explanations by identifying a subgraph of an input graph that has a strong correlation with the prediction...	[ "Mohit Bajaj", "Lingyang Chu", "Zi Yu Xue", "Jian Pei", "Lanjun Wang", "Peter Cho-Ho Lam", "Yong Zhang" ]	[ "Counterfactual", "explanations", "GNN", "robust", "graph neural networks", "interpretation", "explainable AI", "decision logic" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper presents a new method to produce robust counterfactual explanations for the predictions of GNNs. However, reviewers raised several concerns about the method, the experiments, and the writing. Therefore, the paper cannot be accepted in the current form.	3	[{"review_id": "avis1Em2VYO", "reviewer": "Reviewer_ZnuS", "summary": "The paper presents an approach to GNN prediction explanations based on providing modifications to the input samples to change the prediction outcomes. The idea is that those modifications are the ones that can make a human understand the reason for ...	Robust Counterfactual Explanations on Graph Neural Networks Mohit Bajaj** Lingyang Chu Zi Yu Xue Jian Pei Lanjun Wang' Peter Cho-Ho Lam Yong Zhang' Huawei Technologies Canada Co., Ltd. McMMatter University The University of British Columbia Simon Fraser University [mohit. bajaji, zi yu.uuue, lanjun jun wangunn,, cho ho...	47,689
w3x8K0M6sAz	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,616	Topology-Imbalance Learning for Semi-Supervised Node Classification	The class imbalance problem, as an important issue in learning node representations, has drawn increasing attention from the community. Although the imbalance considered by existing studies roots from the unequal quantity of labeled examples in different classes (quantity imbalance), we argue that graph data expose a u...	[ "Deli Chen", "Yankai Lin", "Guangxiang Zhao", "Xuancheng Ren", "Peng Li", "Jie Zhou", "Xu Sun" ]	[ "node classification", "topology imbalance learning", "semi-supervised learning", "graph neural network" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper investigates the issue that nodes in a graph may have different local topology. This imbalance of topology may induce bias in node classification task. The paper proposes a solution inspired by classic label propagation. Nodes far away from decision boundary are assigned heavier weights in learning, whereas ...	3	[{"review_id": "qtP-eFDf87S", "reviewer": "Reviewer_84DB", "summary": "This paper proposes an instance-reweighting scheme for GNNs based on the idea that examples \"inside\" the class's cluster should be more important than ones which are on the boundary. A heuristic (Totoro) is proposed for the class weight, based on...	Topology-Imbalance Learning for Semi-Supervised Node Classification Deli Chen 1.2, Yankai Lin', Guangxiang Zhao', Xuancheng Ren', Peng Li', Jie Zhou', Xu Sun? Pattern Recognition Center, WeChat AI, Tencent Inc., China MOE Key Lab of Computational Linguistics, School of EECS, Peking University {delichen, yankailin, patr...	48,701
wJXWzCsGlZw	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,368	Practical, Provably-Correct Interactive Learning in the Realizable Setting: The Power of True Believers	We consider interactive learning in the realizable setting and develop a general framework to handle problems ranging from best arm identification to active classification. We begin our investigation with the observation that agnostic algorithms \emph{cannot} be minimax-optimal in the realizable setting. Hence, we desi...	[ "JULIAN KATZ-SAMUELS", "Blake Mason", "Kevin Jamieson", "Rob Nowak" ]	[ "Active Learning", "Active Classification", "Multi-Armed Bandits" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper presents an algorithm for solving a set of problems in the realizable interactive learning framework via a new sampling based algorithm. All the reviewers liked the theoretical results in the paper. There was considerable and productive discussion among the authors and the reviewers and in the end all the rev...	4	[{"review_id": "ouh9-TmBgUx", "reviewer": "Reviewer_97Fq", "summary": "The paper considers minimax optimality for some realizable interactive learning settings. A general separation is given between agnostic and realizable learners for the label complexity of active classification. Furthermore efficient algorithms are ...	Practical, Provably-Correct Interactive Learning in the Realizable Setting: The Power of True Believers Julian Katz-Samuels Blake Mason Rice University bm630rice.. edu Kevin Jamieson University of Washington University of Wisconsin, Madison katzsameelsOiisc.cm j jamieson@cs. washington edu Robert Nowak University of Wi...	42,007
xNmhYNQruJX	neurips	2,021	main	NeurIPS.cc/2021/Conference	5,729	Time Discretization-Invariant Safe Action Repetition for Policy Gradient Methods	In reinforcement learning, continuous time is often discretized by a time scale $\delta$, to which the resulting performance is known to be highly sensitive. In this work, we seek to find a $\delta$-invariant algorithm for policy gradient (PG) methods, which performs well regardless of the value of $\delta$. We first i...	[ "Seohong Park", "Jaekyeom Kim", "Gunhee Kim" ]	[ "Reinforcement learning" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper proposes a novel trick to address the time discretization issue with continuous-time policy gradient method. As a small time interval delta is used, the authors shows that variance of PG estimator can exploded when delta->0. To address the issue, they propose a Safe Action Repetition method, which lets the a...	4	[{"review_id": "i9o2suwOxe", "reviewer": "Reviewer_hLjy", "summary": "The paper proposes an alternative to durative actions where instead of outputting an action and how long the action is to persist, it instead outputs an action and a radius. This radius defines a \"safe region\" where the action is to be repeated for...	Time Discretizatin-IIvariint Safe Action Repetition for Policy Gradient Methods Seohong Park Seoul National University artberryx@snu. ac Jaekyeom Kim Seoul National University j jaekyeom@snu. ac.kr Gunhee Kim Seoul National University gunhee@snu. ac Abstract In reinforcement learning, continuous time is often discretiz...	45,501
w5j80GVGFsr	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,601	Stochastic optimization under time drift: iterate averaging, step-decay schedules, and high probability guarantees	We consider the problem of minimizing a convex function that is evolving in time according to unknown and possibly stochastic dynamics. Such problems abound in the machine learning and signal processing literature, under the names of concept drift and stochastic tracking. We provide novel non-asymptotic convergence gua...	[ "Joshua Cutler", "Dmitriy Drusvyatskiy", "Zaid Harchaoui" ]	[ "stochastic gradient", "online tracking", "concept drift", "high-probability bounds" ]	NeurIPS 2021 Poster	Accept (Poster)	All the reviewers were positive about this paper and felt that it presented a good analysis of less-well-studied problem.	4	[{"review_id": "zjZq5Hlc-4A", "reviewer": "Reviewer_WvZE", "summary": "The paper under review is devoted to the analysis of the stochastic approximation algorithm with the presence of the drift term. The main result is to derive finite-time bounds on the expected mean square error. ", "questions": "", "limitations": ""...	Stochastic optimization under time drift: iterate averaging, step decay, and high-probability guarantees Joshua Cutler University of Washington jocutler@uv.edi Dmitriy Drusvyatskly University of Washington ddrusv@uw.co .du Zaid Harchaoui University of Washington zaid@uw.. edu Abstract We consider the problem of minimiz...	34,738
w5fW0TNWPyc	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,079	Machine Learning for Variance Reduction in Online Experiments	We consider the problem of variance reduction in randomized controlled trials, through the use of covariates correlated with the outcome but independent of the treatment. We propose a machine learning regression-adjusted treatment effect estimator, which we call MLRATE. MLRATE uses machine learning predictors of the ou...	[ "Yongyi Guo", "Dominic Coey", "Mikael Konutgan", "Wenting Li", "Chris Schoener", "Matt Goldman" ]	[ "experimentation", "variance reduction", "agnostic statistics", "debiased machine learning", "semiparametrics", "experiment splitting" ]	NeurIPS 2021 Poster	Accept (Poster)	The reviewers praise the quality of the write up (which seems to be a significant achievement wrt to the previous version of the paper) and the crafty combination of existing methods to build an estimator of causal effects. Both theoretical and empirical results are provided that back up the relevance of the contributi...	4	[{"review_id": "m7JLiY2y8jX", "reviewer": "Reviewer_SmDS", "summary": "This paper proposes a new approach for variance reduction in randomized controlled trials. The proposed method first predicts the outcome variable Y from a set of covariates X that are independent of the treatment assignment T, and measures the aver...	Machine Learning for Variance Reduction in Online Experiments Yongyi Guo Department of Operations Research and Financial Engineering Princeton University Princeton, NJ 08544 yongyig@princeton.co Dominic Coey Mikael Konutgan Facebook Facebook Hacker Way, Menlo Park, CA 94025 coeyefb. Hacker Way, Menlo Park, CA 94025 kmi...	43,002
vvi7KqHQiA	neurips	2,021	main	NeurIPS.cc/2021/Conference	5,901	Implicit Bias of SGD for Diagonal Linear Networks: a Provable Benefit of Stochasticity	Understanding the implicit bias of training algorithms is of crucial importance in order to explain the success of overparametrised neural networks. In this paper, we study the dynamics of stochastic gradient descent over diagonal linear networks through its continuous time version, namely stochastic gradient flow. We ...	[ "Scott Pesme", "Loucas Pillaud-Vivien", "Nicolas Flammarion" ]	[ "Implicit bias", "SGD", "diagonal neural networks", "SDE", "implicit regularisation" ]	NeurIPS 2021 Poster	Accept (Poster)	Reviewers generally agreed that the paper is clearly written, technically solid and interesting. Reservations with regards to the significance of the analyzed setting (diagonal linear neural networks) were raised, but given that this setting already received notable attention, the committee reached a decision by which...	4	[{"review_id": "kMSDRc_iVQY", "reviewer": "Reviewer_a887", "summary": "This paper studies the implicit bias of SGD over GD trained solutions for a regression task, with MSE loss, and a quadratic parameterization of the regression coefficients. Building off recent analysis demonstrating how the initialization for GD to...	Implicit Bias of SGD for Diagonal Linear Networks: a Provable Benefit of Stochasticity Scott Pesme EPFL scott coott.pesme@epl.cc Loucas Pillaud-Vivien EPFL loucas loucas.pillauee.. wtw.aapppiiaaddiiiiieeeee...... Nicolas Flammarion EPFL nicolas, . www.mamion@epplllcmmmmmmmm Abstract Understanding the implicit bias of t...	45,180
vrkQ07gp0kq	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,423	TOHAN: A One-step Approach towards Few-shot Hypothesis Adaptation	In few-shot domain adaptation (FDA), classifiers for the target domain are trained with \emph{accessible} labeled data in the source domain (SD) and few labeled data in the target domain (TD). However, data usually contain private information in the current era, e.g., data distributed on personal phones. Thus, the priv...	[ "Haoang Chi", "Feng Liu", "Wenjing Yang", "Long Lan", "Tongliang Liu", "Bo Han", "William Cheung", "James Kwok" ]	[ "transfer learning", "domain adaptation", "few-shot learning" ]	NeurIPS 2021 Spotlight	Accept (Spotlight)	The paper provides a few-shot learning technique that avoids looking at the data obtained to train previous tasks. As such it has promise for privacy preserving settings. The techniques seem solid, as well as the experimental setting. There where a few concerns related to over-promising in the area of differential priv...	4	[{"review_id": "b3g4sjkIyCN", "reviewer": "Reviewer_6L9x", "summary": "This paper focuses on a problem setting where data are not available but classifiers are. The possible benefits of this setting lie in the privacy reservation, i.e., data used in previous few-shot learning may leak the private information. The propo...	TOHAN: A One-step Approach towards Few-shot Hypothesis Adaptation Haoang Chil.2,6* Feng Liu Wenjing Yang'! Long Lan Tongliang Liu", Bo Han William K. Cheung?. James T. Kwok State Key Laboratory of High Performance Computing, College of CS, NUDT CS Department, HKBU DeSI Lab, AAII, Faculty of Engineering and ITT, UTS TML...	46,398
w6U6g5Bvug	neurips	2,021	main	NeurIPS.cc/2021/Conference	11,642	The Utility of Explainable AI in Ad Hoc Human-Machine Teaming	Recent advances in machine learning have led to growing interest in Explainable AI (xAI) to enable humans to gain insight into the decision-making of machine learning models. Despite this recent interest, the utility of xAI techniques has not yet been characterized in human-machine teaming. Importantly, xAI offers the ...	[ "Rohan R Paleja", "Muyleng Ghuy", "Nadun Ranawaka Arachchige", "Reed Jensen", "Matthew Gombolay" ]	[ "Explainable AI", "Human-Machine Teaming", "Situational Awareness", "Ad Hoc Teaming" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper presents a study on the effect of two types of interpretable/explainable models in human-machine teaming. By varying the type of explanation (showing the model for human intention recognition used by the robot and showing the robot's action-selection model), the authors conclude that showing both models incr...	5	[{"review_id": "xC6nN42EVhu", "reviewer": "Reviewer_jmSk", "summary": "The paper presents two user studies to quantify the impact of explanations on Human-AI teaming. The first study experiments with varying cobot (AI) policies to understand how well humans can create a mental model and situational awareness. Results f...	The Utility of Explainable AI in Ad Hoc Human-Machine Teaming Rohan Paleja', Muyleng Ghuy', Nadun R. Arachchige', Reed Jensen', Matthew Gombolay' 'Georgia Institute of Technology, 'MIT Lincoln Laboratory Atllnta, GA 30332, Lexiingonn MA 02420 (rpaleja3, mghuy3, nkra3) n @gatech edu, rjensen@l1. mit. edu mgombolay30gate...	56,829
wF-llA3k32	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,251	Batch Multi-Fidelity Bayesian Optimization with Deep Auto-Regressive Networks	Bayesian optimization (BO) is a powerful approach for optimizing black-box, expensive-to-evaluate functions. To enable a flexible trade-off between the cost and accuracy, many applications allow the function to be evaluated at different fidelities. In order to reduce the optimization cost while maximizing the benefit-...	[ "Shibo Li", "Robert Kirby", "Shandian Zhe" ]	[ "Multi-fidelity Bayesian Optimization", "Batch Acquisition Function", "Full Auto-Regressive Modeling", "Multi-fidelity Max-value Entropy Search" ]	NeurIPS 2021 Poster	Accept (Poster)	Ultimately, I am recommending this paper for acceptance because I believe the approach to multi-fidelity optimization proposed by the authors is sufficiently novel and the experimental evaluation is sufficiently strong, and I am thus in agreement with Reviewer gu5R and Reviewer Hcby that, after substantial incorporatio...	4	[{"review_id": "yu0JbEUDdGe", "reviewer": "Reviewer_Hcby", "summary": "In multi-fidelity black-box optimization, f() can be queried at points x at different levels of precision. Higher levels of precision have higher costs. Prior work on Bayesopt for mfbbo used GP models for multi-fidelity observations that make simpli...	Batch Multi-Fidelity Bayesian Optimization with Deep Auto-Regressive Networks Shibo Li, Robert M. Kirby, and Shandian Zhe School of Computing. University of Utah Salt Lake City, UT 84112 shibolcs. utah. edu, wirkykrrroooo wtw..eaaeeeeo edu, che@cs.utah....................a edu Abstract Bayesian optimization (BO) is pow...	51,289
w-EabDtADg	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,635	Fair Sequential Selection Using Supervised Learning Models	We consider a selection problem where sequentially arrived applicants apply for a limited number of positions/jobs. At each time step, a decision maker accepts or rejects the given applicant using a pre-trained supervised learning model until all the vacant positions are filled. In this paper, we discuss whether the fa...	[ "Mohammad Mahdi Khalili", "Xueru Zhang", "Mahed Abroshan" ]	[ "Supervised Learning", "Fairness", "Sequential Selection" ]	NeurIPS 2021 Poster	Accept (Poster)	Reviewer all agreed that the paper formulates and partially addresses an interesting and practically relevant question: namely, how to define and guarantee an appropriate notion of fairness in sequential decision-making settings where a limited number of positions are available and diversity among the selected is an im...	4	[{"review_id": "IQWlEUrfFW1", "reviewer": "Reviewer_DxNT", "summary": "This paper introduces a new task of making fair decisions in sequential selection problems. The authors have shown that existing fairness notions are not suitable for this problem and could lead to discrimination. Then the authors propose a new fair...	Fair Sequential Selection Using Supervised Learning Models Mohammad Mahdi Khalili Xueru Zhang CSE Department Ohio State University Columbus, OH, USA zhang. 12807 edu Mahed Abroshan Alan Turing Institute London, UK CIS Department University of Delaware Newark, DE, USA khalili@udel. edu mabroshan@turingg.co Abstract We c...	44,891
vrXuRmaU_jM	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,693	Weak-shot Fine-grained Classification via Similarity Transfer	Recognizing fine-grained categories remains a challenging task, due to the subtle distinctions among different subordinate categories, which results in the need of abundant annotated samples. To alleviate the data-hungry problem, we consider the problem of learning novel categories from web data with the support of a c...	[ "Junjie Chen", "Li Niu", "Liu Liu", "Liqing Zhang" ]	[ "transfer learning", "similarity transfer", "weak-shot learning" ]	NeurIPS 2021 Poster	Accept (Poster)	Following the rebuttal and discussion period, all reviewers lean towards acceptance of this work. The reviewers found the problem statement to be of practical importance and the paper well motivated and well written. Questions raised by QXjD and 4wp8 were adequately addressed during the rebuttal/discussion phase relate...	3	[{"review_id": "UzXte_UQ76", "reviewer": "Reviewer_4wp8", "summary": "This paper proposes a new setting called weak-shot fine-grained classification, which contains base classes with clean labels and novel classes with noisy labels. To boost the performance of the noisy novel classes, the paper adopts sample weighting ...	Weak-shot Fine-grained Classification via Similarity Transfer Junjie Chen, Li Niu, Liu Liu, Liqing Zhang" MoE Key Lab of Artificial Intelligence, Department of Computer Science and Engineering, Shanghai Jiao Tong University (chen.bys, ustcnewly, shirlley)asji edu. cn, chang-lq@css saaeelllossssstt... edu. cn Abstract R...	46,731
w1FvEPcwTnI	neurips	2,021	main	NeurIPS.cc/2021/Conference	5,361	Graphical Models in Heavy-Tailed Markets	Heavy-tailed statistical distributions have long been considered a more realistic statistical model for the data generating process in financial markets in comparison to their Gaussian counterpart. Nonetheless, mathematical nuisances, including nonconvexities, involved in estimating graphs in heavy-tailed settings pose...	[ "José Vinícius De Miranda Cardoso", "Jiaxi Ying", "Daniel P. Palomar" ]	[ "graphs", "financial markets", "precision matrix", "Laplacian" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper formulates a graphical model learning problem by maximizing the likelihood of a multivariate student-t distribution subject to constraints on the Laplacian structured parameter matrix. The formulation is non convex. An explicit ADMM iteration is derived to arrive at a stationary point. The primary strength...	3	[{"review_id": "gcEJYtXO8w6", "reviewer": "Reviewer_C54j", "summary": "The paper proposes a new estimator for learning graphical models under the t-distribution assumption for data generation process. For the new estimator, an ADMM-based algorithm is designed to solve the associated optimization problem in order to fin...	Graphical Models in Heavy-Tailed Markets José Vinícius de M. Cardoso, Jiaxi Ying, Daniel P. Palomar Department Electronic and Computer Engineering Department of Industrial Engineering and Decision Analytics The Hong Kong University of Science and Technology Clear Water Bay, Hong Kong SAR China vjvdmc, jx.ying) ust hk, ...	37,311
vqHak8NLk25	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,690	Parameter Prediction for Unseen Deep Architectures	Deep learning has been successful in automating the design of features in machine learning pipelines. However, the algorithms optimizing neural network parameters remain largely hand-designed and computationally inefficient. We study if we can use deep learning to directly predict these parameters by exploiting the pas...	[ "Boris Knyazev", "Michal Drozdzal", "Graham W. Taylor", "Adriana Romero" ]	[ "parameter prediction", "hypernetworks", "graph networks", "computational graphs", "optimization", "meta-learning", "deep networks" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper presents a technique for prediction of parameters in deep architectures. After the initial reviews, the reviewers were concerned about 1) performance in the low-data regime 2) training cost of GHNs 3) relevance/utility for NAS application 4) improvement via finetuning. The authors clarified 1) and 4) well. 2...	4	[{"review_id": "repBJtrMeVX", "reviewer": "Reviewer_mxL9", "summary": "This paper proposes and analyzes a method for generating trained weights already optimized for a given single task for a large class of diverse architectures using Graph hypernetworks. They introduce a new DEEPNETS-1M benchmark with in-distribution ...	Parameter Prediction for Unseen Deep Architectures Boris Knyazev 1.2- Michal Drozdzal Graham W. Taylor 22331t Adriana Romero-Soriano 4.5,f University of Guelph Vector Institute for Artificial Intelligence 3 Canada CIFAR AI Chair Facebook AI Research 5 McGGill University Tequal advising Abstract Deep learning has been s...	68,050
vh7qBSDZW3G	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,576	Efficient constrained sampling via the mirror-Langevin algorithm	We propose a new discretization of the mirror-Langevin diffusion and give a crisp proof of its convergence. Our analysis uses relative convexity/smoothness and self-concordance, ideas which originated in convex optimization, together with a new result in optimal transport that generalizes the displacement convexity of ...	[ "Kwangjun Ahn", "Sinho Chewi" ]	[ "Langevin MCMC", "Sampling", "Optimization", "mirror-Langevin" ]	NeurIPS 2021 Poster	Accept (Poster)	The authors have identified an assumption that extends the work of [Zhang, Peyre, Fadili, Pereyra] to have guarantees for the mirrored Langevin algorithm -- not to be confused from the mirrored Langevin dynamics of [Hsieh et al]. However. the assumption cannot be implemented in practice since the authors cannot do the...	4	[{"review_id": "nFzhYEvaXT6", "reviewer": "Reviewer_UqoF", "summary": "The paper under review is considering a new discretization for the MLD. The motivation comes from solving constrained optimization problem. Under proper assumption, the authors present convergence analysis for the algorithm in terms of the KL and Br...	Efficient constrained sampling via the mirror-Langevin algorithm Kwangjun Ahn Sinho Chewi Department of EECS Department of Mathematics Massachusetts Institute of Technology Massachusetts Institute of Technology Cambridge, MA 02139 kjahn&mit.cc Cambridge, MA 02139 schewi @mit edu Abstract We propose a new discretization...	45,280
wD6GWHqLuhS	neurips	2,021	main	NeurIPS.cc/2021/Conference	7,280	CorticalFlow: A Diffeomorphic Mesh Transformer Network for Cortical Surface Reconstruction	In this paper, we introduce CorticalFlow, a new geometric deep-learning model that, given a 3-dimensional image, learns to deform a reference template towards a targeted object. To conserve the template mesh’s topological properties, we train our model over a set of diffeomorphic transformations. This new implementatio...	[ "Leo Lebrat", "Rodrigo Santa Cruz", "Frederic de Gournay", "Darren Fu", "Pierrick Bourgeat", "Jurgen Fripp", "Clinton Fookes", "Olivier Salvado" ]	[ "3D Deep Learning", "Geometric Deep Learning", "Regular Surface Recosntruction", "Cortical Surface Reconstruction" ]	NeurIPS 2021 Poster	Accept (Poster)	The paper presents a Geometric deep learning model that learns to diffeomorphic deform a regular template mesh towards a targeted object. The paper was found interesting from a methods perspective and it lead to impressive results. The methodology is strong and elegant, making a strong case for a "deep diffeomorphic" a...	4	[{"review_id": "rUHhgPpqs_j", "reviewer": "Reviewer_xE8n", "summary": "This paper focuses on the segmentation of the cortex in 3D brain MRI volumes. The proposed method combines a diffeomorphic registration layer with volumetric U-Net modules to iteratively register a (very smooth, sphere-like) template mes...	CorticalFlow: A Diffeomorphic Mesh Deformation Module for Cortical Surface Reconstruction Léo Lebrat CSIRO, QUT leb0260csiro.. au Rodrigo Santa Cruz CSIRO, QUT Frédéric de Gournay IMT UMRS219 degourna@insa-tool..oo Darren Fu UQ fon0220csiro. Pierrick Bourgeat CSIRO Jurgen Fripp CSIRO Clinton Fookes Olivier Salvado QUT ...	53,118
vsCCDVdTAx	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,534	Human-Adversarial Visual Question Answering	Performance on the most commonly used Visual Question Answering dataset (VQA v2) is starting to approach human accuracy. However, in interacting with state-of-the-art VQA models, it is clear that the problem is far from being solved. In order to stress test VQA models, we benchmark them against human-adversarial exampl...	[ "Sasha Sheng", "Amanpreet Singh", "Vedanuj Goswami", "Jose Alberto Lopez Magana", "Tristan Thrush", "Wojciech Galuba", "Devi Parikh", "Douwe Kiela" ]	[ "vqa", "visual question answering", "adversarial evaluation", "model in the loop" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper introduces a new Adversarial VQA dataset collected with human-and-model in the loop by directly asking humans to write questions to attack SOTA VQA models. After author rebuttal, it has received 4 accept recommendations. All the reviewers are happy about the paper, and agree that this is a solid new benchmar...	4	[{"review_id": "y3WCVkYg4R-", "reviewer": "Reviewer_MEbN", "summary": "This paper proposes AdVQA, a new (adversarial) test set for the Visual Question Answering. Annotators and a SOTA model are put in the loop to make sure the model can be fooled. Experiments show that other VQA models tend to perform poorly on AdVQA,...	Human-Adversarial Visual Question Answering Sasha Sheng! Amanpreet Singh Vedanuj Goswami Jose Alberto Lopez Magana Tristan Thrush Wojeiech Galuba Devi Parikh Douwe Kiela Facebook AI Research Tecnológico de Monterrey 8 Georgia Tech https://adverssaiillq...org Abstract Performance on the most commonly used Visual Questio...	48,262
vwgsqRorzz	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,203	Edge Representation Learning with Hypergraphs	Graph neural networks have recently achieved remarkable success in representing graph-structured data, with rapid progress in both the node embedding and graph pooling methods. Yet, they mostly focus on capturing information from the nodes considering their connectivity, and not much work has been done in representing ...	[ "Jaehyeong Jo", "Jinheon Baek", "Seul Lee", "Dongki Kim", "Minki Kang", "Sung Ju Hwang" ]	[ "Graph Neural Network", "Edge Representation Learning", "Graph Pooling", "Hypergraph" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper proposes edge representation learning framework in graphs based on a hyper graph transformation. The reviewers agreed that the problem motivation has some merits. However, all the reviewers had hard time discerning what’s the novelty of the proposed work, and the significance of the contribution. In future s...	4	[{"review_id": "cyCsgaASaXL", "reviewer": "Reviewer_HuNj", "summary": "This paper introduces an edge representation learning framework using Dual Hypergraph Transformation which transforms input graph to hypergraph and uses graph convolution network on the hypergraph to obtain edge representations.\nThey further introd...	Edge Representation Learning with Hypergraphs Jaehyeong Jo Jinheon Baek'", Seul Lee':, Dongki Kim', Minki Kang', Sung Ju Hwang 1.2 KAIST' AITRICS South Korea harryjo97@kaistt.co ac. kr, jinheon baekOkaist. ac. el1emlee77900gmai. com, cleverki@haist.ac kc. zexc11330kaist.ac. kr, sjhwang620kaist.co ac. ac.kr Abstract Gra...	56,284
vqzAfN-BoA_	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,185	Towards Calibrated Model for Long-Tailed Visual Recognition from Prior Perspective	Real-world data universally confronts a severe class-imbalance problem and exhibits a long-tailed distribution, i.e., most labels are associated with limited instances. The naïve models supervised by such datasets would prefer dominant labels, encounter a serious generalization challenge and become poorly calibrated. W...	[ "Zhengzhuo Xu", "Zenghao Chai", "Chun Yuan" ]	[ "visual recognition", "data imbalance", "mixup", "long tail" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper addresses the problem of long-tailed visual recognition. They proposed two complementary methods to address the imbalance problem: (i) a data augmentation scheme based on the Mixup and (ii) an approach to compensate for bias in class prior. The final proposed pipeline is 1 stage, as opposed to 2-stage which ...	4	[{"review_id": "dJXHPJC8AZ", "reviewer": "Reviewer_1uZE", "summary": "The authors propose two models to improve calibration for long-tailed recognition. 1) a Uniform Mixup is proposed, which adopts an advanced mixing factor and sampler in favor of the minority. 2) the authors propose a Bayias loss to compensate for the...	Towards Calibrated Model for Long-Tailed Visual Recognition from Prior Perspective Zhengzhuo Xu Zenghao Chai'; Chun Yuan 122 Shenzhen International Graduate School, Tsinghua University Peng Cheng Laboratory xzz200mails 1s. tsinghua. edu. cn, zenghaochai @gmail com, yuancesz. tsinghua. edu. cn Abstract Real-world data u...	51,751
wDI6CNTR3yP	neurips	2,021	main	NeurIPS.cc/2021/Conference	2,990	CorticalFlow: A Diffeomorphic Mesh Transformer Network for Cortical Surface Reconstruction	In this paper, we introduce CorticalFlow, a new geometric deep-learning model that, given a 3-dimensional image, learns to deform a reference template towards a targeted object. To conserve the template mesh’s topological properties, we train our model over a set of diffeomorphic transformations. This new implementatio...	[ "Leo Lebrat", "Rodrigo Santa Cruz", "Frederic de Gournay", "Darren Fu", "Pierrick Bourgeat", "Jurgen Fripp", "Clinton Fookes", "Olivier Salvado" ]	[ "3D Deep Learning", "Geometric Deep Learning", "Regular Surface Recosntruction", "Cortical Surface Reconstruction" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper proposes a geometric deep learning approach to process volumetric images to extract mesh representations from them via a set of learned deformations aimed to capture a manifold (namely a surface) from the 3D input image. As the principal application and motivation for this work, it is demonstrated on cortica...	4	[{"review_id": "KRcN48TOliR", "reviewer": "Reviewer_FY8L", "summary": "This paper proposes a method called CorticalFlow, which is a geometric deep learning model to acquire brain surface mesh out of 3D volume images. It preserves topological property of the extracted mesh with relatively small computation using diffeom...	CorticalFlow: A Diffeomorphic Mesh Deformation Module for Cortical Surface Reconstruction Léo Lebrat CSIRO, QUT leb0260csiro.. au Rodrigo Santa Cruz CSIRO, QUT Frédéric de Gournay IMT UMRS219 degourna@insa-tool..oo Darren Fu UQ fon0220csiro. Pierrick Bourgeat CSIRO Jurgen Fripp CSIRO Clinton Fookes Olivier Salvado QUT ...	53,123
w0ZNeU5S-l	neurips	2,021	main	NeurIPS.cc/2021/Conference	1,545	Learning with Algorithmic Supervision via Continuous Relaxations	The integration of algorithmic components into neural architectures has gained increased attention recently, as it allows training neural networks with new forms of supervision such as ordering constraints or silhouettes instead of using ground truth labels. Many approaches in the field focus on the continuous relaxati...	[ "Felix Petersen", "Christian Borgelt", "Hilde Kuehne", "Oliver Deussen" ]	[ "differentiable algorithms", "algorithmic supervision", "continuous relaxation", "weakly supervised", "perturbed", "smooth", "sorting", "differentiable rendering" ]	NeurIPS 2021 Poster	Accept (Poster)	In this paper, the authors propose an approach for smoothing algorithms. While existing works were typically specialised for specific tasks, this works attempts to be more general by smoothing low-level operations such as if-else branches, while loops, and array indexing. The authors successfully demonstrate their appr...	4	[{"review_id": "mysEcdDup3N", "reviewer": "Reviewer_YtiF", "summary": "The paper proposes an approach for continuous relaxations of algorithms that allows their integration into end-to-end trainable neural network architectures. It shows that the general approach can compete with SOTA continuous relaxations and could r...	Learning with Algorithmic Supervision via Continuous Relaxations Felix Petersen Christian Borgelt University of Salzburg christian@borgell net University of Konstanz felix. pew.ix..ttersenuui... Hilde Kuehne Oliver Deussen University of Konstanz oliver siiver.deussen@ui kn University of Frankfurt MIT-IBM Watson AI Lab ...	47,835
vrhNQ7aYSdr	neurips	2,021	main	NeurIPS.cc/2021/Conference	3,273	SubTab: Subsetting Features of Tabular Data for Self-Supervised Representation Learning	Self-supervised learning has been shown to be very effective in learning useful representations, and yet much of the success is achieved in data types such as images, audio, and text. The success is mainly enabled by taking advantage of spatial, temporal, or semantic structure in the data through augmentation. However,...	[ "Talip Ucar", "Ehsan Hajiramezanali", "Lindsay Edwards" ]	[ "representation learning", "self-supervised learning", "contrastive learning", "tabular data", "multi-view learning" ]	NeurIPS 2021 Poster	Accept (Poster)	The authors have addressed most of the reviewer concerns during the review process, and I now suggest the acceptance of the paper. As the reviewers have acknowledged, the performance improvements are notable with an elegant and conceptually appealing method. There is significant content in the author responses that n...	4	[{"review_id": "sOSzBuc2P4", "reviewer": "Reviewer_2zYu", "summary": "The paper proposes SubTab, a self-supervised (pre-training) method for learning good representations of tabular data. The method generates multiple views for each (unlabeled) example by selecting different (possibly overlapping) subsets of features. ...	SubTab: Subsetting Features of Tabular Data for Self-Supervised Representation Learning Talip Uçar, Ehsan Hajiramezanali, Lindsay Edwards Respiratory and Immunology, R&D, AstraZeneca ehsan haj iramezanal,, lindsay edwards {talip. ucar, (talip.ucar, Abstract Self-supervised learning has been shown to be very effective i...	48,691
vlf0zTKa5Lh	neurips	2,021	main	NeurIPS.cc/2021/Conference	4,867	On the Sample Complexity of Learning under Geometric Stability	Many supervised learning problems involve high-dimensional data such as images, text, or graphs. In order to make efficient use of data, it is often useful to leverage certain geometric priors in the problem at hand, such as invariance to translations, permutation subgroups, or stability to small deformations. We study...	[ "Alberto Bietti", "Luca Venturi", "Joan Bruna" ]	[ "invariance", "learning theory", "approximation", "kernels" ]	NeurIPS 2021 Poster	Accept (Poster)	This paper studies kernel ridge regression in the case where the target function is invariant through the group action of $G$, a subgroup of the symmetric group. They explore how making the inner-product kernel invariant too can improve the sample complexity bounds. Furthermore they consider extensions to approximate i...	4	[{"review_id": "z_A-gfoim12", "reviewer": "Reviewer_qjVL", "summary": "This paper proposed a framework for analyzing the generalization behavior of kernel ridge regression under invariance and \"near-invariance\" assumptions with respect to group actions.", "questions": "", "limitations": "", "rating": 8, "confidence":...	On the Sample Complexity of Learning under Invariance and Geometric Stability Alberto Bietti NYU Luea Venturi NYUł 1v8008nyu.ed edu Joan Bruna NYU alberto. alberto.biettinnn wwettiiiit....... brunaâcims nyu. nyu edu Abstract Many supervised learning problems involve high-dimensional data such as images, text, graphs. I...	40,496
vmJs9dyUeWQYe	neurips	2,021	main	NeurIPS.cc/2021/Conference	10,567	Adaptive Denoising via GainTuning	Deep convolutional neural networks (CNNs) for image denoising are typically trained on large datasets. These models achieve the current state of the art, but they do not generalize well to data that deviate from the training distribution. Recent work has shown that it is possible to train denoisers on a single noisy im...	[ "Sreyas Mohan", "Joshua L. Vincent", "Ramon Manzorro", "Peter Crozier", "Carlos Fernandez-Granda", "Eero P Simoncelli" ]	[ "denoising", "adaptation", "out of distribution", "generalization", "gain" ]	NeurIPS 2021 Poster	Accept (Poster)	All four reviewers agreed that the paper was not currently above threshold for acceptance, with a high indicator of confidence in their assessments. In addition to the reviewer comments, one reviewer suggested in a note during discussion three areas for improvement that my be helpful for the authors: "Especially: (1)...	4	[{"review_id": "qTXoHzTzpkW", "reviewer": "Reviewer_2yuX", "summary": "This paper proposes a method to finetune the pre-trained denoising network, which is trained in a supervised manner, in an unsupervised way to adapt the network to the tested noisy image. To avoid overfitting, they propose GainTuning which only adju...	Adaptive Denoising via GainTuning Sreyas Mohan', Joshua L. Vincent", Ramon Manzorro', Peter A. Crozier 2 Carlos Granddl 1.3 Eero P. Simoncelli 1,3,4 Center For Data Science, NYU, Sccooo for Engineering of Matter, Transport and Energy, ASU CCurant Institute of Mathematical Sciences, NYU *Center for Neural Science, NYU a...	50,796

Subsets and Splits

No community queries yet

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