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 | tldr stringlengths 0 250 | primary_area stringclasses 38
values | venue stringclasses 19
values | decision stringclasses 8
values | decision_comment stringlengths 0 9.28k | author_rebuttal stringlengths 0 6k | num_reviews int32 2 8 | reviews_json stringlengths 1.96k 78.7k | markdown stringlengths 0 1.05M | markdown_chars int64 0 1.05M |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
sK6CtXIgKcp | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 463 | Open-set Label Noise Can Improve Robustness Against Inherent Label Noise | Learning with noisy labels is a practically challenging problem in weakly supervised learning. In the existing literature, open-set noises are always considered to be poisonous for generalization, similar to closed-set noises. In this paper, we empirically show that open-set noisy labels can be non-toxic and even benef... | [
"Hongxin Wei",
"Lue Tao",
"RENCHUNZI XIE",
"Bo An"
] | [
"noisy labels",
"weakly supervised learning",
"open-set noisy labels",
"Out-of-Distribution detection",
"deep learning"
] | NeurIPS 2021 Poster | Accept (Poster) | UPDATE: The revision from the authors has been reviewed. After some back-and-forth with the authors to discuss the details of the 300K Random Images dataset that they have chosen to use, the paper has been officially accepted.
----
Reviewers generally appreciated the paper's observation regarding the benefit of open... | 4 | [{"review_id": "sEyGOCFII0Q", "reviewer": "Reviewer_aFDj", "summary": "This work proposes augmenting a training set by using an out of distribution auxiliary dataset to help train a robust classifier through randomly sampling labels for the OOD dataset. Model's trained with this additional data perform better than rela... | Open-set Label Noise Can Improve Robustness Against Inherent Label Noise
Hongxin Wei! Lue Tao Renchunzi Xie Bo An
Schooo of Computer Science and Engineering, Nanyang Technological University, Singapore
2 College of Computer Science and Technology.
Nanjing University of Aeronautics and Astronautics, China 3 MIIT key Lab... | 58,644 | |||
rYhBGWYm6AU | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 11,269 | Intriguing Properties of Contrastive Losses | We study three intriguing properties of contrastive learning. First, we generalize the standard contrastive loss to a broader family of losses, and we find that various instantiations of the generalized loss perform similarly under the presence of a multi-layer non-linear projection head. Second, we study if instance-b... | [
"Ting Chen",
"Calvin Luo",
"Lala Li"
] | [
"contrastive learning",
"contrastive loss",
"feature suppression",
"self-supervised learning",
"computer vision"
] | NeurIPS 2021 Poster | Accept (Poster) | The paper studies three properties of contrastive learning. Reviewers found that although these properties were shown in experiments the paper did not explain why they happen, but thought they can guide better algorithm design in contrastive learning and of interest to a wide audience. Reviewers agreed that this is a ... | 4 | [{"review_id": "odBLoaU_zQe", "reviewer": "Reviewer_cf57", "summary": "This work presents three empirical analyses on contrastive learning algorithms. The first analysis shows that when using a multi-layer non-linear project head, the different ways of instantiating the generalized contrastive learning loss lead to fin... | Intriguing Properties of Contrastive Losses
Ting Chen Google Research namtinnghee&@goglee.. com
Calvin Luo Google Research cal calvinluo@googeecom com
Lala Li Google Research
lala@google. com
Abstract
We study three intriguing properties of contrastive learning. First, generalize the standard contrastive loss to broade... | 40,409 | |||
rHCzkRd0UK | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 9,713 | Linear-Time Gromov Wasserstein Distances using Low Rank Couplings and Costs | The ability to compare and align related datasets living in heterogeneous spaces plays an increasingly important role in machine learning. The Gromov-Wasserstein (GW) formalism can help tackle this problem. Its main goal is to seek an assignment (more generally a coupling matrix) that can register points across otherwi... | [
"Meyer Scetbon",
"Gabriel Peyré",
"marco cuturi"
] | [
"Optimal Transport",
"Gromov-Wasserstein",
"Low-rank methods"
] | NeurIPS 2021 Submitted | Reject | The focus of the submission is speeding up the computation of the Gromov-Wasserstein (GW) distance (capable of capturing the discrepancy of distributions defined on different spaces) using a variant of the Sinkhorn technique. The approach improves the original cubic complexity (w.r.t. the sample size) of GW computation... | 4 | [{"review_id": "joWTOcnypN", "reviewer": "Reviewer_FxAF", "summary": "The paper proposes a new approximating algorithm for Gromov-Wasserstein distance with $L_2$-norm loss that runs in a linear time regarding the number of samples. This algorithm is rooted in the combination of two low-rank techniques: one is the speed... | Linear-Time Gromov Wasserstein Distances using Low Rank Couplings and Costs
Anonymous Author(s)
Affiliation Address
email
Abstract
The ability to compare and align related datascts living in heterogeneous spaces plays an increasingly important role in machine learning. The Gromov- Wasserstein (GW) formalism can help ta... | 44,629 | |||
rKNOWYRaXaJ | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 9,078 | Universal Approximation Using Well-Conditioned Normalizing Flows | Normalizing flows are a widely used class of latent-variable generative models with a tractable likelihood. Affine-coupling models [Dinh et al., 2014, 2016] are a particularly common type of normalizing flows, for which the Jacobian of the latent-to-observable-variable transformation is triangular, allowing the likelih... | [
"Holden Lee",
"Chirag Pabbaraju",
"Anish Prasad Sevekari",
"Andrej Risteski"
] | [
"Generative Models",
"Normalizing Flows",
"Universal Approximation",
"Langevin",
"Dynamical Systems",
"Deep Learning Theory"
] | NeurIPS 2021 Poster | Accept (Poster) | The paper demonstrates that log-concave distributions can be well approximated by well-conditioned affine-coupling flows with Gaussian base density.
This paper is clearly written and technically sound. It provides novel insights and interesting connections between normalising flows, Langevin diffusions, Hénon maps and... | 7 | [{"review_id": "vB063fnMjE9", "reviewer": "Reviewer_PyGo", "summary": "This paper proves that _well-conditioned_ affine-coupling-based normalizing flows can approximate any _log-concave_ distribution. The proof techinques used in this paper combine underdamped Langevin dynamics and methods from dynamical systems in a n... | Universal Approximation For Log-concave Distributions Using Well-conditioned Normalizing Flows
Holden Lee Mathematics Department Duke University Durham, NC 27708 holden. leedduke. edu
Chirag Pabbaraju
Computer Science Department
Stanford University
Stanford, CA 94305
cpabbara@cs stanf ord. edu
Anish Sevekari
Andrej Ris... | 39,106 | |||
rm0I5y2zkG8 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 7,840 | Learning to delegate for large-scale vehicle routing | Vehicle routing problems (VRPs) form a class of combinatorial problems with wide practical applications. While previous heuristic or learning-based works achieve decent solutions on small problem instances, their performance deteriorates in large problems. This article presents a novel learning-augmented local search f... | [
"Sirui Li",
"Zhongxia Yan",
"Cathy Wu"
] | [
"machine learning",
"combinatorial optimization",
"vehicle routing",
"decomposition"
] | NeurIPS 2021 Spotlight | Accept (Spotlight) | This paper proposes and analyzes a learning augmented local search algorithm to solve large-scale VRPs. Overall, all reviewers found novelty and merit in the work that was presented, but a number of questions and potential concerns of the work were posed. In their rebuttal, the author(s) were very thorough and convinci... | 4 | [{"review_id": "m8p1ALihJPV", "reviewer": "Reviewer_oqMs", "summary": "This paper proposes an iterative framework to solve large-scale Vehicle Routing Problems. At each iteration, given the current feasible solution, a learned component selects subroutes that form a VRP subproblem. Then the subproblem is solved using a... | Learning to Delegate for Large-scale Vehicle Routing
Sirui Li" MIT siruil@m memiil@mitteeeee
Zhongxia Yan
MIT zxyanCmit.eet edu
Cathy Wu MIT
cathyru@mit. edu
Abstract
Vehicle routing problems (VRPs) form class of combinatorial problems with wide practical applications. While previous heuristic or learning-basee works a... | 48,736 | |||
r6cNUjS8cm0 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 6,379 | Closing the Gap: Tighter Analysis of Alternating Stochastic Gradient Methods for Bilevel Problems | Stochastic nested optimization, including stochastic compositional, min-max, and bilevel optimization, is gaining popularity in many machine learning applications.
While the three problems share a nested structure, existing works often treat them separately, thus developing problem-specific algorithms and analyses.
A... | [
"Tianyi Chen",
"Yuejiao Sun",
"Wotao Yin"
] | [
"Stochastic bilevel optimization",
"min-max optimization",
"compositional optimization",
"convergence analysis"
] | NeurIPS 2021 Spotlight | Accept (Spotlight) | This paper unifies several SGD-type updates for stochastic nested problems into a single nested SGD approach and presents a tighter sample complexity analysis that improves or matches the existing results. All reviewers give positive evaluation on this paper. So do I. Please update the final version according to the re... | 4 | [{"review_id": "yRaUucka9JF", "reviewer": "Reviewer_r5Xc", "summary": "The author derives a general algorithm framework applicable to minmax, composition, and bi-level optimization. The convergence rates are improved for different classes by leveraging the hidden smoothness of the problem. This paper is well organized ... | where f a'dnd g are differentiable functions; and, & and are random variables. In the optimization is referred to as the stochastic bileve! problem, where the upper-level Closing the Gap: Tighter Analysis of Alternating Stochastic Gradient Methods for Bilevel Problems
Tianyi Chen Rensselaer Polytechnic Institute chenti... | 46,246 | |||
r2uzPR4AYo | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 470 | Accurately Solving Rod Dynamics with Graph Learning | Iterative solvers are widely used to accurately simulate physical systems. These solvers require initial guesses to generate a sequence of improving approximate solutions. In this contribution, we introduce a novel method to accelerate iterative solvers for rod dynamics with graph networks (GNs) by predicting the initi... | [
"Han Shao",
"Tassilo Kugelstadt",
"Torsten Hädrich",
"Wojtek Palubicki",
"Jan Bender",
"Soren Pirk",
"Dominik Michels"
] | [
"Dynamical Systems",
"Representation Learning"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper considers accelerating the simulation of rod dynamics by training a graph neural network to act as an initialisation for an iterative solver. This is an interesting and original paper.
While reviewers were generally in favour of this paper, they had a number of concerns. The primary question, including after... | 4 | [{"review_id": "w-e90unbu8P", "reviewer": "Reviewer_ixu1", "summary": "This paper aims to accelerate the iterative physical simulators using graph neural networks, by combining the traditional iterative rod solver and the graph networks. Instead of an end-to-end way, this paper uses the trained GNN to provide the initi... | Accurately Solving Rod Dynamics with Graph Learning
Han Shao
Tassilo Kugelstadt
KAUST
RWTH Aachen University kugelstadt@cs rwt.ss.thhaaaaee aachhaaahhen.co de
han. shao@kaust edu,
Torsten Hädrich
Wojeiech Palubicki AMU
Jan Bender
KAUST
RWTH Aachen University
torsten. www.ichkkiiitttttccmmm edu.
wp06@amu. edu, pl
Sender... | 51,327 | |||
sLVJXf-BkIt | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 6,635 | Scalable Intervention Target Estimation in Linear Models | This paper considers the problem of estimating the unknown intervention targets in a causal directed acyclic graph from observational and interventional data. The focus is on soft interventions in linear structural equation models (SEMs). Current approaches to causal structure learning either work with known interventi... | [
"Burak Varici",
"Karthikeyan Shanmugam",
"Prasanna Sattigeri",
"Ali Tajer"
] | [
"linear SEMs",
"interventions",
"causal inference"
] | NeurIPS 2021 Poster | Accept (Poster) | The paper proposes an algorithm for estimating the unknown intervention targets in a causal linear structural equation model (SEM) with Gaussian noise. The paper considers observational and interventional data generated under soft interventions.
This paper was viewed as a borderline paper.
The authors learn the set o... | 3 | [{"review_id": "vVYBQ38WIA", "reviewer": "Reviewer_wJ1B", "summary": "The paper consider the problem of learning the interventional Markov equivalence class of causal directed acyclic graph from observational and interventional data where the intervention targets are unknown and there are no unobserved variables. This ... | Scalable Intervention Target Estimation in Linear Models
Burak VarieI Rensselaer Polytechnic Institute varicberpi. edu
Karthikeyan Shanmugam
IBM Research AI
karthikeyan. shanmugam2 iim. com
Prasanna Sattigeri
Ali Tajer
IBM Research AI
Rensselaer Polytechnic Institute
psattiglus ibm. com
tajerêecse. rpi.
Abstract
This p... | 40,843 | |||
rA9HFxFT7th | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 7,375 | Sageflow: Robust Federated Learning against Both Stragglers and Adversaries | While federated learning (FL) allows efficient model training with local data at edge devices, among major issues still to be resolved are: slow devices known as stragglers and malicious attacks launched by adversaries. While the presence of both of these issues raises serious concerns in practical FL systems, no kno... | [
"Jungwuk Park",
"Dong-Jun Han",
"Minseok Choi",
"Jaekyun Moon"
] | [
"Federated Learning",
"Stragglers",
"Adversaries"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper proposes a robust approach to dealing with stragglers and adversaries in federated learning by model grouping and staleness-dependent weighting of models at the server, combined with entropy-based filtering. Stragglers and adversaries are two key concerns in federated learning, and this paper makes progress ... | 4 | [{"review_id": "s1FBb7ak4F", "reviewer": "Reviewer_QeNU", "summary": "This paper proposes a robust FL approach to deal with stragglers and adversaries in federated learning. Model grouping and staleness dependent weighting of models is done at the server to deal with stragglers. For model poising attacks, entropy-based... | Sageflow: Robust Federated Learning against
Both Stragglers and Adversaries
Jungwuk Park' KAIST savertmOkaist.aco ac.
Dong-Jun Han" KAIST djhan930kaist.ac.kk
Minseok Choi
Jeju National University e jaqmf @ e junu, ac kr
Jaekyun Moon KAIST
jmoonkkaist. edu
Abstract
While federated learning (FL) allows efficient model tr... | 44,309 | |||
r1pprsDm185 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 3,612 | SBO-RNN: Reformulating Recurrent Neural Networks via Stochastic Bilevel Optimization | In this paper we consider the training stability of recurrent neural networks (RNNs) and propose a family of RNNs, namely SBO-RNN, that can be formulated using stochastic bilevel optimization (SBO). With the help of stochastic gradient descent (SGD), we manage to convert the SBO problem into an RNN where the feedforwar... | [
"Ziming Zhang",
"Yun Yue",
"Guojun Wu",
"Yanhua Li",
"Haichong Zhang"
] | [
"recurrent neural networks",
"stochastic bilevel optimization",
"training stability"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper proposes a new type of RNN where the hidden state at the current time step is taken to be the result of a single optimizer step applied to a function involving the previous hidden state and the current observation. It is argued that if the optimizer step defining the dynamics uses a learning rate that scales... | 4 | [{"review_id": "seju3_8LowL", "reviewer": "Reviewer_AbCy", "summary": "This paper considered bi-level gradient-based meta-learning for training Recurrent Neural Networks (RNNs). In this paper, the author first formulated the training problem of RNNs as a bi-level meta-learning problem, where the inner level learns the ... | SBO-RNN: Reformulating Recurrent Neural Networks via Stochastic Bilevel Optimization
Ziming Zhang, Yun Yue, Guojun Wu, Yanhua Li, Haichong Zhang
Worcester Polytechnic Institute
Worcester, MA 01609
(zzhang15, yyue, gwu, yli15, hzhang10 0 @wpi edu
Abstract
In this paper consider the training stability of recurrent neural... | 50,624 | |||
qz0MLeaTP1C | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 4,122 | Coresets for Classification – Simplified and Strengthened | We give relative error coresets for training linear classifiers with a broad class of loss functions, including the logistic loss and hinge loss. Our construction achieves $(1\pm \epsilon)$ relative error with $\tilde O(d \cdot \mu_y(X)^2/\epsilon^2)$ points, where $\mu_y(X)$ is a natural complexity measure of the data... | [
"Tung Mai",
"Cameron N Musco",
"Anup Rao"
] | [
"coresets",
"logistic regression",
"importance sampling",
"leverage score sampling",
"lewis weight sampling"
] | NeurIPS 2021 Poster | Accept (Poster) | After the rebuttal phase, all reviewers see the merits of the paper and this coincides with
my own impressions. The paper is worth being published.
| 4 | [{"review_id": "hysVWnsrT7", "reviewer": "Reviewer_mLvM", "summary": "This paper studies construction of coresets for training linear classifiers with objective functions such as logistic loss and hinge loss. The coresets they construct are of size $d\\mu_y(X)^2/\\epsilon^2$ upto polylogarithmic factors where $\\mu_y$ ... | Coresets for Classification - -Simliffied and Strengthened
Tung Mai Adobe Research tumai Cadobe. com
Cameron Musco University of Massachusetts Amherst cmusco@cs. umass. edu
Anup Rao Adobe Research
anuprao@adobe.comm com
Abstract
We give relative егrоr coresets for training linear classifiers with a broad class loss fun... | 33,895 | |||
r7UC-b67YkO | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 5,238 | Rebooting ACGAN: Auxiliary Classifier GANs with Stable Training | Conditional Generative Adversarial Networks (cGAN) generate realistic images by incorporating class information into GAN. While one of the most popular cGANs is an auxiliary classifier GAN with softmax cross-entropy loss (ACGAN), it is widely known that training ACGAN is challenging as the number of classes in the data... | [
"Minguk Kang",
"Woohyeon Joseph Shim",
"Minsu Cho",
"Jaesik Park"
] | [
"Generative Adversarial Networks",
"Conditional Image Generation",
"Adversarial Learning",
"Image Synthesis"
] | NeurIPS 2021 Poster | Accept (Poster) | The authors identify that the auxiliary classifier GAN (ACGAN) fails to scale to more classes and is biased towards easily classifiable examples because ACGAN can suffer from gradient explosion when training the classifier, causing the collapse of training. They then propose to normalize the feature vectors and add a d... | 4 | [{"review_id": "caBRfbggH5f", "reviewer": "Reviewer_t8Ys", "summary": "This paper tackles limitation of ACGAN.\n1. ACGAN struggles when # classes is large.\n2. ACGAN generates easily classifiable samples (=lacks diversity).\n\nTh authors argue that\n1. Gradient exploding in the classifier leads to collapse. Normalizing... | Rebooting ACGAN: Auxiliary Classifier GANs with Stable Training
Minguk Kang
Woohyeon Shim
Minsu Cho
Jaesik Park
Pohang University of Science and Technology (POSTECH), South Korea {mgkang, wh. shim, mscho, beesik.parke @postech. partskaeeeerp ac.kr
Abstract
Conditional Generative Adversarial Networks (cGAN) generate rea... | 52,131 | |||
r6Khc1Lq9z1 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 4,212 | Approximate optimization of convex functions with outlier noise | We study the problem of minimizing a convex function given by a zeroth order oracle that is possibly corrupted by {\em outlier noise}. Specifically, we assume the function values at some points of the domain are corrupted arbitrarily by an adversary, with the only restriction being that the total volume of corrupted po... | [
"Anindya De",
"Sanjeev Khanna",
"Huan Li",
"Hesam Nikpey"
] | [
"gradient descent",
"convex optimization",
"optimization with noise",
"convex interpolation"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper studies the problem of minimizing a convex function given an evaluation oracle to it -- with the twist that the oracle has outlier-noise: for some parameter K, a volume equal to that of a ball of radius K can be corrupted adversarially (thus, when evaluated at any point inside this ball, the oracle can retur... | 3 | [{"review_id": "tzabXqr0_2Q", "reviewer": "Reviewer_CLVB", "summary": "The authors consider a convex minimization problem with a certain noisy oracle. Specifically, the function to be minimized is assumed to be alpha-strongly convex, and beta-Lipschitz (referred to as alpha-beta nice). It is also assumed that we have a... | Approximate optimization of convex functions with outlier noise
Anindya De University of Pennsylvania anindyad@cis upenn edu
Sanjeev Khanna University of Pennsylvania sanjee sanjeevOcis. upenn edu
Huan Li University of Pennsylvania huanli@cis upenn edu
Hesam Nikpey University of Pennsylvania hesam0cis upenn. edu
Abstra... | 38,853 | |||
qQAtFdyDr- | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 6,276 | BatchQuant: Quantized-for-all Architecture Search with Robust Quantizer | As the applications of deep learning models on edge devices increase at an accelerating pace, fast adaptation to various scenarios with varying resource constraints has become a crucial aspect of model deployment. As a result, model optimization strategies with adaptive configuration are becoming increasingly popular. ... | [
"Haoping Bai",
"Meng Cao",
"Ping Huang",
"Jiulong Shan"
] | [
"Joint Neural Architecture Search and Quantization",
"Neural Architecture Search",
"Mixed-Precision Quantization",
"Model Compression"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper combines the idea of mixed precision quantization and single shot NAS method, and proposed a single-shot weight sharing quantization method that requires no training during search. 3 out of 4 reviewers acknowledges that the idea is solid and simple. Concerns remain about the novelty. This paper is recommende... | 5 | [{"review_id": "aURKgEj5o0m", "reviewer": "Reviewer_Q9hH", "summary": "This paper proposed to conduct quantization-aware NAS: in the meanwhile of neural architecture search (basically it is searching subnet within a supernet), it searches for bitwidth allocation for different layer. To stablize the training for quantiz... | BatchQuant: Quantized-for-al A Search with Robust Quantizer
Haoping Bai"
Meng Cao
Ping Huang
Jiulong Shan
Apple
hhopingg bai
mengcao,
huang- -ping
jiulong_shan wwwpssannnnpppppecco com
Abstract
As the applications deep learning models on edge devices increase at an acceler- ating pace, fast adaptation to various scenar... | 45,059 | |||
qZpOqPbwhy | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 3,488 | Learning Riemannian metric for disease progression modeling | Linear mixed-effect models provide a natural baseline for estimating disease progression using longitudinal data. They provide interpretable models at the cost of modeling assumptions on the progression profiles and their variability across subjects. A significant improvement is to embed the data in a Riemannian manif... | [
"Samuel Gruffaz",
"Pierre-Emmanuel Poulet",
"Etienne Maheux",
"Bruno Michel Jedynak",
"Stanley Durrleman"
] | [
"Riemannian Geometry",
"RKHS",
"mixed-effect model",
"Disease progression modelling",
"Longitudinal data"
] | NeurIPS 2021 Poster | Accept (Poster) | Thanks to the authors for their engaging submission on an important topic. There was a lot of positive response in the reviews to this work, and all reviewers were positive about the underlying idea. However, there are areas in which this submission could be strengthened.
A major reviewer concern is understanding t... | 4 | [{"review_id": "vBJjVyh0CuY", "reviewer": "Reviewer_nBtD", "summary": "In this work, the authors proposed a mixed effect model in a Riemannian manifold to estimate disease progression. The proposed model is built upon previous work with an improvement to estimate a Riemannian metric from data. An alternating maximizati... | Learning Riemannian metric for disease progression modeling
Samuel Gruffaz
Pimrne-Emmanuel Poulet
Inria Paris
Paris Brain Institute
Ecole normale supérieure Paclay
Inria Paris
samuel. -ww.ffareens--prri--aaaalaaeeeee sacis-saalaall.l......a fr
pierre-enmanuel
poulet@inria.ff. hwwee--mmmaeelpooleeeenrr.......
Etienne Ma... | 45,972 | |||
qb0qTdxPWzY | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 7,354 | List-Decodable Mean Estimation in Nearly-PCA Time | Robust statistics has traditionally focused on designing estimators tolerant to a minority of contaminated data. {\em List-decodable learning}~\cite{CharikarSV17} studies the more challenging regime where only a minority $\tfrac 1 k$ fraction of the dataset, $k \geq 2$, is drawn from the distribution of interest, and n... | [
"Ilias Diakonikolas",
"Daniel Kane",
"Daniel Kongsgaard",
"Jerry Li",
"Kevin Tian"
] | [
"robust statistics",
"learning theory",
"mixture models",
"semidefinite programming",
"list-decodable learning"
] | NeurIPS 2021 Spotlight | Accept (Spotlight) | This paper gives a nearly "PCA" time algorithm for the problem of list-decodable mean estimation. There are two relevant parameters in the problem: the underlying dimension d, the number of samples n, and the fraction of inliers \alpha. There's a significant history of works on this problem -- starting with polynomial ... | 3 | [{"review_id": "wxPrFJHBV8n", "reviewer": "Reviewer_F5h5", "summary": "This paper studies the list-decodable mean estimation problem and proposes an algorithm that runs in nearly $k$-PCA time, which can be seen as a natural barrier to this problem. ", "questions": "", "limitations": "", "rating": 8, "confidence": 3, "s... | List-Decodable Mean Estimation in Nearly-PCA Time
Ilias Diakonikolas
Daniel M. Kane
Department of Computer Science University of Wisconsin, Madison
Department of Computer Science University of California, San Diego La Jolla, CA 92093 dakane@cs. ucsd. edu
Madison, WI 53706 iliasecs. wwwc.ediii..
Daniel Kongsgaard
Jerry ... | 45,845 | |||
qWtmNpgjD5K | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 10,951 | Impossibility results for fair representation | With the growing awareness to fairness in machine learning and the realization of the central role that data representation has in data processing tasks, there is an obvious interest in notions of fair data representations. We provide a formal framework for examining the fairness of data representations through the len... | [
"Tosca Lechner",
"Nivasini Ananthakrishnan",
"Sushant Agarwal",
"Shai Ben-David"
] | [
"Fairness",
"learning",
"data representation",
"theory",
"sample complexity"
] | NeurIPS 2021 Submitted | Reject | Reviewers thought that the paper addresses an important question in Fair-ML (that is, conditions under which fair data representation leads to fair predictions), and offers potentially significant insights. That said, reviewers found the paper challenging to follow and suggested several concrete steps to improve readab... | 4 | [{"review_id": "zIl585nRiHZ", "reviewer": "Reviewer_58vf", "summary": "This paper formalizes impossibility results for fair representation learning. The authors present a series of settings where various desirable properties cannot be achieved.", "questions": "", "limitations": "", "rating": 6, "confidence": 4, "soundn... | Impossibility results for fair representations
Anonymous Author(s)
Affiliation Address email
Abstract
With the growing awareness to fairesss in machine learning and the realization of the central role that data representation has in data processing tasks, there is obvious interest in notions of fair data representation... | 41,088 | |||
qdphcA9jEbJ | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 6,797 | Recursive Bayesian Networks: Generalising and Unifying Probabilistic Context-Free Grammars and Dynamic Bayesian Networks | Probabilistic context-free grammars (PCFGs) and dynamic Bayesian networks (DBNs) are widely used sequence models with complementary strengths and limitations. While PCFGs allow for nested hierarchical dependencies (tree structures), their latent variables (non-terminal symbols) have to be discrete. In contrast, DBNs al... | [
"Robert Lieck",
"Martin Alois Rohrmeier"
] | [
"Bayesian networks",
"graphical models",
"probabilistic context-free grammars",
"structure learning",
"parsing",
"mixed discrete-continuous",
"sequence models",
"tree induction"
] | NeurIPS 2021 Poster | Accept (Poster) | There have been mixed opinions about the manuscript, even after the short discussion. The paper describes an interesting model, some properties and algorithms. I share some concerns of a reviewer who is unsatisfied about how the work is presented, as it can give the impression to do more than it actually does. Indeed t... | 4 | [{"review_id": "KvQMW7kQng", "reviewer": "Reviewer_1oXa", "summary": "As the title indicates a new formalism: recursive Bayesian networks is\npresented which is basically the union of PCFGs and DBNs. RBNs are\nproperly defined and their relation to PCFGs and DBNs clearly\nexplained (with examples). Since RBNs have both... | Recursive Bayesian Networks: Generalising and Unifying Probabilistic Context-Free Grammars and Dynamic Bayesian Networks
Robert Lieck
Martin Rohrmeier
Digital and Cognitive Musicology Lab
Digital and Cognitive Musicology Lab École Polytechnique Fédérale de Lausanne 1015 Lausanne, Switzerland martin. rwhrmeier@epll..ccm... | 52,527 | |||
rC3zu-OqnII | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 6,495 | A Closer Look at the Worst-case Behavior of Multi-armed Bandit Algorithms | One of the key drivers of complexity in the classical (stochastic) multi-armed bandit (MAB) problem is the difference between mean rewards in the top two arms, also known as the instance gap. The celebrated Upper Confidence Bound (UCB) policy is among the simplest optimism-based MAB algorithms that naturally adapts to ... | [
"Anand Kalvit",
"assaf zeevi"
] | [
"multi-armed bandits",
"learning algorithms",
"UCB",
"Thompson Sampling",
"minimax regret",
"diffusion approximation",
"distribution of arm-pulls"
] | NeurIPS 2021 Spotlight | Accept (Spotlight) | This is strong paper that I would like to see accepted as a spotlight. There is a minor criticism concerning the restriction to two arms but the reviewers feel that this is not a major concern. There are some suggestions by the reviewers of how to improve the paper further and it would be good if the authors could inco... | 4 | [{"review_id": "pm_RS3rq2CH", "reviewer": "Reviewer_15pu", "summary": "The paper studies the arm sampling behavior of UCB and Thompson sampling algorithms. For the two-arm case, the asymptotic behavior of arm sampling is characterized for different regimes (small, large, and medium) of suboptimality gap. Using this cha... | A Closer Look at the Worst-case Behavior of Multi-armed Bandit Algorithms
Anand Kalvit' and Assaf Zeevi²
Columbia University
New York, USA
{ ('akalvit22," assaf Aagsb. col.mbia...leeeeoeooooo.o.ee. edu
Abstract
One of the key drivers of complexity the classical (stochastic) multi-armed bandit (MAB) problem is the diffe... | 49,992 | |||
qPOeyokHXT8 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 7,965 | Provably efficient multi-task reinforcement learning with model transfer | We study multi-task reinforcement learning (RL) in tabular episodic Markov decision processes (MDPs). We formulate a heterogeneous multi-player RL problem, in which a group of players concurrently face similar but not necessarily identical MDPs, with a goal of improving their collective performance through inter-player... | [
"Chicheng Zhang",
"Zhi Wang"
] | [
"Multi-task learning",
"Provably efficient reinforcement learning",
"Model transfer"
] | NeurIPS 2021 Poster | Accept (Poster) | The paper presents a method to transform a multi-task MDP into a multi-player game where agents act in related but slightly environments, and share information. Several reviewers commented on the quality and clarity of writing, and while there were many specific technical questions, they were mostly clarified in the re... | 5 | [{"review_id": "uqALhXrKLnU", "reviewer": "Reviewer_tFPt", "summary": "This paper formulates a multi-task concurrent RL problem, even the tasks are not identical. it also proposes a prototype algorithm for tabular MDPs and provides theory bounds.", "questions": "", "limitations": "", "rating": 5, "confidence": 2, "soun... | Provably Efficient Multi-Tasi Reinforcement Learning with Model Transfer
Chicheng Zhang University of Arizona chi chichengoss arizona. edu
Zhi Wang
University of California San Diego zhiwangdeng ucsd. edu
Abstract
We study multi-task reinforcement learning (RL) in tabular episodic Markov deci- sion processes (MDPs). We... | 42,987 | |||
qwtfY-3ibt7 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 3,172 | MAU: A Motion-Aware Unit for Video Prediction and Beyond | Accurately predicting inter-frame motion information plays a key role in video prediction tasks. In this paper, we propose a Motion-Aware Unit (MAU) to capture reliable inter-frame motion information by broadening the temporal receptive field of the predictive units. The MAU consists of two modules, the attention modul... | [
"Zheng Chang",
"Xinfeng Zhang",
"Shanshe Wang",
"Siwei Ma",
"Yan Ye",
"Xinguang Xiang",
"Wen Gao"
] | [
"Temporal receptive field",
"motion-aware",
"attention mechanism",
"video prediction"
] | NeurIPS 2021 Poster | Accept (Poster) | There was a robust discussion between the reviewers on the merits of this work. The author response helped clarify many of the things that the reviewers asked for. All in all, I feel this work should be accepted at NeurIPS. While the reviewers as a whole felt that there is some lack of novelty in the work, and had some... | 4 | [{"review_id": "wa3tdWik88A", "reviewer": "Reviewer_LrcY", "summary": "The paper proposes the Motion-Aware Unit (MAU) for video prediction. This module attempts to model the temporal information in videos by enlarging the temporal receptive field and aggregating it through an attention mechanism. The proposed module is... | MAU: A Motion-Aware Unit for Video Prediction and Beyond
Zheng Chang
University of Chinese Academy of Sciences Institute of Computing Technology, Chinese Academy of Sciences changzhengl 80ma i ils. ucas.ac.cn
Xinfeng Zhang
Shanshe Wang Institute of Digital Media, Peking University sswangopku. edu cn
School of Computer ... | 42,875 | |||
qpdc7sCpbi | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 1,059 | A Biased Graph Neural Network Sampler with Near-Optimal Regret | Graph neural networks (GNN) have recently emerged as a vehicle for applying deep network architectures to graph and relational data. However, given the increasing size of industrial datasets, in many practical situations, the message passing computations required for sharing information across GNN layers are no longer... | [
"Qingru Zhang",
"David Wipf",
"Quan Gan",
"Le Song"
] | [
"Graph Neural Network",
"Neighbor Sampling",
"Multi-Armed Bandit"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper proposes a novel bandit algorithm for sampling neighbourhoods for GNNs. The reviewers agreed that the work was of good quality, significance and originality. Some reviewers noted, and the authors are encouraged to take on board, that the current write up is quite dense and exposition could be improved by add... | 4 | [{"review_id": "hez7MKKFxob", "reviewer": "Reviewer_rPqa", "summary": "The authors propose Thanos, a bandit-based approach for subsampling graph neighbourhoods, to be used as context for graph neural network predictions. It improves on the prior art (BanditSampler) by proposing trading off increased reward bias for les... | A Biased Graph Neural Network Sampler with Near-Optimal Regret
Qingru Zhang
David Wipf
Quan Gan
Le Song 4.3
Georgia Institute of Technology
2 Amazon Shanghai AI Lab
3 Mohaamed bin Zayed University of Artificial Intelligence qingru. zhangggathhe edu, daviwipf@amazon com quagan@amazon. com, le. se.ong@mmzuai. ac.ae
Abstr... | 45,782 | |||
r-oRRT-ElX | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 10,149 | On Provable Benefits of Depth in Training Graph Convolutional Networks | Graph Convolutional Networks (GCNs) are known to suffer from performance degradation as the number of layers increases, which is usually attributed to over-smoothing. Despite the apparent consensus, we observe that there exists a discrepancy between the theoretical understanding of over-smoothing and the practical capa... | [
"Weilin Cong",
"Morteza Ramezani",
"Mehrdad Mahdavi"
] | [
"graph neural network",
"over-smoothing",
"generalization",
"expressive power",
"optimization"
] | NeurIPS 2021 Poster | Accept (Poster) | This work provides theoretical and empirical evidence that over-smoothing does not necessarily happen in practice: it is shown that a deep GCN is expressive as long as properly trained, as well as that it can converge to a globally optimal solution. The paper also discusses the generalization capability of GCNs.
The r... | 4 | [{"review_id": "ZDVwiynGvp_", "reviewer": "Reviewer_aKcs", "summary": "The authors argue that over-smoothing does not necessarily happen in practice and prove that a deep model is expressive as long as properly trained and can converge to global optimal with linear convergence rate. They also analyze the generalization... | On Provable Benefits of Depth in Training Graph Convolutional Networks
Weilin Cong Penn State
Morteza Ramezani
Mehrdad Mahdavi Penn State mzm6160psu. edu
Penn State morteza@cse. psu. edu
wxc2720psu.ede edu
Abstract
Graph Convolutional Networks (GCNs) known to suffer from performance degradation as the number of layers ... | 57,332 | |||
rh0vIXw6i33 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 2,228 | Multiple Descent: Design Your Own Generalization Curve | This paper explores the generalization loss of linear regression in variably parameterized families of models, both under-parameterized and over-parameterized. We show that the generalization curve can have an arbitrary number of peaks, and moreover, the locations of those peaks can be explicitly controlled. Our result... | [
"Lin Chen",
"Yifei Min",
"Misha Belkin",
"amin karbasi"
] | [
"generalization",
"double descent",
"multiple descent"
] | NeurIPS 2021 Poster | Accept (Poster) | We thank the authors for this submission. Overall, the paper presents an interesting perspective on the fact that the generalization curve can have multiple descents, and that the locations can be explicitly controlled.
The paper well-motivates the approach. The authors have provided extensive responses to the concer... | 4 | [{"review_id": "xTNIJtSK0x", "reviewer": "Reviewer_qLTq", "summary": "I have read other reviewers' comments and the author response.\nI lower my score due to the following reasons: 1) changing the feature dimension $d$ doesn't change the groundtruth function, which makes the problem setting questionable. 2) the respons... | Multiple Descent: Design Your Own Generalization Curve
Lin Chen
Yifei Min
Simons Institute for the Theory of Computing University of California, Berkeley CA 94720 1in. wwwn@berkeeey...... edu
Department of Statistics and Data Science Yale University CT 06511 yifei weteiiimi@yall...o
Mikhail Belkin
Amin Karbasi
Halıroğl... | 45,342 | |||
qcjOWDHAc4J | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 2,798 | Social Processes: Self-Supervised Forecasting of Nonverbal Cues in Social Conversations | The default paradigm for the forecasting of human behavior in social conversations is characterized by top-down approaches. These involve identifying predictive relationships between low level nonverbal cues and future semantic events of interest (e.g. turn changes, group leaving). A common hurdle however, is the limit... | [
"Chirag Raman",
"Hayley Hung",
"Marco Loog"
] | [
"Social Behavior Forecasting",
"Situated Interactions",
"Neural Processes",
"Free-standing Conversations",
"Self-supervised Learning"
] | NeurIPS 2021 Submitted | Reject | This paper proposes a novel problem involving the prediction of nonverbal behaviors in social settings, and introduces a neural process model fit to the problem. Reviewers had no major methodological concerns, but raised concerns about novelty in the context of a machine learning methods venue, and about clarity. An ex... | 4 | [{"review_id": "xSylz7W0xkb", "reviewer": "Reviewer_u6ot", "summary": "This work proposes the Social Process (SP) models to predict human behaviors in social conversations.", "questions": "", "limitations": "", "rating": 4, "confidence": 4, "soundness": null, "presentation": null, "contribution": null, "strengths": "",... | Social Processes: Self-Supervised Forecasting of Nonverbal Cues in Social Conversations
Anonymous Author(s)
Affiliation
Address
ema i
Abstract
The default paradigm for the forecasting of human behavior in social conversations is characterized by top-down approaches. These involve identifying predictive relationships be... | 61,658 | |||
qxKh67NNJ2I | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 3,490 | Safe Reinforcement Learning with Natural Language Constraints | While safe reinforcement learning (RL) holds great promise for many practical applications like robotics or autonomous cars, current approaches require specifying constraints in mathematical form. Such specifications demand domain expertise, limiting the adoption of safe RL. In this paper, we propose learning to interp... | [
"Tsung-Yen Yang",
"Michael Hu",
"Yinlam Chow",
"Peter Ramadge",
"Karthik R Narasimhan"
] | [
"Safe reinforcement learning",
"Language grounding"
] | NeurIPS 2021 Spotlight | Accept (Spotlight) | This paper presents a framework wherein constraints (particularly safety constraints) on a reinforcement learning agent can be specified using natural language. The paper is well written, clear, and was well-received by the reviewers, all of whom recommend acceptance. The AC finds the idea of natural language constrain... | 3 | [{"review_id": "wAfyZkFLepe", "reviewer": "Reviewer_8Sns", "summary": "This paper contributes to the area of constrained RL by adding machinery to interpret and apply constraints specified in natural language. They also design and implement a testbed world to evaluate such systems. In this world, they demonstrate tha... | Safe Reinforcement Learning with Natural Language Constraints
T'sun-Yeen Yang" Princeton University ty3@princeton.. edu
Michael Hu"
Princeton University
michael hw@yobi..bb... ventures
Yinlam Chow
Google Research
yin 1 Manllamhh@@@@oogle..co com
Peter J. Ramadge
Karthik Narasimhan
Princeton University
Princeton Univers... | 54,732 | |||
qe9z54E_cqE | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 452 | Post-Training Sparsity-Aware Quantization | Quantization is a technique used in deep neural networks (DNNs) to increase execution performance and hardware efficiency. Uniform post-training quantization (PTQ) methods are common, since they can be implemented efficiently in hardware and do not require extensive hardware resources or a training set. Mapping FP32 mo... | [
"Gil Shomron",
"Freddy Gabbay",
"Samer Kurzum",
"Uri Weiser"
] | [
"quantization",
"post-training",
"hardware",
"systolic array",
"tensor core",
"performance",
"sparsity"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper explores the activation sparsity at both bit-level and numerical-level for 8-bit post-training quantization. For bit-level, a dynamic quantization technique is proposed to dynamically examine and trim leading zero-bits, which enabling a multi-scale quantization and a 2x4bit-8bit MAC unit. For numerical level... | 4 | [{"review_id": "z-xE8rJgF6", "reviewer": "Reviewer_8ZWQ", "summary": "This paper presents a sparsity-aware post-training quantization approach, which considers the sparsity on both the digit representation level and the activation level. The algorithm thus can select the most significant bits from the original 8-bit va... | Sparsity-Aware Quantization
Gil Shomron
Freddy Gabbay
Samer Kurzum
Uri Weiser
Technion - Israel Institute Technology, Haifa, Israel
Ruppin Academic Center, Emek Hefer, Israel
{gilsho@campue, ssamer15@campus, wri.weisereee.... wwwhnion.cc........
freddygōruppin.ac.o
Abstract
Quantization technique used in deep neural ne... | 41,314 | |||
q_fMLfwTAJY | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 3,108 | Scalable Rule-Based Representation Learning for Interpretable Classification | Rule-based models, e.g., decision trees, are widely used in scenarios demanding high model interpretability for their transparent inner structures and good model expressivity. However, rule-based models are hard to optimize, especially on large data sets, due to their discrete parameters and structures. Ensemble method... | [
"Zhuo Wang",
"Wei Zhang",
"Ning Liu",
"Jianyong Wang"
] | [
"interpretable classification",
"rule-based model",
"representation learning",
"scalability"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper presents a new scalable classifier, named Rule-based Representation Learner (RRL), that can automatically learn interpretable rules for data representation and classification. A new gradient-based discrete model training method, i.e., Gradient Grafting, is proposed as well that directly optimizes the discret... | 4 | [{"review_id": "XytOkTTwaM8", "reviewer": "Reviewer_v1eM", "summary": "The paper introduces a gradient-descent and network-based training method to learn a rule-based interpretable classifier by training concurrently a continuous model and a discretized version.", "questions": "", "limitations": "", "rating": 7, "confi... | Scalable Rule-Based Representation Learning for Interpretable Classification
Zhuo Wang', Wei Zhang'; Ning Liu', Jianyong Wangl
'Deparmeent of Computer Science and Technology, Tsinghua University
Jiangsu Collaborative Innovation Center for Language Ability, Jiangsu Normal University Schooo of Computer Science and Techno... | 48,092 | |||
rD6ulZFTbf | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 11,626 | Successor Feature Landmarks for Long-Horizon Goal-Conditioned Reinforcement Learning | Operating in the real-world often requires agents to learn about a complex environment and apply this understanding to achieve a breadth of goals. This problem, known as goal-conditioned reinforcement learning (GCRL), becomes especially challenging for long-horizon goals. Current methods have tackled this problem by au... | [
"Christopher Hoang",
"Sungryull Sohn",
"Jongwook Choi",
"Wilka Torrico Carvalho",
"Honglak Lee"
] | [
"successor features",
"goal-conditioned RL",
"graph-based planning"
] | NeurIPS 2021 Poster | Accept (Poster) | All reviewers agree that this is a strong, clearly described empirical work. Although it uses mostly pre-existing concepts, it combines them in an effective way, and its well-executed experiments demonstrate impressive results on pixel-based benchmarks. The ablation studies conducted as part of the discussion with revi... | 4 | [{"review_id": "wb_VDIjQJwP", "reviewer": "Reviewer_YnN2", "summary": "This paper presents Successor Feature Landmarks (SFL), a new graph-based planning framework based on successor features for long-horizon goal-conditioned RL (GCRL). The key idea is to use successor features to define a new metric, Successor Feature ... | Successor Feature Landmarks for Long-Horizon Goal-Conditioned Reinforcement Learning
Christopher Hoang
Sungryull Sohn 12
Jongwook Choi
Wilka Carvalho
Honglak Lee 12
University of Michigan [choang, srsohn, juook, wcarvalh, honglak @umi ch. edu
LG Al Research
Abstract
Operating in the real-world often requires agents to ... | 45,545 | |||
qKRr_rNCEPz | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 8,514 | BooVI: Provably Efficient Bootstrapped Value Iteration | Despite the tremendous success of reinforcement learning (RL) with function approximation, efficient exploration remains a significant challenge, both practically and theoretically. In particular, existing theoretically grounded RL algorithms based on upper confidence bounds (UCBs), such as optimistic least-squares val... | [
"Boyi Liu",
"Qi Cai",
"Zhuoran Yang",
"Zhaoran Wang"
] | [
"Reinforcement Learning",
"Exploration"
] | NeurIPS 2021 Poster | Accept (Poster) | The paper improves the state-of-the-art analysis of randomized RL algorithms with function approximation. The authors should improve the presentation and add more empirical support. | 4 | [{"review_id": "n-fTCl3JvQo", "reviewer": "Reviewer_PuQ1", "summary": "This paper proposes a modified version of LSVI based on bootstrapping, called BooVI. Different from the famous Bootstrapped DQN, it shows that BooVI is provably efficient under linear MDP; meanwhile, although not theoretically supported, an implemen... | Boo VI: Provably Efficient Bootstrapped Value Iteration
Boyi Liu"
Qi Cait
Zhuoran Yang!
Zhaoran Wang
Abstract
Despite the tremendous success reinforcement learning (RL) with function ap- proximation, efficient exploration remains significant challenge, both practically and theoretically. In particular, existing theoret... | 38,237 | |||
qGeqg4_hA2 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 6,488 | Post-processing for Individual Fairness | Post-processing in algorithmic fairness is a versatile approach for correcting bias in ML systems that are already used in production. The main appeal of post-processing is that it avoids expensive retraining. In this work, we propose general post-processing algorithms for individual fairness (IF). We consider a settin... | [
"Felix Petersen",
"Debarghya Mukherjee",
"Yuekai Sun",
"Mikhail Yurochkin"
] | [
"algorithmic fairness",
"graph Laplacian",
"post-processing",
"fairness",
"individual fairness"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper proposes a post-processing method for enforcing individual fairness via graph-based smoothness. All four reviewers ultimately recommended acceptance. The main concern was novelty, which was partially (but not completely) addressed in the author response, but was not deemed significant enough to prevent accep... | 4 | [{"review_id": "xWYlhTRYvPY", "reviewer": "Reviewer_YBcz", "summary": "This paper proposes a post-processing method GLIF in enforcing individual fairness. The key idea is to build a similarity graph based on an oracle fair metric and then apply laplacian regularization on model output and the similarity graph. The opti... | Post-processing for Individual Fairness
Felix Petersen" University of Konstanz felix. www.reeeuuiiiii..... kn
Debarghya Mukherjee' University Michigan mdeb@umich ch. edu
Yuekai Sun
Mikhail Yurochkin
University of Michigan yuekai @umi ch. edu
IBM Research, MIT-IBM Watson AI Lab mikhail, yurocckioiih1n1 bm. com
Abstract
... | 45,406 | |||
qDrpme0FAi | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 1,109 | Mining the Benefits of Two-stage and One-stage HOI Detection | Two-stage methods have dominated Human-Object Interaction~(HOI) detection for several years. Recently, one-stage HOI detection methods have become popular. In this paper, we aim to explore the essential pros and cons of two-stage and one-stage methods. With this as the goal, we find that conventional two-stage methods ... | [
"Aixi Zhang",
"Yue Liao",
"Si Liu",
"Miao Lu",
"Yongliang Wang",
"Chen Gao",
"XIAOBO LI"
] | [
"Human-object Interaction Detection"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper presents work on human-object interaction (HOI) detection. The main contributions include an analysis of two-stage versus one-stage HOI detection methods and a multi-layer transformer architecture that achieves state of the art results. The initial reviews pointed to concerns over clarity of some terminolo... | 4 | [{"review_id": "svptOumhwOO", "reviewer": "Reviewer_ez6V", "summary": "Inspired by its success on various vision and NLP tasks, Transformer based architectures have recently been adopted and achieved impressive performance on the task of HOI detection [3, 27, 38]. This paper builds upon these work and further innovates... | Mining the Benefits of Two-stage and One-stage HOI Detection
Aixi Zhang'*
Yue Liao Si Liu
Miao Lu
Yongliang Wang'
Chen Gao Xiaobo Li'
Alibaba Group
"Beihang University
Abstract
Two-stage methods have dominated Human-Object Interaction (HOI) detection for several years. Recently, one-stage HOI detection methods have bec... | 46,731 | |||
q1yLPNF0UFV | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 4,452 | On the Out-of-distribution Generalization of Probabilistic Image Modelling | Out-of-distribution (OOD) detection and lossless compression constitute two problems that can be solved by the training of probabilistic models on a first dataset with subsequent likelihood evaluation on a second dataset, where data distributions differ. By defining the generalization of probabilistic models in terms o... | [
"Mingtian Zhang",
"Andi Zhang",
"Steven McDonagh"
] | [
"OOD detection",
"OOD generalizationm",
"Lossless compression",
"Generative model"
] | NeurIPS 2021 Poster | Accept (Poster) | The main concerns about this work shared by the reviewers were around novelty and presentation. One reviewer felt that a key idea underlying the work has already featured in several other works in recent years, and thus the novelty of the work is limited. I am inclined to agree with this, however, I believe this work c... | 4 | [{"review_id": "wu5R3og72Sk", "reviewer": "Reviewer_t8Bq", "summary": "The paper discussed the counterintuitive phenomenon in OOD detection using generative models, and proposed a new method called non-local feature density which fixed the issue and achieved the SOTA performance, because non-local features are shown to... | On the Out-of-distributio Generalization of Probabilistic Image Modelling
Mingtian Zhang 1,2+
Andi Zhang 2,3+
Steven McDonagh 2
1 AI Center, University College London, Huawei Noah's 's Ark Lab, 3 Department of Computer Science and Technology, University of Cambridge steven stteen. www.eemccoona@hhuuee...com
mingtian. z... | 46,445 | |||
qL_juuU4P3Y | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 10,921 | Partition and Code: learning how to compress graphs | Can we use machine learning to compress graph data? The absence of ordering in graphs poses a significant challenge to conventional compression algorithms, limiting their attainable gains as well as their ability to discover relevant patterns. On the other hand, most graph compression approaches rely on domain-dependen... | [
"Giorgos Bouritsas",
"Andreas Loukas",
"Nikolaos Karalias",
"Michael M. Bronstein"
] | [
"lossless graph compression",
"neural compression",
"graph neural networks"
] | NeurIPS 2021 Poster | Accept (Poster) | The paper presents learning-based method for compressing graphs. The method proceeds by partitioning the input graph into simpler parts, and using a learned dictionary to encode frequently appearing subgraphs. Empirical evaluation demonstrates improvements over multiple baselines.
The main concern was that the sizes ... | 4 | [{"review_id": "XC89GeGcDHX", "reviewer": "Reviewer_Q2XE", "summary": "The paper studies the lossless compression of graphs with deep learning. The authors propose to compress the graphs in 3 steps: a partitioning algorithm decomposes the graph into elementary structures, which are mapped to the elements of a small d... | Partition and Code: learning how to compress graphs
Giorgos Bouritsas" Imperial College London, UK g. bouritsas@imperial. ac uk
Andreas Loukas EPFL, Switzerland andreas. www.kluueuflllccomm ch
Nikolaos Karalias
Michael M. Bronstein
EPFL. Switzerland nikolaos www.
Imperial College London Twitter, UK wwwnsteiiiinnerri...... | 64,107 | |||
qGvMv3undNJ | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 10,022 | Moser Flow: Divergence-based Generative Modeling on Manifolds | We are interested in learning generative models for complex geometries described via manifolds, such as spheres, tori, and other implicit surfaces.
Current extensions of existing (Euclidean) generative models are restricted to specific geometries and typically suffer from high computational costs. We introduce Moser F... | [
"Noam Rozen",
"Aditya Grover",
"Maximilian Nickel",
"Yaron Lipman"
] | [
"generative models",
"manifolds",
"normalizing flows"
] | NeurIPS 2021 Oral | Accept (Oral) | The following is a summary of the pros and cons that resulted from the reviews and the discussion period.
pros:
* novel and nontrivial contributions to the flow literature. (R6hn, oTmD, STbS, VH2c, 96Cx)
- reduces computational cost to train CNFs by no longer needing to solve an ODE.
- likelihood computation do... | 5 | [{"review_id": "uDh2xjItd-g", "reviewer": "Reviewer_VH2c", "summary": "The paper tackles the problem of learning flow-based generative models. The proposed approach, a new instance of continuous normalizing-flow methods, models the density as a difference between the prior and a divergence term, parameterized by a neur... | Moser Flow: Divergence-based Generative Modeling on Manifolds
Noam Rozen'
Aditya Grover
Maximilian Nickel?
Yaron Lipman" 1.2 *CLA
Weizmann Institute of Science
Facebook Al Research
Abstract
We are interested in learning generative models for complex geometries described via manifolds, such spheres, tori, and other impl... | 36,950 | |||
q6h7jVe0wE3 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 3,459 | Preserved central model for faster bidirectional compression in distributed settings | We develop a new approach to tackle communication constraints in a distributed learning problem with a central server. We propose and analyze a new algorithm that performs bidirectional compression and achieves the same convergence rate as algorithms using only uplink (from the local workers to the central server) comp... | [
"Constantin Philippenko",
"Aymeric Dieuleveut"
] | [
"Federated Learning",
"Bidirectional Compression",
"Data Heterogeneity",
"Optimization"
] | NeurIPS 2021 Poster | Accept (Poster) | This work proposes MCM, a method that performs bidirectional (i.e., both uplink and downlink) compression in distributed learning. A benefit of the approach is that it is able to match convergence rates of methods that use compression in only a single direction. The reviewers were generally in agreement that the work w... | 5 | [{"review_id": "cQAxq1fd7iW", "reviewer": "Reviewer_NENJ", "summary": "This paper proposed to compress model different for improving downlink compression. The theoretical analysis on convex problems shows that the dominant term does not depend on the compression error of the downlink, thus improving upon on the existin... | Preserved central model for faster bidirectional compression in distributed settings
Constantin Philippenko Aymeric Dieuleveut CMAP, École Polytechnique, Institut Polytechnique de Paris [f [fistname] [lastname] @polytechnique. edu
Abstract
We develop new approach tackle communication constraints in distributed learning... | 49,391 | |||
q0h6av9Vi8 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 2,982 | Light Field Networks: Neural Scene Representations with Single-Evaluation Rendering | Inferring representations of 3D scenes from 2D observations is a fundamental problem of computer graphics, computer vision, and artificial intelligence. Emerging 3D-structured neural scene representations are a promising approach to 3D scene understanding. In this work, we propose a novel neural scene representation, L... | [
"Vincent Sitzmann",
"Semon Rezchikov",
"William T. Freeman",
"Joshua B. Tenenbaum",
"Fredo Durand"
] | [
"Neural Fields",
"Neural Scene Representations",
"Neural Implicit Representations",
"Coordinate-based representations",
"Neural Rendering",
"Light Fields"
] | NeurIPS 2021 Spotlight | Accept (Spotlight) | The submission was thoroughly reviewed and discussed. All four reviewers support acceptance. The work was found to be stimulating and can usefully inform follow-up efforts in this area. The AC supports the reviewers' recommendation. The authors are encouraged to thoroughly address the reviewers' concerns and recommenda... | 4 | [{"review_id": "wfpn8vxvw0f", "reviewer": "Reviewer_9CQx", "summary": "The paper presents a new neural scene representation based on the idea of light fields. Rather than predicting properties (e.g. occupancies, colors) for points in space, the paper proposes to predict such entities for all rays in a scene using simpl... | Light Field Networks: Neural Scene Representations with Single-Evaluation Rendering
Vincent Sitzmann +* si tzmann@mi edu
Semon Rezchikoy skr@math. columbia wwluuaaaaeiii..... edu
William T. Freeman
1.4.5 Joshua B. Tenenbaum 1.4.5
Frédo Durand'
billffmit. edd
jbt@mit. edu
fredo@mit edu
MIT CSAIL
Columbia University IAFI... | 49,225 | |||
q1eCa1kMfDd | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 5,909 | Flattening Sharpness for Dynamic Gradient Projection Memory Benefits Continual Learning | The backpropagation networks are notably susceptible to catastrophic forgetting, where networks tend to forget previously learned skills upon learning new ones. To address such the 'sensitivity-stability' dilemma, most previous efforts have been contributed to minimizing the empirical risk with different parameter regu... | [
"Danruo DENG",
"Guangyong Chen",
"Jianye HAO",
"Qiong Wang",
"Pheng-Ann Heng"
] | [
"continual learning",
"weight loss landscape",
"dynamic gradient projection memory",
"sharpness flatten"
] | NeurIPS 2021 Poster | Accept (Poster) | Echoing what the reviewers highlighted as well. I think the idea of this work is quite novel, and the paper is well executed, where there was due diligence in terms of explaining the hypothesis, and doing ablation studies. As it stands, I think the work is definitely interesting for the community and should be accepted... | 4 | [{"review_id": "ekh1Pa9V_ws", "reviewer": "Reviewer_zbwu", "summary": "The article proposes a continual learning method that innovatively improves on an existing algorithm. Precisely, the current work takes GPM (Gradient Projection Memory) which they show is good at preserving performance on old tasks (good stability),... | Flattening Sharpness for Dynamic Gradient Projection Memory Benefits Continual Learning
Danruo Deng', Guangyong Chen?; Jianye Hao Qiong Wang", Pheng-Ann Heng 'The Chinese University of Hong Kong, *Guangdong-Hong Kong-Macao Joint Laboratory of Human-Machine Intelligence-Syeeet Synergy Systems, Shenzhen Institute of Adva... | 44,204 | |||
pzmwfDLoANS | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 6,018 | Proxy-Normalizing Activations to Match Batch Normalization while Removing Batch Dependence | We investigate the reasons for the performance degradation incurred with batch-independent normalization. We find that the prototypical techniques of layer normalization and instance normalization both induce the appearance of failure modes in the neural network's pre-activations: (i) layer normalization induces a coll... | [
"Antoine Labatie",
"Dominic Masters",
"Zach Eaton-Rosen",
"Carlo Luschi"
] | [
"Batch-Independent Normalization",
"Batch Normalization",
"Layer Normalization",
"Group Normalization",
"Instance Normalization",
"Deep Learning Theory",
"CNNs",
"ResNets",
"ResNeXts",
"EfficientNets"
] | NeurIPS 2021 Poster | Accept (Poster) | The paper proposes a theoretical analysis on different normalization schemes. It demonstrates that normalizing schemes which don't rely on batch-statistics suffer from channel collapse or lack or expressivity. It then proposes a novel normalization scheme to address those issues.
Although the practicality of the algo... | 3 | [{"review_id": "yVZmwC-Ftm", "reviewer": "Reviewer_AgAg", "summary": "This work proposes some theoretical analyses on different normalization schemes in deep neural networks. It found that LN suffers from the collapsing effect over the number of layers, which makes the computation more and more linear, whereas IN suffe... | Proxy-Normalizing Activations to Match Batch Normalization while Removing Batch Dependence
Antoine Labatie
Zach Eaton-Rosen Carlo Luschi Graphcore Research, UK antoine wabatiieecaarralens.... {dominicm, zacher, carlo))graphcoree .i
Dominic Masters
Abstract
We investigate the reasons for the performance degradation incu... | 63,961 | |||
pu6loAVvBZb | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 7,337 | Robust Regression Revisited: Acceleration and Improved Estimation Rates | We study fast algorithms for statistical regression problems under the strong contamination model, where the goal is to approximately optimize a generalized linear model (GLM) given adversarially corrupted samples. Prior works in this line of research were based on the \emph{robust gradient descent} framework of \cite{... | [
"Arun Jambulapati",
"Jerry Li",
"Tselil Schramm",
"Kevin Tian"
] | [
"robust statistics",
"stochastic optimization",
"linear regression",
"acceleration"
] | NeurIPS 2021 Poster | Accept (Poster) | The reviewers all agree that this paper makes solid contributions to the robust estimation problem and improves upon existing work in several aspects. The authors should incorporate the reviewers' suggestions into the final version of the paper. | 4 | [{"review_id": "r2McGEyR6G8", "reviewer": "Reviewer_Xq7Q", "summary": "This paper resolves an open question of the recent work [PSBR 20], which uses robust gradient descent for robust estimation tasks, to achieve faster accelerated convergence rate and better statistical accuracy. This is done via a clever modification... | Robust Regression Revisited: Acceleration and Improved Estimation Rates
Arun Jambulapati Stanford University jublpati@staan ord. edu
Jerry Li Microsoft Research jerr1@mi www.emmmccoott.oommm com
Tselil Schramm Stanford University tselil drelil@stanfor ord. edu
Kevin Tian
Stanford University kjtian@stanford.edh
Abstract... | 50,164 | |||
q7wQ3Z6_keU | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 5,549 | Towards Understanding Cooperative Multi-Agent Q-Learning with Value Factorization | Value factorization is a popular and promising approach to scaling up multi-agent reinforcement learning in cooperative settings, which balances the learning scalability and the representational capacity of value functions. However, the theoretical understanding of such methods is limited. In this paper, we formalize a... | [
"Jianhao Wang",
"Zhizhou Ren",
"Beining Han",
"Jianing Ye",
"Chongjie Zhang"
] | [
"Multi-Agent Reinforcement Learning",
"Reinforcement Learning Theory",
"Value Factorization",
"Fitted Q-Iteration"
] | NeurIPS 2021 Poster | Accept (Poster) | The reviewers and AC discussed the paper. There was agreement that the approach leads to useful insight about value function decomposition in MARL. The author response was also very helpful in clarifying several points about the paper. Still, a number of improvements need to be made. Some of these updates have already ... | 4 | [{"review_id": "dOaEUyUbrf2", "reviewer": "Reviewer_8PFC", "summary": "The paper proposes the multi-agent version of fitted q learning to analyze two kinds of value decomposition. They prove that on-policy training or richer joint value function classes can improve its local or global convergence properties.", "questio... | Towards Understanding Cooperative Multi-Agent Q-Learning with Value Factorization
Jianhao Wang', Zhizhou Ren Beening Han', Jianing Ye', Chongjie Zhang'
Institute for Interdisciplinary Information Sciences, Tsinghua University
2 Department of Computer Science, University of Illinois at Urbana-Champaign wjh19@maill tsing... | 53,861 | |||
pvCLqcsLJ1N | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 3,660 | Remember What You Want to Forget: Algorithms for Machine Unlearning | We study the problem of unlearning datapoints from a learnt model. The learner first receives a dataset $S$ drawn i.i.d. from an unknown distribution, and outputs a model $\widehat{w}$ that performs well on unseen samples from the same distribution. However, at some point in the future, any training datapoint $z \in S... | [
"Ayush Sekhari",
"Jayadev Acharya",
"Gautam Kamath",
"Ananda Theertha Suresh"
] | [
"Machine unlearning",
"theory",
"differential privacy",
"generalization guarantee",
"right to be forgotten"
] | NeurIPS 2021 Poster | Accept (Poster) | The paper considers the problem of machine unlearning: after a model is trained on a dataset, there is a request to delete a point from the dataset. The goal is to design an efficient method to update the trained model s.t. is nearly indistinguishable from what we would have obtained had we trained on the dataset witho... | 4 | [{"review_id": "j9POb7JNYsi", "reviewer": "Reviewer_yPiR", "summary": "The paper considers the problem of machine unlearning: after a model is trained on a dataset, there is a request to delete a point from the dataset. The goal is to design an *efficient* method to update the trained model such that is nearly indistin... | Remember What You Want to Forget:
Algorithms for Machine Unlearning
Ayush Sekhari Cornell University as36630corme11. edu
Jayadev Acharya' Cornell University acharya@cornell..o edu
Gautam Kamath University of Waterloo gocsail mii.. edu
Ananda Theertha Suresh" Google Research, NY cheertha@google.com com
Abstract
We study... | 42,555 | |||
q2JWz371le | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 6,722 | How to transfer algorithmic reasoning knowledge to learn new algorithms? | Learning to execute algorithms is a fundamental problem that has been widely studied. Prior work (Veličković et al., 2019) has shown that to enable systematic generalisation on graph algorithms it is critical to have access to the intermediate steps of the program/algorithm. In many reasoning tasks, where algorithmic-s... | [
"Louis-Pascal A. C. Xhonneux",
"Andreea Deac",
"Petar Veličković",
"Jian Tang"
] | [
"Learning algorithms",
"Transfer",
"Graph Neural Networks"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper investigates how algorithms for which we have access to the execution trace can be leveraged to learn to solve similar tasks for which we do not have execution traces. The authors create a dataset that covers 9 algorithms and 3 different graph types to investigate transfer learning in two major classes of gr... | 4 | [{"review_id": "Zgu_57tlU4", "reviewer": "Reviewer_JA8z", "summary": "The paper studies a transfer learning setting for learning graph algorithms, where we do not have the execution traces for the target algorithm, but we have execution traces for a related algorithm. The paper uses an adapted version of NeuralExecutor... | How to transfer algorithmic reasoning knowledge to learn new algorithms?
Louis-Pascal A. Xhonneux*
Andreea Deac Université de Montréal
Université de Montréal Mila xhonneul @malammmaa....mllll emila.quebec
Mila deacandr@mila.com Quebe
Petar Veličković DeepMind, London UK petarvggoogle.cmm com
Jian Tang
HEC Montréal
Mila... | 44,588 | |||
pk4q0SD_r1X | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 3,891 | COCO-LM: Correcting and Contrasting Text Sequences for Language Model Pretraining | We present a self-supervised learning framework, COCO-LM, that pretrains Language Models by COrrecting and COntrasting corrupted text sequences. Following ELECTRA-style pretraining, COCO-LM employs an auxiliary language model to corrupt text sequences, upon which it constructs two new tasks for pretraining the main mod... | [
"Yu Meng",
"Chenyan Xiong",
"Payal Bajaj",
"saurabh tiwary",
"Paul N. Bennett",
"Jiawei Han",
"Xia Song"
] | [
"Language Model Pretraining"
] | NeurIPS 2021 Poster | Accept (Poster) | This work improves ELECTRA language pretraining approach by introducing Corrective Language Model task (so the model can generate words) and Sequence Contrastive Learning (so sentence representations are more informative). I agree with the reviewers that the authors have conducted extensive experiments to show that the... | 4 | [{"review_id": "wnaifbo287w", "reviewer": "Reviewer_SaZ9", "summary": "This paper introduces COCO-LM, a pretraining method. The pretraining has two components: *COrrecting*, and *COntrasting*. The COrrecting objective shares most of the efficiency advantages of ELECTRA, while still allowing some language modelling abil... | COCO-LM: Correcting and Contrasting Text
Sequences for Language Model Pretraining
Yu Meng'; Chenyan Xiong', Payal Bajaj', Saurabh Tiwary?,
Paul Bennett", Jiawel Han", Xia Song"
University of Illinois at Urbana-Champaign
Microsoft
1 [yumeng5, (hanj)@illll
2 { chenyan xiong, payal. bajaj, satiwary, paul.n. www.ettt,,tttt... | 47,570 | |||
pZHGKM9mAp | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 1,436 | Learning Interpretable Decision Rule Sets: A Submodular Optimization Approach | Rule sets are highly interpretable logical models in which the predicates for decision are expressed in disjunctive normal form (DNF, OR-of-ANDs), or, equivalently, the overall model comprises an unordered collection of if-then decision rules. In this paper, we consider a submodular optimization based approach for lear... | [
"Fan Yang",
"Kai He",
"Linxiao Yang",
"Hongxia Du",
"Jingbang Yang",
"Bo Yang",
"Liang Sun"
] | [
"interpretable",
"rule set",
"submodular",
"rule-based model"
] | NeurIPS 2021 Spotlight | Accept (Spotlight) | The authors provide an interesting connection between an interpretable decision rule set and submodular optimization. Technically, they either consider regularized submodular maximization framework or the difference of two submodular functions.
I found the formalization very interesting. Also, the authors had a succe... | 3 | [{"review_id": "oR3RbK9VvUA", "reviewer": "Reviewer_4Q8k", "summary": "This paper proposes an approach for learning decision rule sets that combines both rule generatino and rule selection. this is accomplished by formulating the an outer subproblem as regularized submodular maximization and an inner problem as maximiz... | Learning Interpretable Decision Rule Sets: A Submodular Optimization Approach
Fan Yang, Kai He, Linxiao Yang, Hongxia Du, Jingbang Yang, Bo Yang, Liang Sun DAMO Academy, Alibaba Group, Hangzhou, China (fanyang . yf, kai. he, linxia. ylx, hongxia dhx, jingbang. mwwai. yb, ww..lii...... sun] Galibaba-ing com
Abstract
Rul... | 46,139 | |||
pl2WX3riyiq | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 2,743 | How Should Pre-Trained Language Models Be Fine-Tuned Towards Adversarial Robustness? | The fine-tuning of pre-trained language models has a great success in many NLP fields. Yet, it is strikingly vulnerable to adversarial examples, e.g., word substitution attacks using only synonyms can easily fool a BERT-based sentiment analysis model. In this paper, we demonstrate that adversarial training, the prevale... | [
"Xinshuai Dong",
"Anh Tuan Luu",
"Min Lin",
"Shuicheng YAN",
"Hanwang Zhang"
] | [
"Natural Language Processing",
"Pre-trained Language Models",
"Adversarial Robustness"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper describes a new training technique more robust adversarial training approach. The method performs well under a variety of scenarios. The paper is thorough and clear. In post-rebuttal discussion a few points were raised, which I would like the authors to include in their paper.
Several reviewers wondered ... | 3 | [{"review_id": "oskL_m1pJy", "reviewer": "Reviewer_jFCo", "summary": "This paper argues that applying adversarial training into the fine-tuning of the pre-trained language model suffers severely from catastrophic forgetting. It then proposes Robust Informative Fine-Tuning (RIFT) to encourage an objective model to retai... | How Should Pre-Trained Language Models Be Fine-Tuned Towards Adversarial Robustness?
Xinshuai Dong Nanyang Technological University & Sea Al Lab Setggiinsuai@outloooo com
Luu Anh Tuan
Nanyang Technological University anhtuan, luu@ntu. edu. sg
Min Lin Sea AI Lab
Shuicheng Yan
Hanwang Zhang
Sea AI Lab yansc&sea. com
Nany... | 47,184 | |||
q6Kknb68dQf | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 817 | Curriculum Offline Imitating Learning | Offline reinforcement learning (RL) tasks require the agent to learn from a pre-collected dataset with no further interactions with the environment.
Despite the potential to surpass the behavioral policies, RL-based methods are generally impractical due to the training instability and bootstrapping the extrapolation e... | [
"Minghuan Liu",
"Hanye Zhao",
"Zhengyu Yang",
"Jian Shen",
"Weinan Zhang",
"Li Zhao",
"Tie-Yan Liu"
] | [
"reinforcement learning",
"offline reinforcement learning"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper generated an interesting discussion between the reviewers and the authors. Especially, the authors made a very good job at addressing reviewers' concerns. They ran new experiments, finetuned the baselines and conducted a thorough ablation study. The paper is now considered as presenting a SotA algorithm for ... | 3 | [{"review_id": "Nm5dAygtZB7", "reviewer": "Reviewer_8Yop", "summary": "This paper proposed and interesting algorithm to do imitation learning with an automatic curriculum learning technique. In particular an automatic filtering of better trajectories under the current policy and a selection of the best ones in that se... | Curriculum Offline Imitating Learning
Minghuan Liu' Hanye Zhao Zhengyu Yang' Jian Shen'
Weinan Zhang"t Li Zhao Tan Liu
Shanghai Jiao Tong University, 2 Microsoft Research
(minghuanliu, fineartz, zyyang, rockyshen, ynzhang]@s1 @sjtu. besjttu.du.cco cn, [lizo,tyliu] www.fff....c.comm
Abstract
Offline reinforcement learni... | 43,833 | |||
pbfAgoc_l2w | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 10,234 | Sequence-to-Sequence Learning with Latent Neural Grammars | Sequence-to-sequence learning with neural networks has become the de facto standard for sequence modeling. This approach typically models the local distribution over the next element with a powerful neural network that can condition on arbitrary context. While flexible and performant, these models often require large d... | [
"Yoon Kim"
] | [
"sequence-to-sequence learning",
"compositional generalization",
"synchronous grammars"
] | NeurIPS 2021 Spotlight | Accept (Spotlight) | The proposed hierarchical approach to sequence-to-sequence learning with synchronous grammars is interesting and looks flexible. Empirical results on compositional generalization benchmarks are good. All reviewers agree the acceptance. I hope the authors find the reviews are helpful to further polish this submission wh... | 4 | [{"review_id": "v0EetB2k-F4", "reviewer": "Reviewer_sHgm", "summary": "This work proposes neural QCFG (quasi-synchronous context-free grammars) for sequence-to-sequence learning. By explicitly modeling the alignment between source and target trees, they aim to improve the generalization and interpretability of the mode... | Sequence-to-Sequenee Learning with Latent Neural Grammars
Yoon Kim MIT CSAIL yoonkimemit.co edu
Abstract
Sequence-to-sequee learning with neural networks has become the de facto standard for sequence prediction tasks. This approach typically models the local distribution over the next word with powerful neural network ... | 64,240 | |||
q4Dln9kWFA0 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 7,433 | Heterogeneous Multi-player Multi-armed Bandits: Closing the Gap and Generalization | Despite the significant interests and many progresses in decentralized multi-player multi-armed bandits (MP-MAB) problems in recent years, the regret gap to the natural centralized lower bound in the heterogeneous MP-MAB setting remains open. In this paper, we propose BEACON -- Batched Exploration with Adaptive COmmuni... | [
"Chengshuai Shi",
"Wei Xiong",
"Cong Shen",
"Jing Yang"
] | [
"Multi-agent System",
"Multi-armed Bandits",
"Decentralized Learning"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper looks at the multi-player multi-arm bandit problem, with heterogenous rewards. This specific problem of reaching the centralized case with decentralized protocols was still open, while it was settled with homogenous rewards.
This result requires combining additional, non trivial techniques from combinatori... | 4 | [{"review_id": "mLBouzjqAA", "reviewer": "Reviewer_Vyvg", "summary": "Overall, this paper introduces a new algorithm BEACON for multi-agent multi-armed bandits that closes the gap between the state-of-the-art upper bound and information theoretic lower bound of the regrets by implicit communication via adaptive differe... | Heterogeneous Multi-player Multi-armed Bandits: Closing the Gap and Generalization
Chengshuai Shi University of Virginia cs7ync@virginia edu
Wei Xiong The Hong Kong University of Science and Technology wxiongae@connect. ust hk
Cong Shen University of Virginia cong@virginia. edu
Jing Yang
The Pennsylvania State Universi... | 47,329 | |||
prVxS4W_ds | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 3,817 | Near-Optimal Lower Bounds For Convex Optimization For All Orders of Smoothness | We study the complexity of optimizing highly smooth convex functions. For a positive integer $p$, we want to find an $\epsilon$-approximate minimum of a convex function $f$, given oracle access to the function and its first $p$ derivatives, assuming that the $p$th derivative of $f$ is Lipschitz.
Recently, three indep... | [
"Ankit Garg",
"Robin Kothari",
"Praneeth Netrapalli",
"Suhail Sherif"
] | [
"Convex optimization",
"Oracle complexity",
"Lower bounds",
"Acceleration"
] | NeurIPS 2021 Spotlight | Accept (Spotlight) | Thank you for submitting your paper to NeurIPS'21.
Despite some minor concerns that should be addressed in the camera-ready version, the reviewers agreed that the paper is worthy of acceptance and is presented at NeurIPS'21. Congratulations! | 3 | [{"review_id": "qSop2F1R9aR", "reviewer": "Reviewer_2gRK", "summary": "This paper stuides the complexity of optimizing highly smooth convex functions. For a positive integer p, one wants to find an $\\epsilon$-approximate minimum of a convex function f, given oracle aceess to the function and its first p derivatives, a... | Near-Optimal Lower Bounds For Convex Optimization For All Orders of Smoothness
Ankit Garg Microsoft Research India Bengaluru, KA, India gar pargammicrosoft. com
Robin Kothari
Microsoft Quantum and Microsoft Research Redmond, WA, USA
robin.kothar www.aokthhrriie fitchari@miie crosoft.commmmmmooo..oom com
Praneeth Netrap... | 32,426 | |||
pvjfA4wogD6 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 1,944 | Video Instance Segmentation using Inter-Frame Communication Transformers | We propose a novel end-to-end solution for video instance segmentation (VIS) based on transformers.
Recently, the per-clip pipeline shows superior performance over per-frame methods leveraging richer information from multiple frames.
However, previous per-clip models require heavy computation and memory usage to achi... | [
"Sukjun Hwang",
"Miran Heo",
"Seoung Wug Oh",
"Seon Joo Kim"
] | [
"video",
"instance segmentation",
"video instance segmentation",
"tracking",
"transformers"
] | NeurIPS 2021 Poster | Accept (Poster) | None of the reviewers recommended accepting this paper.
After reading the author response and other reviews one of the reviewers also reduced their score.
One of the common critiques of the work was around the degree of novelty provided by the work.
One of the initially more positive reviewers did recognize that the me... | 4 | [{"review_id": "ZMErSbcgMdm", "reviewer": "Reviewer_pxiA", "summary": "The paper extends transformer-based video instance segmentation framework VisTR in three aspects. First, the intra- (i.e., space) and inter-frame (i.e., time) relation computations in the transformer encoder are decomposed. Specifically, the extra m... | Video Instance Segmentation using
Inter-Frame Communication Transformers
Sukjun Hwang'
Miran Heo' Yonsei University
Seoung Wug Oh Adobe Research
Seon Joo Kim
{sj hwang, miran, ponjoookim@@yoneeee ac. kr
seoh@adobe.com com
Abstract
We propose novel end-to-end solution for video instance segmentation (VIS) based on trans... | 38,273 | |||
pZ5X_svdPQ | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 790 | Understanding Negative Samples in Instance Discriminative Self-supervised Representation Learning | Instance discriminative self-supervised representation learning has been attracted attention thanks to its unsupervised nature and informative feature representation for downstream tasks. In practice, it commonly uses a larger number of negative samples than the number of supervised classes. However, there is an incons... | [
"Kento Nozawa",
"Issei Sato"
] | [
"representation learning",
"self-supervised representation learning"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper aims to understand a discrepancy between the theory literature (namely, Arora et al., 2019) and recent empirical results in self-supervised learning. Theory suggests that adding more negatives to contrastive learning should degrade downstream classification performance while empirical results have shown the ... | 3 | [{"review_id": "bWxncVN58Xc", "reviewer": "Reviewer_QhKs", "summary": "This paper provides a new theoretical framework to analyze an empirical observation common in self-supervised representation learning, i.e., using a large number of negative samples than the number of supervised classes can improve performance. Howe... | Understanding Negative Samples in Instance Diseriminative Self-supervised Representation Learning
Kento Nozawa
Issei Sato The University Tokyo satoßg. ecc. u-tokyo ac ac.
The University of Tokyo RIKEN AIP nzwēg. ecc. u-tokuo. acc.u-tokyo.ac.jpp jp.
Abstract
Instance discriminative self-supervised representation learnin... | 48,348 | |||
qGn3Rlgul5F | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 8,446 | Parametrized Quantum Policies for Reinforcement Learning | With the advent of real-world quantum computing, the idea that parametrized quantum computations can be used as hypothesis families in a quantum-classical machine learning system is gaining increasing traction. Such hybrid systems have already shown the potential to tackle real-world tasks in supervised and generative ... | [
"Sofiene Jerbi",
"Casper Gyurik",
"Simon Callum Marshall",
"Hans J Briegel",
"Vedran Dunjko"
] | [
"reinforcement learning",
"quantum computing",
"parametrized quantum circuits",
"quantum neural networks",
"policy gradient",
"quantum machine learning",
"quantum reinforcement learning",
"quantum",
"variational quantum circuits"
] | NeurIPS 2021 Poster | Accept (Poster) | The authors present a quantum reinforcement learning regime that utilizes a parameterized quantum circuit. The paper is clearly written. There are a few concerns with the scientific contributions and side points from the reviewers, especially why the PQC plays an essential role in the learning model. However, the autho... | 3 | [{"review_id": "iOVgeGxBKV-", "reviewer": "Reviewer_e14U", "summary": "## Summary \n\nThe paper proposes a policy-based training framework with variational quantum circuits (VQC). Although the VQC-based deep reinforcement learning algorithm has been first studied in [20] and [23], this paper further addresses a policy-... | Parametrized Quantum Policies for Reinforcement Learning
Sofiene Jerbi
Casper Gyurik LIACS, Leiden University
Simon C. Marshall LIACS, Leiden University
Institute for Theoretical Physics, University of Innsbruck sof ac.
Hans J. Briegel
Vedran Dunjko LIACS, Leiden University
Institute for Theoretical Physics, University... | 52,611 | |||
pSitk34qYit | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 8,077 | Risk Bounds and Calibration for a Smart Predict-then-Optimize Method | The predict-then-optimize framework is fundamental in practical stochastic decision-making problems: first predict unknown parameters of an optimization model, then solve the problem using the predicted values. A natural loss function in this setting is defined by measuring the decision error induced by the predicted ... | [
"Heyuan Liu",
"Paul Grigas"
] | [
"contextual stochastic optimization",
"predict-then-optimize",
"learning theory",
"risk bounds"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper builds upon a prior work on Smart Predict-Then-Optimize (SPO) framework. Many practical problems can be described as 2 step procedures: 1. predict some quantity 2. solve optimization problem with predictions from step 1 being inputs. Elmachtoub and Grigas earlier proposed SPO where we take into account step ... | 3 | [{"review_id": "wLC7RjI-RNO", "reviewer": "Reviewer_7rEw", "summary": "The paper develops the risk bounds and uniform calibration results for the surrogate SPO+ loss relative to the SPO loss in the predict-then-optimize framework under the assumption of a linear loss function. The results are first obtained for a polyh... | Risk Bounds and Calibration for a Smart Predict-then-Optimize Method
Heyuan Liu University of California, Berkeley
Paul Grigas
University of California, Berkeley Berkeley, CA 94720
Berkeley, CA 94720 heyuan_ leweeerkeleyeeeeeeeee..... edu edu
pgrigas@berkelley edu
Abstract
The predict-then-optimize framework is fundame... | 45,891 | |||
pTe-8qCdDqy | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 2,748 | Improving Conditional Coverage via Orthogonal Quantile Regression | We develop a method to generate prediction intervals that have a user-specified coverage level across all regions of feature-space, a property called conditional coverage. A typical approach to this task is to estimate the conditional quantiles with quantile regression---it is well-known that this leads to correct cove... | [
"Shai Feldman",
"Stephen Bates",
"Yaniv Romano"
] | [
"Quantile Regression",
"Pinball Loss",
"Conditional Coverage",
"Conformal Prediction",
"Uncertainty Estimation"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper proposes a new type of regularization for quantile regression such that the accuracy of finite sample conditional coverage is improved. The proposed regularization is built on an interesting notion about independence between the size of the intervals and the indicator of a miscoverage event. All reviewers ag... | 3 | [{"review_id": "wEHQRRG1M6e", "reviewer": "Reviewer_CaJn", "summary": "The authors propose a novel method to construct prediction intervals with pre-specified conditional coverage probability. Starting point of the paper is the observation that the length $|\\hat{C}(X)|$ and conditional coverage indicator $1\\{Y \\in \... | Improving Conditional Coverage via Orthogonal Quantile Regression
Shai Feldman
Stephen Bates
Department of Computer Science Technion, Israel
Departments of Statistics and of EECS
UC Berkeley
shai. fe 1 dman@cs. technion.ac.il
stephenbaateslcs.beerre
edu
Yaniv Romano
Departments of Electrical and Computer Engineering an... | 43,750 | |||
pbAmqUUHsQ | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 2,024 | Continuous Mean-Covariance Bandits | Existing risk-aware multi-armed bandit models typically focus on risk measures of individual options such as variance. As a result, they cannot be directly applied to important real-world online decision making problems with correlated options.
In this paper, we propose a novel Continuous Mean-Covariance Bandit (CMCB)... | [
"Yihan Du",
"Siwei Wang",
"Zhixuan Fang",
"Longbo Huang"
] | [
"risk-aware bandits",
"mean-covariance metric"
] | NeurIPS 2021 Poster | Accept (Poster) | The paper considers the problem of regret minimization on bandits with continuous arms. The key idea is to use the correlation between arms. It shows matching upper and lower regret bounds for full information and semi-bandit settings, and upper bound for the full bandit setting.
Three expert reviewers considered the ... | 3 | [{"review_id": "iyFPbWPUdn8", "reviewer": "Reviewer_nzbx", "summary": "The focus of this paper is the management of risk in stochastic multi-armed bandits, specifically when the reward distributions of the arms are correlated (as can happen in practical portfolio management scenarios). This problem requires a solution ... | Continuous Mean-Covariance Bandits
Yihan Du
Siwei Wang CST, Tsinghua University
MSS, Tsinghua University
Beijing, China
Beijing, China
duyh1 1oyy1880mils tsinghua. edu.
wangsw20200mail wtt.imaa...tstnnnaaaeeae tsinghua.een edu. en
Zhixuan Fang
Longbo Huang"
III.. Tsinghua University, Beijing, China Shanghai Qi Zhi Inst... | 42,965 | |||
qLpJ0VWRuWk | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 6,894 | Universal Approximation Using Well-Conditioned Normalizing Flows | Normalizing flows are a widely used class of latent-variable generative models with a tractable likelihood. Affine-coupling models [Dinh et al., 2014, 2016] are a particularly common type of normalizing flows, for which the Jacobian of the latent-to-observable-variable transformation is triangular, allowing the likelih... | [
"Holden Lee",
"Chirag Pabbaraju",
"Anish Prasad Sevekari",
"Andrej Risteski"
] | [
"Generative Models",
"Normalizing Flows",
"Universal Approximation",
"Langevin",
"Dynamical Systems",
"Deep Learning Theory"
] | NeurIPS 2021 Poster | Accept (Poster) | The paper introduces a universal approximation proof of normalizing for log-concave distribution when the Jacobians are not ill-conditioned.
While the scope of the paper remains limited (log-concave distribution), the submission gives a clear exposition of their techniques and could provide a strong foundation for fut... | 3 | [{"review_id": "XpZ9Et6sa3", "reviewer": "Reviewer_Hnha", "summary": "The authors provide a proof for the universal approximation capability of well-conditioned affine coupling network. Their proof requires Gaussian padding, and uses techniques from dynamical systems and stochastic differential equations.", "questions"... | Universal Approximation For Log-concave Distributions Using Well-conditioned Normalizing Flows
Holden Lee Mathematics Department Duke University Durham, NC 27708 holden. leedduke. edu
Chirag Pabbaraju
Computer Science Department
Stanford University
Stanford, CA 94305
cpabbara@cs stanf ord. edu
Anish Sevekari
Andrej Ris... | 39,083 | |||
pZQrKCkbas | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 9,830 | Extending Lagrangian and Hamiltonian Neural Networks with Differentiable Contact Models | The incorporation of appropriate inductive bias plays a critical role in learning dynamics from data. A growing body of work has been exploring ways to enforce energy conservation in the learned dynamics by encoding Lagrangian or Hamiltonian dynamics into the neural network architecture. These existing approaches are b... | [
"Yaofeng Desmond Zhong",
"Biswadip Dey",
"Amit Chakraborty"
] | [
"Deep Model Learning",
"Physics-based Priors",
"Contact Models"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper proposes a differentiable contact model that can be used together with previously proposed Lagrangian and Hamiltonian neural networks, enabling gradient-based system identification and model-based control for hybrid dynamical systems. All reviewers recommend acceptance and by the end of the rebuttal process ... | 5 | [{"review_id": "nlxu-U-aD9X", "reviewer": "Reviewer_GQFZ", "summary": "The paper proposes a differentiable contact model, whose unknown parameters (restitution and friction coefficients) can be learned together with other system parameters using Hamiltonian and/or Lagrangian neural networks. The proposed contact model ... | Extending Lagrangian and Hamiltonian Neural Networks with Differentiable Contact Models
Yaofeng Desmond Zhongi, Biswadip Dey Amit Chakrabortyl
SSiemens Technology. Princeton, NJ 08536, USA.
Kyaofeng.zhone riswadip.dey amit. haat.cccarbbboryy@iieeeeeee com
Abstract
The incorporation of appropriate inductive bias plays c... | 50,342 | |||
pUZBQd-yFk7 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 3,232 | Robustifying Algorithms of Learning Latent Trees with Vector Variables | We consider learning the structures of Gaussian latent tree models with vector observations when a subset of them are arbitrarily corrupted. First, we present the sample complexities of Recursive Grouping (RG) and Chow-Liu Recursive Grouping (CLRG) without the assumption that the effective depth is bounded in the numbe... | [
"Fengzhuo Zhang",
"Vincent Tan"
] | [
"graphical model",
"latent trees",
"arbitrary corruptions"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper studies the problem of learning Gaussian latent tree models where the nodes are vector-valued random variables and there are corruptions. There were differing opinions among the reviewers, but in aggregate they agreed that the extension of existing works (notably the algorithms of Choi et al.) from the scala... | 3 | [{"review_id": "nxVJcjvjYdb", "reviewer": "Reviewer_fher", "summary": "In this paper, the authors consider the problem of learning Gaussian latent tree models when the nodes of the tree are vector variables and the observations are arbitrarily corrupted. The paper uses the algorithms of Choi et al. (2011) as the base a... | Robustifying Algorithms of Learning Latent Trees with Vector Variables
Fengzhuo Zhang
Department of Electrical and Computer Engineering National University of Singapore fzzhang@u. nus edu
.us.edu
Vincent Y. F. Tan Department of Electrical and Computer Engineering Department of Mathematics
National University of Singapo... | 38,355 | |||
rG2ponW2Si | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 230 | Direct Multi-view Multi-person 3D Pose Estimation | We present Multi-view Pose transformer (MvP) for estimating multi-person 3D poses from multi-view images. Instead of estimating 3D joint locations from costly volumetric representation or reconstructing the per-person 3D pose from multiple detected 2D poses as in previous methods, MvP directly regresses the multi-perso... | [
"Tao Wang",
"Jianfeng Zhang",
"Yujun Cai",
"Shuicheng YAN",
"Jiashi Feng"
] | [
"Human Pose Estimation",
"3D",
"Multi-view",
"Transformer",
"End-to-End",
"Direct",
"Camera Ray"
] | NeurIPS 2021 Poster | Accept (Poster) | Reviewers agreed this is an interesting paper and assigned scores ranging from 5 to 8. The rebuttal successfully clarified some of the key reviewers’ initial concerns and the ACs reached consensus that this paper can be accepted for publication. Authors are highly encouraged to address the key comments reported by revi... | 7 | [{"review_id": "twEtZVviH81", "reviewer": "Reviewer_NJn8", "summary": "The paper presents a method for multi-view multi-human 3D pose estimation. The method extracts multi-view feature information using a Transformer architecture and directly regresses the 3D joint locations using the aggregated features. The two propo... | Direct Multi-view Multi-person 3D Pose Estimation
Tao Wang's .2 Jianfeng Zhang²* Yujun Shuicheng Yan Jiashi Jiashi Feng',
Sea Al Lab "National University of Singapore,
twangnh@gmail. com, zhangj ianfjjinnnngeee nus. edu, {caiyj yansc, fngg @eeee
Abstract
We present Multi-view Pose transformer (MvP) for estimating multi... | 43,086 | |||
pZCYG7gjkKz | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 512 | Skyformer: Remodel Self-Attention with Gaussian Kernel and Nystr\"om Method | Transformers are expensive to train due to the quadratic time and space complexity in the self-attention mechanism. On the other hand, although kernel machines suffer from the same computation bottleneck in pairwise dot products, several approximation schemes have been successfully incorporated to considerably reduce t... | [
"Yifan Chen",
"Qi Zeng",
"Heng Ji",
"Yun Yang"
] | [
"efficient transformers",
"attention",
"NLP applications"
] | NeurIPS 2021 Poster | Accept (Poster) | The reviewers agree that this is a solid paper re-introducing the Nystrom method for Transformers to address the quadratic space and time complexity of the regular attention module. Thus, it leverages the fruitful area of research on using kernel approaches to improve regular Transformers. The novelty given that Nystro... | 6 | [{"review_id": "kYTwuHkY4un", "reviewer": "Reviewer_5adD", "summary": "The paper proposes a method (Skyformer) to apply Nystrom approximation to the attention matrix. It embeds the attention matrix inside a larger PSD matrix (unlike Nystromformer), allowing Nystrom method to work well. Theoretical and empirical validat... | Skyformer: Remodel Self-Attention with Gaussian Kernel and Nyström Method
Yifan Chen; Qi Zeng, Heng Ji, Yun Yang
University of Illinois trbana-Champaign (yifanc10, qizeng2, hengji, yy84)eillinois.co edu
Abstract
Transformers are expensive to train due to the quadratic time and space complexity in the self-attention mec... | 50,647 | |||
pX7gwTNljqa | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 8,826 | Non-Gaussian Gaussian Processes for Few-Shot Regression | Gaussian Processes (GPs) have been widely used in machine learning to model distributions over functions, with applications including multi-modal regression, time-series prediction, and few-shot learning. GPs are particularly useful in the last application since they rely on Normal distributions and enable closed-form ... | [
"Marcin Sendera",
"Jacek Tabor",
"Aleksandra Nowak",
"Andrzej Bedychaj",
"Massimiliano Patacchiola",
"Tomasz Trzcinski",
"Przemysław Spurek",
"Maciej Zieba"
] | [
"Meta-Learning",
"Few-Shot Learning",
"Few-Shot Regression",
"Normalizing Flows",
"Gaussian Processes"
] | NeurIPS 2021 Poster | Accept (Poster) | The reviewers ultimately agree that the key idea proposed in this paper, namely the use of continuous normalizing flows (CNFs) to achieve highly flexible non-Gaussian predictive posteriors, applicable to few-shot regression and achieving strong results in that domain. This core idea is both natural and novel, and the e... | 4 | [{"review_id": "o8Jhq5L9ukZ", "reviewer": "Reviewer_3o7L", "summary": "Authors propose a method for incorporating continuous normalising flows (CNFs) into the learning mechanism of Gaussian processes (GP). The work is motivated by the scenario of few-shot learning, where they need more flexibility to model subsequent t... | Non-Gaussian Gaussian Processes for Few-Shot Regression
Marcin Sendera Jagiellonian University
Jacek Tabor Jagiellonian University
Aleksandra Nowak Jagiellonian University
Andrzej Bedychaj Jagiellonian University
Massimiliano Patacchiola University of Cambridge
Tomasz Trzcinski
Przemyslaw Spurek Jagiellonian University... | 48,277 | |||
pTmYjQadg9 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 5,370 | Permutation-Invariant Variational Autoencoder for Graph-Level Representation Learning | Recently, there has been great success in applying deep neural networks on graph structured data. Most work, however, focuses on either node- or graph-level supervised learning, such as node, link or graph classification or node-level unsupervised learning (e.g. node clustering). Despite its wide range of possible appl... | [
"Robin Winter",
"Frank Noe",
"Djork-Arne Clevert"
] | [
"variational autoencoder",
"permutation invariance",
"graph",
"self-attention",
"representation learning",
"graph autoencoder"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper proposed a new variational encoder for unsupervised graph representation learning. The main contribution is the permutation invariance, which is missing in most existing works on graph generative modeling. After the rebuttal, the reviewers found that most of the concerns have been properly addressed, and dur... | 4 | [{"review_id": "vB8k7By211W", "reviewer": "Reviewer_bsmb", "summary": "This work proposes a new graph autoencoder approach, which is claimed to have invariant graph embedding. The work claims to solve the permutation matching challenge during the decoding stage.", "questions": "", "limitations": "", "rating": 6, "confi... | Permutation-Invarian Variational Autoencoder for Graph-Level Representation Learning
Robin Winter
Frank Noé Freie Universität Berlin frank. www.oooeereeennn....mm
Bayer AG
Freie Universität Berlin
robin. winter@bayer.co com
Djork-Arné Clevert
Bayer AG
djork- www.emm.cccleeettbaerr.ooo
Abstract
Recently, there has been ... | 53,114 | |||
pMvBiSLGTeU | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 8,210 | A Topological Perspective on Causal Inference | This paper presents a topological learning-theoretic perspective on causal inference by introducing a series of topologies defined on general spaces of structural causal models (SCMs). As an illustration of the framework we prove a topological causal hierarchy theorem, showing that substantive assumption-free causal in... | [
"Duligur Ibeling",
"Thomas F Icard"
] | [
"causal inference",
"statistical learning theory",
"topology"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper proposes a novel topological approach to the causal hierarchy. All of the reviewers found the approach interesting, although some valid concerns were raised regarding clarity and presentation. In particular, a serious concern regarding comparisons with related work was raised and sorted out during the discus... | 5 | [{"review_id": "nHMUYwpqF1B", "reviewer": "Reviewer_V1rj", "summary": "This paper studies the weak topology on three spaces of probability measures, each corresponding to a level of a causal hierarchy introduced by Pearl (Pearl and Mackenzie, 2018). The three levels, in order of increasing expressivity, are described ... | A Topological Perspective on Causal Inference
Duligur Ibeling Department of Computer Science Stanford University dul dul.igursttanoordd edu
Thomas Icard Department Philosophy Stanford University icard@stanf ord. edu
Abstract
This paper presents a topological serming-theoretic perspective on causal inference by introduc... | 43,166 | |||
pR3dPOHrbfy | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 5,714 | Do Input Gradients Highlight Discriminative Features? | Post-hoc gradient-based interpretability methods [Simonyan et al., 2013, Smilkov et al., 2017] that provide instance-specific explanations of model predictions are often based on assumption (A): magnitude of input gradients—gradients of logits with respect to input—noisily highlight discriminative task-relevant feature... | [
"Harshay Shah",
"Prateek Jain",
"Praneeth Netrapalli"
] | [
"instance-specific explanations",
"post-hoc interpretability",
"feature attributions",
"input gradients",
"adversarial robustness"
] | NeurIPS 2021 Poster | Accept (Poster) | Thank you for your submission to NeurIPS. The reviewers and I are in agreement that the paper presents interesting new insights into the problem of understanding the significance of input gradients as they concern robust and nonrobust models. The addition of the BlockMNIST data set seems particularly helpful to bette... | 4 | [{"review_id": "yiRo87Jiosq", "reviewer": "Reviewer_L2GP", "summary": "The authors test the current assumption for gradient-based explanation methods, i.e., a high magnitude of vanilla gradients highlight more discriminative task-relevant features and vice-versa. The authors theoretically and empirically test the above... | Do Input Gradients Highlight
Discriminative Features?
Harshay Shah' Microsoft Research India harshayggoogle. com
Prateek Jain" Microsoft Research India prajain aiggiingggogee..oo com
Praneeth Netrapalli Microsoft Research India
pnetrapalli@goglee. com
Abstract
Post-hoc gradient-based interpretability methods 1 that pro... | 54,479 | |||
pDgN3l3EAF | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 8,078 | Learning Collaborative Policies to Solve NP-hard Routing Problems | Recently, deep reinforcement learning (DRL) frameworks have shown potential for solving NP-hard routing problems such as the traveling salesman problem (TSP) without problem-specific expert knowledge. Although DRL can be used to solve complex problems, DRL frameworks still struggle to compete with state-of-the-art heur... | [
"Minsu Kim",
"Jinkyoo Park",
"Joungho Kim"
] | [
"Deep reinforcement learning",
"NP-hard",
"Policy collaboration",
"Hierarchical solving strategy"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper proposes Learning Collaborative Policies (LCP) for learning to optimize routing problems. LCP is a hierarchical strategy that consists of two RL policies called the seeder and reviser. The seeder generates many valid but diverse solutions (using an entropy maximization term in the reward). The reviser lear... | 4 | [{"review_id": "m1D4UGz6rT", "reviewer": "Reviewer_Atn1", "summary": "This paper introduces Learning Collaborative Policies (LCP) for learning to optimize TSP-style routing problems. The goal of this was not to outperform all other optimizers, but to outperform RL optimizers, which are usually less effective than tradi... | Learning Collaborative Policies to Solve NP-hard Routing Problems
Minsu Kim
Jinkyoo Park Joungho Kim
Korea Advanced Institute of Science and Technology (KAIST)
i School of Electrical Engineering. f Dept. Industrial & Systems Engineering {min-su, jinkyoo parky, joungho)&kaise ac.kr
Abstract
Recently, deep reinforcement ... | 43,211 | |||
pPbrtkTHe9 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 2,928 | Privately Publishable Per-instance Privacy | We consider how to privately share the personalized privacy losses incurred by objective perturbation, using per-instance differential privacy (pDP). Standard differential privacy (DP) gives us a worst-case bound that might be orders of magnitude larger than the privacy loss to a particular individual relative to a fix... | [
"Rachel Emily Redberg",
"Yu-Xiang Wang"
] | [
"differential privacy",
"private ERM",
"per-instance privacy",
"objective perturbation"
] | NeurIPS 2021 Poster | Accept (Poster) | The authors thought that data dependent privacy properties are an interesting direction of study. The main concern in discussion was the motivation for the results. The reviewers questioned when this notion of privacy would be useful, commenting (among other things) that it could not be directly used by the algorithm d... | 4 | [{"review_id": "msy5UbH69sq", "reviewer": "Reviewer_fBsx", "summary": "The paper looks at how to release per-instance differentially private (pDP) privacy bounds, focusing on empirical risk minimization (ERM) with a convex loss function. The paper proposes methods for releasing either a looser data independent estimate... | Privately Publishable Per-instance Privacy
Rachel Redberg
Yu-Xiang Wang Department Computer Science
Department of Computer Science
UC Santa Barbara Santa Barbara, CA 93106 rredbergduss. edu
UC Santa Barbara Santa Barbara, CA 93106 yuxiangwecs ucsb. edu
Abstract
We consider how to privately share the personalized privac... | 37,869 | |||
ppv5yqhpNyE | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 4,549 | EditGAN: High-Precision Semantic Image Editing | Generative adversarial networks (GANs) have recently found applications in image editing. However, most GAN-based image editing methods often require large-scale datasets with semantic segmentation annotations for training, only provide high-level control, or merely interpolate between different images. Here, we propos... | [
"Huan Ling",
"Karsten Kreis",
"Daiqing Li",
"Seung Wook Kim",
"Antonio Torralba",
"Sanja Fidler"
] | [
"generative adversarial networks",
"image editing"
] | NeurIPS 2021 Poster | Accept (Poster) | This submission tackles the problem of semantic editing of images leveraging a generative model that modulates a joint distribution of images with segmentation masks. Thanks to this joint generation, the editing approach is based on the segmentation masks manipulation.
This allows highly localized editing that were d... | 3 | [{"review_id": "uenONwHVR9G", "reviewer": "Reviewer_6weF", "summary": "The paper proposes to perform image editing via segmentation maps. First, a joint representation of image and segmentation map is learned with a GAN such that the GAN's latent representation w represents both the image and the segmentation map. This... | EditGAN: High-Precision Semantic Image Editing
Huan Lingl
122,,
Karsten Kreis!
Daiqing Li
Seung Wook Kim 1,2.3
Antonio Torralba
Sanja Fidler
'VVIDIA University of Toronto
Vecttor Institute
MIT
(hul ing, kiree a, dal ging), seungwookk, sfldlrrt w cccom torra laagmit..eds
Abstract
Generative adversarial networks (GANs) h... | 55,398 | |||
p9dySshcS0q | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 3,265 | STEP: Out-of-Distribution Detection in the Presence of Limited In-Distribution Labeled Data | Existing semi-supervised learning (SSL) studies typically assume that unlabeled and test data are drawn from the same distribution as labeled data. However, in many real-world applications, it is desirable to have SSL algorithms that not only classify the samples drawn from the same distribution of labeled data but als... | [
"Zhi Zhou",
"Lan-Zhe Guo",
"Zhanzhan Cheng",
"Yu-Feng Li",
"Shiliang Pu"
] | [
"weakly supervised learning",
"semi-supervised learning",
"out-of-distribution detection"
] | NeurIPS 2021 Poster | Accept (Poster) | The paper studied out-of-distribution detection problem and proposed semi-supervised OOD detection setting with a new technique called structure-keep unzipping to solve the new setting. The writing is clear, the motivation is strong, the idea is novel, and the results are significant. Thus, I think it should be accepte... | 4 | [{"review_id": "ljx7R9M95Sd", "reviewer": "Reviewer_oqBv", "summary": "**Summary**\nThis paper proposes a new OOD detection method in a semi-supervised OOD detection setting. The problem formulation is based on two characteristics: (1) labeled ID data is limited and (2) unlabeled data can contain a mixture of ID and OO... | STEP Out-of-Distribution Detection in the Presence of Limited In-distribution Labeled Data
Zhi Zhou', Lan-Zhe Guol,, Zhanzhan Cheng", Yu-Feng Li Shiliang Pu²
National Key Laboratory for Novel Software Technology, Nanjing University, Nanjing, China {zhouz, guolz, 1iyf)=lamda.nni .jj. www..du..uu.. edu. cn Hikvision Rese... | 45,316 | |||
ptw0Soe8W8B | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 10,663 | A Gaussian Process-Bayesian Bernoulli Mixture Model for Multi-Label Active Learning | Multi-label classification (MLC) allows complex dependencies among labels, making it more suitable to model many real-world problems. However, data annotation for training MLC models becomes much more labor-intensive due to the correlated (hence non-exclusive) labels and a potential large and sparse label space. We pr... | [
"Weishi Shi",
"Dayou Yu",
"Qi Yu"
] | [
"multi-label active learning",
"Bayesian Bernoulli mixture",
"label correlation"
] | NeurIPS 2021 Poster | Accept (Poster) | The submission considers the problem of active learning for multi-label classification, and propose a two phase Bayesian approach that takes label correlation into account.
Three reviewers carefully considered the submission, and found several weaknesses in the presentation and proposed idea. In particular, reviewer i... | 3 | [{"review_id": "yMvxEhN_h1-", "reviewer": "Reviewer_dgB5", "summary": "The paper proposes a novel model for multi-label active learning that could learn label correlations from both limited labels and data features. This is achieved through a two-phase learning process that uses a Bayesian Bernoulli mixture of label cl... | A Gaussian Process-Bayesian Bernoulli Mixture Model for Multi-Label Active Learning
Weishi Shi'
Dayou Yu * Qi Yu
Golisano College of Computing and Information Sciences Rochester Institute of Technology {ws7586, dy2507, qi qii.yu)@rit.edi
Abstract
Multi-lahel classification (MLC) allows complex dependencies among labels... | 46,090 | |||
paxcakYWwIu | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 3,259 | Simple Stochastic and Online Gradient Descent Algorithms for Pairwise Learning | Pairwise learning refers to learning tasks where the loss function depends on a pair of instances. It instantiates many important machine learning tasks such as bipartite ranking and metric learning. A popular approach to handle streaming data in pairwise learning is an online gradient descent (OGD) algorithm, where ... | [
"Zhenhuan Yang",
"Yunwen Lei",
"Puyu Wang",
"Tianbao Yang",
"Yiming Ying"
] | [
"Pairwise Learning",
"Online (Stochastic) Gradient Descent",
"Stability and Generalization",
"Differential Privacy"
] | NeurIPS 2021 Poster | Accept (Poster) | All the reviewers agree upon the quality of the paper and support its acceptance. There is no argument to go against this consensus and I recommend the acceptance of this paper.
We count on the authors to take into account the small fixes recommended by the reviewers.
| 4 | [{"review_id": "vYmw8hH3r6T", "reviewer": "Reviewer_7nrA", "summary": "The paper proposed \"stochastic gradient descent\" (SGD) and \"online gradient descent\" (OGD) algorithms for pairwise learning problems. Stability results, optimization, and generalization error bounds for both convex and nonconvex as well as both ... | Simple Stochastic and Online Gradient Descent Algorithms for Pairwise Learning
Zhenhuan Yang!*
Yunwen Lei Puyu Wang? Tianbao Yang* Yiming Ying' University at Albany, SUNY, Albany, NY University of Birmingham, Birmingham CCity University of Hong Kong, Hong Kong *University of lowa City, IA zyang6@ albany.edu, y. @ hhbam... | 45,885 | |||
pHCuidXEinv | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 5,725 | Rebounding Bandits for Modeling Satiation Effects | Psychological research shows that enjoyment of many goods is subject to satiation, with short-term satisfaction declining after repeated exposures to the same item. Nevertheless, proposed algorithms for powering recommender systems seldom model these dynamics, instead proceeding as though user preferences were fixed in... | [
"Liu Leqi",
"Fatma Kilinc-Karzan",
"Zachary Chase Lipton",
"Alan L Montgomery"
] | [
"Bandits",
"Boredom",
"Satiation",
"Recommender Systems",
"Dynamical Systems"
] | NeurIPS 2021 Poster | Accept (Poster) | The rebounding bandits model proposed and analyzed in the paper was appreciated by all the reviewers as an interesting and relevant problem, and by and large the algorithm and the analysis and the presentation was considered to be a decent contribution. The reviewers raised a number of questions in their reviews, and t... | 4 | [{"review_id": "hJ6khh-ByuS", "reviewer": "Reviewer_rYye", "summary": "This paper introduces a problem of modeling satiation effects in multi-armed bandits. For example, if we assume the arms are restaurants that an agent is going to over time. Said agent may have a preference for pizza hut, but eating there every nigh... | Rebounding Bandits for Modeling Satiation Effects
Liu Leqi
Fatma Kılmg-Karzan
Machine Learning Department Carnegie Mellon University Pittsburgh, PA 15213 leqiêcs cmu. edu
Tepper School of Business Carnegie Mellon University
Pittsburgh, PA 15213 fkilinc@andrew cmu edu
Zachary C. Lipton
Alan L. Montgomery Tepper School o... | 43,550 | |||
p2XgjS3Qp4X | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 2,145 | baller2vec++: A Look-Ahead Multi-Entity Transformer For Modeling Coordinated Agents | In many multi-agent spatiotemporal systems, the agents are under the influence of shared, unobserved variables (e.g., the play a team is executing in a game of basketball). As a result, the trajectories of the agents are often statistically dependent at any given time step; however, almost universally, multi-agent mode... | [
"Michael A. Alcorn",
"Anh Totti Nguyen"
] | [] | NeurIPS 2021 Submitted | Reject | All ratings were "reject". The key problems the reviewers found with this paper is that it appears to be very close to baller2vec, the experimental comparisons to other models are not broad enough, there are insufficient ablation experiments, and generally there's a sense that the contribution is not substantial enough... | 4 | [{"review_id": "b0vCBC7ptEq", "reviewer": "Reviewer_K2zL", "summary": "This work presents an extension to the existing baller2vec transformer architecture. In particular, a new attention mask is used internally that is lower-triangular (offset by one block diagonal) rather than block-lower-triangular. This enables ball... | baller2vectt; A Look-Ahead Multi-Entity Transformer For Modeling Coordinated Agents
Anonymous Author(s)
Affiliation Address
email
Abstract
In many multi-agent spatiotemporal systems, agents operate under the influence of shared, unobserved variables (e.g., the play a team is executing in a game of bas- ketball). result... | 39,466 | |||
oz3t1BrfNO | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 4,408 | Estimating Multi-cause Treatment Effects via Single-cause Perturbation | Most existing methods for conditional average treatment effect estimation are designed to estimate the effect of a single cause - only one variable can be intervened on at one time. However, many applications involve simultaneous intervention on multiple variables, which leads to multi-cause treatment effect problems. ... | [
"Zhaozhi Qian",
"Alicia Curth",
"Mihaela van der Schaar"
] | [
"Causal Inference",
"Treatment Effect"
] | NeurIPS 2021 Poster | Accept (Poster) | The authors propose a data augmentation method for learning conditional average treatment effects with multiple treatments. Assuming sequential ignorability and sequential overlap, the authors propose to first fit models that describe how changing a single cause affects causal descendants; then to augment the data set ... | 5 | [{"review_id": "elVl-vgbwD", "reviewer": "Reviewer_yyUp", "summary": "The authors propose a data augmentation method for CATE estimation in the setting of multiple causes. In this setting, there is sparsity in the observed treatment assignments, which makes estimation harder. The authors address this sparsity by aggreg... | Estimating Multi-cause Treatment Effects via Single-cause Perturbation
Zhaozhi Qian University of Cambridge zhaozhi qi anomaths cam ac uk
Alicia Curth University of Cambridge amc2530cam ac.
Mihaela van der Schaar University of Cambridge
UCLA The Alan Turing Institute mv4720cam. ac uk
Abstract
Most existing methods for ... | 48,726 | |||
pSNs0PKx0Mw | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 7,061 | Representation Costs of Linear Neural Networks: Analysis and Design | For different parameterizations (mappings from parameters to predictors), we study the regularization cost in predictor space induced by $l_2$ regularization on the parameters (weights). We focus on linear neural networks as parameterizations of linear predictors. We identify the representation cost of certain sparse... | [
"Zhen Dai",
"Mina Karzand",
"Nathan Srebro"
] | [
"Overparameterized Learning",
"Regularization",
"Representation Cost",
"Linear Neural Networks"
] | NeurIPS 2021 Poster | Accept (Poster) | The paper gives a fairly complete list of results covering the representation cost for linear networks of many different architectures. The paper then studies how to design parametrization to encourage certain regularization effects. The reviewers find the results to be a solid theoretical contribution to the understan... | 4 | [{"review_id": "vJwzfzrlwg7", "reviewer": "Reviewer_n4py", "summary": "This paper investigates the correspondence between the architecture of a linear neural network and the complexity measure induced by putting $l_2$ regularization on the weights. On one hand, the authors study the complexity measure induced by fully-... | Representation Costs of Linear Neural Networks: Analysis and Design
Zhen Dai
Mina Karzand
Committee on Computational and Applied Mathematics University of Chicago
Department of Statistics University California, Davis
Chicago, IL 60637 zhen9@uchicago.com
Davis, CA 95616
mkarzand@ucdavias edu
Nathan Srebro
Toyota Technol... | 40,642 | |||
q88AMOYEKLa | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 6,998 | Distribution-free inference for regression: discrete, continuous, and in between | In data analysis problems where we are not able to rely on distributional assumptions, what types of inference guarantees can still be obtained? Many popular methods, such as holdout methods, cross-validation methods, and conformal prediction, are able to provide distribution-free guarantees for predictive inference, b... | [
"Yonghoon Lee",
"Rina Barber"
] | [
"distribution-free inference",
"nonparametric inference"
] | NeurIPS 2021 Poster | Accept (Poster) | The consensus of the reviewing committee is that the nice theoretical contributions of this paper are sufficient for a Neurips publication. There are some concerns about the importance of these results in applications and the authors are encouraged to address this concern in their revision. | 5 | [{"review_id": "hX12bSN44Id", "reviewer": "Reviewer_uYq5", "summary": "The paper studies distribution-free prediction regions in regression problems, more precisely, it analyses confidence interval constructions, based on an i.i.d. data sample, for the true conditional expectation function (of the output given an input... | Distribution-free inference for regression: discrete, continuous, and in between
Yonghoon Lee Department of Statistics University of Chicago Chicago, IL 60637 yhoony31@uchi hatpsteechhcaaae cago. edu
Rina Foygel Barber Department of Statistics University of Chicago Chicago, IL 60637 rina@uchicago edu
Abstract
In data a... | 37,388 | |||
pLk9yRbRRtF | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 8,509 | Versatile Learned Video Compression | Learned video compression methods have demonstrated great promise in catching up with traditional video codecs in their rate-distortion (R-D) performance. However, existing learned video compression schemes are limited by the binding of the prediction mode and the fixed network framework. They are unable to support var... | [
"Runsen Feng",
"Zongyu Guo",
"Zhizheng Zhang",
"Zhibo Chen"
] | [
"Video Compression",
"Deep Learning"
] | NeurIPS 2021 Submitted | Reject | The work proposes and implements a learning-based video compression solution, which is able to handle multiple prediction modes and achieves good rate-distortion performance. The key building blocks are not new, but the actual model and its successful implementation bring an interesting contribution to the video compre... | 5 | [{"review_id": "ynWDRgtLYap", "reviewer": "Reviewer_hMDd", "summary": "The authors introduce VLVC, a new neural video compression codec. Thanks to a new 3D motion compensation structure based on spatio-temporal interpolation, a single trained model can be run in arbitrary prediction modes (such as low-delay or random-a... | Versatile Learned Video Compression
Anonymous Author(s)
Affiliation Address
email
Abstract
Learned video compression methods have demonstrated great promise in catching up with traditional video codecs in their rate-distortion (R-D) performance. How- ever, existing learned video compression schemes are limited by the b... | 42,938 | |||
oumDUrf2dAB | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 1,121 | On the Value of Infinite Gradients in Variational Autoencoder Models | A number of recent studies of continuous variational autoencoder (VAE) models have noted, either directly or indirectly, the tendency of various parameter gradients to drift towards infinity during training. Because such gradients could potentially contribute to numerical instabilities, and are often framed as a probl... | [
"Bin Dai",
"Li Kevin Wenliang",
"David Wipf"
] | [
"variational autoencoders",
"sparse representations",
"latent variable models"
] | NeurIPS 2021 Spotlight | Accept (Spotlight) | This is a strong paper that all reviewers felt should be accepted to the conference; a viewpoint I concur with myself. Some constructive criticisms have though been raised and I encourage the authors to incorporate these suggestions for the final version of the work.
A few extra comments from my end to help update th... | 3 | [{"review_id": "_hupuBO3W9t", "reviewer": "Reviewer_9vZA", "summary": "The author(s) challenge the conventional notion that infinite gradients are bad. Specifically, they examine infinite gradients in the context of auto-encoders and VAEs. Their theoretical findings suggest infinite gradients are necessary at global op... | On the Value of Infinite Gradients in Variational Autoencoder Models
Bin Dai Institue for Advanced Study
K. Wenliang
Gatsby Computational Neuroscience Unit University College London
Tsinghua University
caibo@plyysicc@@hoomall...coo
kevinli@gatsly ucl. ac uk
David Wipf
Shanghai Research Lab
Amazon Web Services
davidwipf... | 47,173 | |||
p99rWde9fVJ | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 4,742 | Federated Hyperparameter Tuning: Challenges, Baselines, and Connections to Weight-Sharing | Tuning hyperparameters is a crucial but arduous part of the machine learning pipeline. Hyperparameter optimization is even more challenging in federated learning, where models are learned over a distributed network of heterogeneous devices; here, the need to keep data on device and perform local training makes it diffi... | [
"Mikhail Khodak",
"Renbo Tu",
"Tian Li",
"Liam Li",
"Nina Balcan",
"Virginia Smith",
"Ameet Talwalkar"
] | [
"federated learning",
"hyperparameter optimization",
"meta-learning",
"personalization"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper studies hyperparameter optimization in FL, which is often neglected but a crucial part of FL. The paper provides interesting ideas to tackle this challenging problem. After the discussion phase, the reviewers are all in favor of accepting the paper. I recommend acceptance. I suggest the authors revise the pa... | 3 | [{"review_id": "fef8H8_IJo2", "reviewer": "Reviewer_onB2", "summary": "This paper proposes a method for learning client-side hyperparameters in the federated setting by parameterizing the distribution of these hyperparameters and estimating gradients with respect to this parametererization. This approach is connected t... | Federated Hyperparameter Tuning: Challenges, Baselines, and Connections to Weight-Sharing
Mikhail Khodak, Renbo Tu, Tian Li Carnegie Mellon University [ichodakrrenbo tianli)@cmu.ce
Liam Li
Hewlett Packard Enterprise me01 m00liamcli..m com
Maria-Florina Balcan, Virginia Smith Carnegie Mellon University ninamf$cs. cmu ed... | 54,050 | |||
owQmPJ9q9u | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 7,829 | Controlling Neural Networks with Rule Representations | We propose a novel training method that integrates rules into deep learning, in a way the strengths of the rules are controllable at inference. Deep Neural Networks with Controllable Rule Representations (DeepCTRL) incorporates a rule encoder into the model coupled with a rule-based objective, enabling a shared represe... | [
"Sungyong Seo",
"Sercan O Arik",
"Jinsung Yoon",
"Xiang Zhang",
"Kihyuk Sohn",
"Tomas Pfister"
] | [
"Controlling neural networks",
"Incorporating rules",
"Unsupervised adaptation"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper considers deep learning models for applications where rules are important (e.g. physics). The idea is to construct a "rule encoder" to incorporate user-specified rules. Specifically, this integration is done by perturbing the observations and enforce some user-specified rules to "regulate" between the model'... | 4 | [{"review_id": "xmtnhShYkl", "reviewer": "Reviewer_9N6z", "summary": "This paper proposes DeepCTRL, a method for infusing domain knowledge in the form of rules into the representation learning models. The method independently encodes the input instance and the supplied rule and couples the input and rule encodings wher... | Controlling Neural Networks with Rule Representations
Sungyong Seo, Sercan Ô. Arik, Jinsung Yoon, Xiang Zhang, Kihyuk Sohn, Tomas Pfister
Google Cloud Al Sunnyvale, CA, USA
Ssungyongs,ssarte ftn.aiiii.iiuugggooooooooor
kihyuks,, htg.a/aawaaaa.........o.ooooooo.
Abstract
We propose novel training method that integrates ... | 44,821 | |||
p5rMPjrcCZq | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 2,271 | Only Train Once: A One-Shot Neural Network Training And Pruning Framework | Structured pruning is a commonly used technique in deploying deep neural networks (DNNs) onto resource-constrained devices. However, the existing pruning methods are usually heuristic, task-specified, and require an extra fine-tuning procedure. To overcome these limitations, we propose a framework that compresses DNNs ... | [
"Tianyi Chen",
"Bo Ji",
"Tianyu DING",
"Biyi Fang",
"Guanyi Wang",
"Zhihui Zhu",
"Luming Liang",
"Yixin Shi",
"Sheng Yi",
"Xiao Tu"
] | [
"model compression",
"structured pruning",
"group sparsity"
] | NeurIPS 2021 Poster | Accept (Poster) | The authors propose a fine-tuning free structured pruning method (OTO). The idea is to first partition the parameters into zero-invariant groups, pruning the zero groups, and solving for a structured sparsity optimization problem with projections. The experiment results on CIFAR10, ImageNet, and SQuAD show the competit... | 4 | [{"review_id": "zNxJOjLP3Pt", "reviewer": "Reviewer_N3gq", "summary": "This paper first proposes the zero-invariant group for neural network. Zero-invariant group is the disjoint group of trainable parameters which results in corresponding output to be zero. Then, this paper formulates a optimization problem with group... | Only Train Once: A One-Shot Neural Network Training And Pruning Framework
Tianyi Chen'
Bo Ji
Microsoft
National University of Singapore jibolcomp. nus edu. sg
Blaccen@@iicrooof..com
Tianyu Ding
Biyi Fang
Guanyi Wang Georgia Institute of Technology
Johns Hopkins University tdingle @ jhu.ce. edu
Microsoft
biffmmicrosoft.... | 57,736 | |||
p7GujbewmRY | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 9,797 | Grounding inductive biases in natural images: invariance stems from variations in data | To perform well on unseen and potentially out-of-distribution samples, it is desirable for machine learning models to have a predictable response with respect to transformations affecting the factors of variation of the input. Here, we study the relative importance of several types of inductive biases towards such pred... | [
"Diane Bouchacourt",
"Mark Ibrahim",
"Ari S. Morcos"
] | [
"data augmentation",
"invariance",
"transformations",
"factors of variation"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper brought about very divergent opinions from the reviewers — they all agreed that the analyses across the three sections are technically solid, and the topic is of broad interest to the community, but the main points of disagreement were over the novelty of the results, the paper being hard to read since it is... | 4 | [{"review_id": "qk-pVqUoUqv", "reviewer": "Reviewer_WyYA", "summary": "This paper provides an analysis of the invariance of neural networks (ResNet) trained on image classification (ImageNet) to image transformations, and of the role of data augmentation in encouraging invariance. The paper provides different types of ... | Grounding inductive biases in natural images: invariance stems from variations in data
Diane Bouchacourt; Mark Ibrahim; Ari S. Morcos Facebook AI Research
{dianeb, marksibrahinn, effo com com
Abstract
To perform well on unseen and potentially ant-of-distributioo samples, is desir- able for machine learning models to ha... | 53,542 | |||
omDF-uQ_OZ | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 11,132 | Identification of the Generalized Condorcet Winner in Multi-dueling Bandits | The reliable identification of the “best” arm while keeping the sample complexity as low as possible is a common task in the field of multi-armed bandits. In the multi-dueling variant of multi-armed bandits, where feedback is provided in the form of a winning arm among as set of k chosen ones, a reasonable notion of be... | [
"Björn Haddenhorst",
"Viktor Bengs",
"Eyke Hüllermeier"
] | [
"Best arm identification",
"Condorcet Winner",
"Dvoretzky–Kiefer–Wolfowitz inequality",
"Multi-armed Bandits",
"Preference Learning"
] | NeurIPS 2021 Poster | Accept (Poster) | The reviewers came to consensus that the theoretical strength overtakes the concerns such as the assumption on the existence of GCW and the heaviness of the technical materials. I agree with these opinions and please sincerely address the concerns raised by the reviewers in the final version such as the relation with t... | 3 | [{"review_id": "qTkCeYkkFpt", "reviewer": "Reviewer_RVxf", "summary": "This work studies the BAI problem in the multi-dueling bandits. \nIt proposes algorithms with theoretical guarantee and provides related lower bounds.\nThere are detailed comparisons to existing works.\nMoreover, it compare the proposed algorithm wi... | Identification of the Generalized Condorcet Winner in Multi-dueling Bandits
Björn Haddenhorst
Viktor Bengs Institute of Informatics
Department of Computer Science
Paderborn University Paderborn, Germany
University of Munich (LMU)
Munich, Germany viktor. www.seeng.mmmmmm.m
bjo bjoernha@mail. upb.de
Eyke Hüllermeier
Inst... | 41,220 | |||
oqKC5A7iq_k | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 1,078 | Low-Fidelity Video Encoder Optimization for Temporal Action Localization | Most existing temporal action localization (TAL) methods rely on a transfer learning pipeline: by first optimizing a video encoder on a large action classification dataset (i.e., source domain), followed by freezing the encoder and training a TAL head on the action localization dataset (i.e., target domain). This resul... | [
"Mengmeng Xu",
"Juan-Manuel Perez-Rua",
"Xiatian Zhu",
"Bernard Ghanem",
"Brais Martinez"
] | [
"End-to-End Pre-training",
"Temporal Action Localization"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper presents work on temporal action localization. The main idea is to use low fidelity (e.g. lower temporal resolution) to enable end-to-end training within the constraints imposed by large video batches / models in GPU memory. The reviewers appreciate the simplicity and effectiveness of this idea. While bas... | 4 | [{"review_id": "TD4NdfTzib", "reviewer": "Reviewer_mQTb", "summary": "Updating after the rebuttal:\nThe rebuttal well addressed some of my concerns. I would raise the rating to 6: Marginally above the acceptance threshold\n\n---\n\nThe paper presents a low-fidelity video encoder optimization approach to relieve the lar... | Low-Fidelity Video Encoder Optimization for Temporal Action Localization
Mengmeng Xu
Juan-Manuel Pérez-Rúa'
Xiatian Zhu'
mengmeng, xu@kaust. edu. sa
PerezRua. PerezRua.JNogmail JM@gmail com
xiatian zwu@samsung com
Bernard Ghanem
Brais Martinez! brais. a@samsung............... com
bernard, ghanem@kaust edu sa
Samsung Al... | 49,222 | |||
pBKOx_dxYAN | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 2,217 | SNIPS: Solving Noisy Inverse Problems Stochastically | In this work we introduce a novel stochastic algorithm dubbed SNIPS, which draws samples from the posterior distribution of any linear inverse problem, where the observation is assumed to be contaminated by additive white Gaussian noise. Our solution incorporates ideas from Langevin dynamics and Newton's method, and ex... | [
"Bahjat Kawar",
"Gregory Vaksman",
"Michael Elad"
] | [
"Inverse Problems",
"Image Restoration",
"Denoising",
"Langevin Dynamics",
"Diffusion",
"Super Resolution",
"Deblurring",
"Compressive Sensing"
] | NeurIPS 2021 Poster | Accept (Poster) | The paper considers noisy linear inverse problems y = Hx +noise, and specifically, the problem of generating samples from the posterior p(x|y) using a Langevin dynamics approach. This has been considered before quite recently, but the novelty of the current paper is working with noise as well as extending ref [19] to w... | 4 | [{"review_id": "wEyaUkJ2YqE", "reviewer": "Reviewer_y4bh", "summary": "This paper proposes to combine Langevin dynamics and Newton's method to arrive at a posterior sampling algorithm, where the prior distribution is given by an MMSE Gaussian denoiser. The proposed method is evaluated on multiple imaging inverse proble... | SNIPS: Solving Noisy Inverse Problems Stochastically
Bahjat Kawar, Gregory Vaksman, Michael Elad Computer Science Department, Technion, Haifa, Israel {bahjat. kawar, grishav, elad}-Ccs technion ac.
Abstract
In this work we introduce novel stochastic algorithm dubbed SNIPS, which draws samples from the posterior distrib... | 42,098 | |||
otDgw7LM7Nn | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 5,490 | Analytic Insights into Structure and Rank of Neural Network Hessian Maps | The Hessian of a neural network captures parameter interactions through second-order derivatives of the loss. It is a fundamental object of study, closely tied to various problems in deep learning, including model design, optimization, and generalization. Most prior work has been empirical, typically focusing on low-ra... | [
"Sidak Pal Singh",
"Gregor Bachmann",
"Thomas Hofmann"
] | [
"hessian",
"rank",
"neural networks",
"overparameterization",
"degeneracy",
"singularity",
"loss landscape",
"degrees of freedom"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper gives an exact characterization on the rank of the Hessian for linear networks, and show that in many settings the rank can be much smaller than the number of parameters. The paper then verify these claims empirically in nonlinear networks and show that the Hessian is still approximately low rank. Although t... | 4 | [{"review_id": "ZTuaRX3BsrP", "reviewer": "Reviewer_bE99", "summary": "This paper analyzes the rank of the Hessian matrix of fully connected neural networks. Theoretically, for a linear fully-connected neural network with square loss, the authors provided an upper bound for the rank of Hessian with respect to the popul... | Analytic Insights into Structure and Rank of Neural Network Hessian Maps
Sidak Pal Singh*
Gregor Bachmann"* and Thomas Hofmann
ETH Zürich
Max Planck ETH Center for Learning Systems
Abstract
The Hessian a neural network captures parameter interactions through second- order derivatives of the loss. It is fundamental obje... | 51,624 | |||
oi08QWKs84 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 10,819 | Adversarial Robustness with Non-uniform Perturbations | Robustness of machine learning models is critical for security related applications, where real-world adversaries are uniquely focused on evading neural network based detectors. Prior work mainly focus on crafting adversarial examples (AEs) with small uniform norm-bounded perturbations across features to maintain the r... | [
"Ecenaz Erdemir",
"Jeffrey Bickford",
"Luca Melis",
"Sergul Aydore"
] | [
"Adversarial robustness",
"adversarial attacks",
"projected gradient descent",
"certified robustness",
"randomized smoothing"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper proposes to improve the adversarial robustness by using non-uniform perturbations in adversarial training. After the rebuttal and discussion, the reviewers raised several concerns on the novelty, the marginal improvement over baselines, and some experimental settings. The paper could be strengthened if these... | 5 | [{"review_id": "nUgz1tayZzL", "reviewer": "Reviewer_QUCv", "summary": "In domains other than computer vision, a uniform bound for adversarial perturbations may not make much sense as different input features can have divergent value ranges and meanings. For instance, in malware detection, certain features are not allow... | Adversarial Robustness with Non-uniform Perturbations
Ecenaz Erdemir Imperial College London e, erdemir17@@mppriall ac. uk
Jeffrey Bickford Amazon Web Services
Luca Melis Amazon Web Services
jbickOamazon.com com
micmeli@amazen com
Sergiil Aydöre Amazon Web Services saydore@amazon.co com
Abstract
Robustness of machine l... | 49,195 | |||
oyHWvdvkZDv | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 10,957 | Efficient Truncated Linear Regression with Unknown Noise Variance | Truncated linear regression is a classical challenge in Statistics, wherein a label, $y = w^T x + \varepsilon$, and its corresponding feature vector, $x \in \mathbb{R}^k$, are only observed if the label falls in some subset $S \subseteq \mathbb{R}$; otherwise the existence of the pair $(x, y)$ is hidden from observatio... | [
"Constantinos Costis Daskalakis",
"Patroklos Stefanou",
"Rui Yao",
"Emmanouil Zampetakis"
] | [
"regression",
"theory of computation",
"truncated statistics",
"linear regression",
"truncation bias",
"stochastic gradient descent",
"asymptotic normality"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper studies high-dimensional linear regression in a truncated setting, where only examples with labels in a given set are observed.
Learning from truncated samples is a classical topic in statistics that has received renewed interest in the last few years. Prior work had given an efficient algorithm for this pro... | 3 | [{"review_id": "q9OkK4TElTK", "reviewer": "Reviewer_Hgq6", "summary": "The paper extends the study of Daskalakis et al. (2019) into the setting where the noise variance is unknown, and gives two theoretical contributions. First, the paper proves the convergence of the projected SGD algorithm when applied to minimizing ... | Efficient Truncated Linear Regression with Unknown Noise Variance
Constantinos Daskalakis 'EECS and CSAIL, MIT costis@csail. mit. edu
Patroklos Stefanou EECS and CSAIL, MIT stefanou@mit edu
Rui Yao EECS and CSAIL, MIT rayyao@mit.c edu
Manolis Zampetakis EECS, UC Berkeley mzampet@berkelee edu
Abstract
Truncated linear r... | 38,266 | |||
ohfi44BZPC4 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 9,501 | Three-dimensional spike localization and improved motion correction for Neuropixels recordings | Neuropixels (NP) probes are dense linear multi-electrode arrays that have rapidly become essential tools for studying the electrophysiology of large neural populations. Unfortunately, a number of challenges remain in analyzing the large datasets output by these probes.
Here we introduce several new methods for extra... | [
"Julien Boussard",
"Erdem Varol",
"Hyun Dong Lee",
"Nishchal Dethe",
"Liam Paninski"
] | [
"Spike-sorting",
"Neuropixels",
"Localization",
"Registration"
] | NeurIPS 2021 Poster | Accept (Poster) | This work brings together a number of algorithms to provide a method to detect the 3D location of cells from data collected with new high-density electrode arrays: Neuropixels. All reviewers appreciated the analysis developed by the authors in this work. The main discussion centered around the scope of NeurIPS with res... | 4 | [{"review_id": "rbCJHmJCwch", "reviewer": "Reviewer_PJEK", "summary": "The authors develop a new analysis pipeline for the detection and localization of spikes in Neuropixel recordings. The pipeline is based on a denoising step applied to all channels, detection of spikes via their amplitudes, triangulation across chan... | Three-dimensional spike localization and improved motion correction for Neuropixels recordings
Julien Boussard'
Erdem Varol"
Hyun Dong Lee
Nishchal Dethe
Liam Paninski
Department of Statistics and Neuroscience, Center for Theoretical Neuroscience, Grossman Center for the Statistics of Mind, Zuckerman Institute, Columbi... | 38,317 | |||
on2DNSz2Qg | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 5,253 | Implicit Deep Adaptive Design: Policy-Based Experimental Design without Likelihoods | We introduce implicit Deep Adaptive Design (iDAD), a new method for performing adaptive experiments in real-time with implicit models. iDAD amortizes the cost of Bayesian optimal experimental design (BOED) by learning a design policy network upfront, which can then be deployed quickly at the time of the experiment. The... | [
"Desi R. Ivanova",
"Adam Foster",
"Steven Kleinegesse",
"Michael U. Gutmann",
"Tom Rainforth"
] | [
"Experimental Design",
"Mutual Information",
"Implicit Models",
"Bayesian methods",
"Bayesian Optimal Experimental Design",
"BOED",
"Sequential BOED",
"Adaptive experiments",
"Likelihood-free methods",
"Likelihood-free inference",
"Parameter estimation",
"Variational methods",
"Mutual inform... | NeurIPS 2021 Poster | Accept (Poster) | The paper is on Bayesian experimental design, building on the recently proposed Deep Adaptive Design (DAD). The paper's contribution is an extension iDAD (I for implicit) that does not require explicit likelihoods, does not require that experiments are conditionally independent, and is fast enough for real-time applica... | 4 | [{"review_id": "cidH7IUGuAt", "reviewer": "Reviewer_k7ct", "summary": "This paper proposes iDAD, an amortized learning framework for implicit (likelihood-free) experimental design. The iDAD framework extends the previously proposed deep adaptive design (DAD) method (Foster et al., 2013), which is an amortized learning ... | Implicit Deep Adaptive Design: Policy-Based Experimental Design without Likelihoods
Desi R. Ivanova' Adam Foster' Steven Kleinegesse* Michael U. Gutmann' Tom Rainforth"
'Department of Statistics, University of Oxford School of Informatics, University of Edinburgh
desi www.ova@statteeoooomm ac,
Abstract
We introduce imp... | 54,940 | |||
ohZjthN1ncg | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 3,301 | Overinterpretation reveals image classification model pathologies | Image classifiers are typically scored on their test set accuracy, but high accuracy can mask a subtle type of model failure. We find that high scoring convolutional neural networks (CNNs) on popular benchmarks exhibit troubling pathologies that allow them to display high accuracy even in the absence of semantically sa... | [
"Brandon Carter",
"Siddhartha Jain",
"Jonas Mueller",
"David Gifford"
] | [
"computer vision",
"benchmarks",
"datasets",
"convolutional neural networks",
"interpretability",
"robustness",
"overinterpretation"
] | NeurIPS 2021 Poster | Accept (Poster) | This work argues that deep neural networks overinterpret the input, i.e. they rely on too small portions of the image to make the decision. This is novel insight and I am enthusiastic that it will help the community understand better deep networks. All reviewers agree that the paper should be accepted, and I am glad to... | 4 | [{"review_id": "sdFc6VIV64k", "reviewer": "Reviewer_dNig", "summary": "The paper presents an interesting finding that CNNs overinterpret a few non-meaningful pixels in the image to make accurate predictions. The authors use SIS and extend it to Batched Gradient SIS to discover the limited number of pixels such CNNs rel... | Overinterpretation reveals image classification model pathologies
Brandon Carter MIT CSAIL
Siddhartha Jain MIT CSAIL
Jonas Mueller Amazon Web Services
David Gifford MIT CSAIL gi ifffordmii..cmm
bcarter@csail mit.
Abstract
Image classifiers typically scored their test accuracy, but high accuracy can mask a subtle type o... | 49,927 | |||
oErdeq9ajjX | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 7,067 | Learning Generalized Gumbel-max Causal Mechanisms | To perform counterfactual reasoning in Structural Causal Models (SCMs), one needs to know the causal mechanisms, which provide factorizations of conditional distributions into noise sources and deterministic functions mapping realizations of noise to samples. Unfortunately, the causal mechanism is not uniquely identifi... | [
"Guy Lorberbom",
"Daniel D. Johnson",
"Chris J. Maddison",
"Daniel Tarlow",
"Tamir Hazan"
] | [
"Gumbel max",
"reparameterization trick",
"structural causal model"
] | NeurIPS 2021 Spotlight | Accept (Spotlight) | The authors study couplings between discrete distributions that allow to infer counterfactuals. Overall, this is an important topic and in many applications, counterfactuals are desired. An inherent difficulty of inferring counterfactuals is the need to make unverifiable assumptions. The authors argue to choose a causa... | 3 | [{"review_id": "vujIK7LM43m", "reviewer": "Reviewer_XBKm", "summary": "This paper proposes a parameterized family of causal mechanisms that generalize Gumbel-max, which can be trained to minimize counterfactual-effect variance. It is motivated by the fact that the causal mechanism is not uniquely identifiable with Gumb... | Learning Generalized Gumbel-max Causal Mechanisms
Guy Lorberbom"
Daniel D. JJohnson" Google Research Toronto, ON, Canada dd jaijohonnggogle... com
Technion
Haifa, Israel
guy juy_lorber@camuus hww.obbrrrrammp.tteeeiiiooooo. technion. ac. il
Chris J. Maddison
Daniel Tarlow Google Research Montreal, QC, Canada
Tamir Hazan... | 45,487 | |||
oRMRIR4qPC1 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 10,435 | Faster Perturbed Stochastic Gradient Methods for Finding Local Minima | Escaping from saddle points and finding local minimum is a central problem in nonconvex optimization.
Perturbed gradient methods are perhaps the simplest approach for this problem. However, to find $(\epsilon, \sqrt{\epsilon})$-approximate local minima, the existing best stochastic gradient complexity for this type of ... | [
"Zixiang Chen",
"Dongruo Zhou",
"Quanquan Gu"
] | [] | NeurIPS 2021 Submitted | Reject | The paper proposes and analyzes a stochastic first-order optimization method called “Pullback,” proving that it achieves the optimal rate of convergence to $(\epsilon,\sqrt{\epsilon})$-approximate local minima under appropriate assumptions.
After reading the paper and discussing it with the reviewers, I concur with r... | 4 | [{"review_id": "vnuTLuQGhn", "reviewer": "Reviewer_zSqf", "summary": "The paper proposes a new approach called \"Pullback\" for escaping saddle points and finding local minimum that matches the same complexity as SGD. The approach perturbs a stochastic gradient estimators (SARAH/SPIDER) and STORM by an element uniforml... | For the case where is deterministic function, it has been shown that vanilla gradient descent fails 24 to find local minima efficiently since the iterates will get stuck at saddle points for exponential time Faster Perturbed Stochastic Gradient Methods for Finding Local Minima
Anonymous Author(s)
Affiliation Address em... | 37,044 | |||
oIhzg4GJeOf | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 11,628 | Learning Semantic Representations to Verify Hardware Designs | Verification is a serious bottleneck in the industrial hardware design cycle, routinely requiring person-years of effort. Practical verification relies on a "best effort" process that simulates the design on test inputs. This suggests a new research question: Can this simulation data be exploited to learn a continuous ... | [
"Shobha Vasudevan",
"Wenjie Jiang",
"David Bieber",
"Rishabh Singh",
"HAMID SHOJAEI",
"C. Richard Ho",
"Charles Sutton"
] | [
"Hardware Design",
"Verification",
"Graph Convolutional Networks",
"Test generation"
] | NeurIPS 2021 Poster | Accept (Poster) | The idea of predictive coverage/testing for RTL designs using modern representation learning methods is interesting and potentially useful. Most of the reviewer discussion surrounded the best ways to evaluate such a method; eventually reviewers agreed that the authors did as well as could be expected given the state of... | 3 | [{"review_id": "mPTkr6Z6nm", "reviewer": "Reviewer_z6UE", "summary": "This paper introduces Design2Vec, which models the semantics of RTL programs using a graph neural network. The representation is used for 2 tasks: coverage prediction and test generation. The experiments on different designs show that Design2Vec can ... | Learning Semantic Representations to Verify Hardware Designs
Shobha Vasudevan Google Research, Brain Team shovasu@google. com
Wenjie Jiang
David Bieber
Google Research, Brain Team
Google Research, Brain Team
@google.eogl...ooooo com
dbieber@google.cm com
Rishabh Singh
Hamid Shojaei
Richard Ho Google
Google X
Google ham... | 52,678 | |||
oa1AMhWKrS | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 1,452 | Fast Approximation of the Sliced-Wasserstein Distance Using Concentration of Random Projections | The Sliced-Wasserstein distance (SW) is being increasingly used in machine learning applications as an alternative to the Wasserstein distance and offers significant computational and statistical benefits. Since it is defined as an expectation over random projections, SW is commonly approximated by Monte Carlo. We adop... | [
"Kimia Nadjahi",
"Alain Durmus",
"Pierre Jacob",
"Roland Badeau",
"Umut Simsekli"
] | [
"optimal transport",
"central limit theorem",
"random projections",
"high-dimensional data",
"generative modeling"
] | NeurIPS 2021 Poster | Accept (Poster) | The focus of the submission is the fast approximation of the sliced-Wasserstein distance (SW). Particularly, the authors present a new scheme to tackle this task relying on specifically designed Gaussian projections as an alternative to the widely-used Monte Carlo approximation (where the projection directions are dist... | 4 | [{"review_id": "sa12KXVsK8I", "reviewer": "Reviewer_sFFp", "summary": "The paper presents a method to estimate the Sliced-Wasserstein distance. The method is based on the Gaussian approximation of the projected data, thus the SW distance could be approximated through the SW distance between Gaussians which has a closed... | Fast Approximation of the Sliced- wasse--Wasserstein Distance Using Concentration of Random Projections
Kimia Nadjahi Alain Durmus', Pierre Jacob Roland Badcau', Umut Şimşekli
1: LTCI, Télécom Paris, Institut Polytechnique de Paris, France Université Paris-Saclay, ENS Paris-Saclay, CNRS, Centre Borelli, F-91 190 Gif-su... | 49,482 | |||
pmWeMLm411_ | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 20 | Learning High-Precision Bounding Box for Rotated Object Detection via Kullback-Leibler Divergence | Existing rotated object detectors are mostly inherited from the horizontal detection paradigm, as the latter has evolved into a well-developed area. However, these detectors are difficult to perform prominently in high-precision detection due to the limitation of current regression loss design, especially for objects w... | [
"Xue Yang",
"Xiaojiang Yang",
"Jirui Yang",
"Qi Ming",
"Wentao Wang",
"Qi Tian",
"Junchi Yan"
] | [
"Rotated Object Detection",
"High-Precision",
"Kullback-Leibler Divergence"
] | NeurIPS 2021 Poster | Accept (Poster) | The reviewers have discussed the paper and have not come to an agreement.
Reasons to accept: the proposed approach is noted to be simple and yet effective, and it can improve the performance of various existing object detectors on different datasets significantly. Most of the issues raised in the reviews were properly ... | 3 | [{"review_id": "wmt5dUBtsiq", "reviewer": "Reviewer_7RYn", "summary": "The paper proposes an algorithm that can detect slender objects,\nlike text in images or boats in aerial images. The work mainly focuses on\nthe bounding box regression task and includes a parameter for box rotation.\nBounding boxes are represented ... | Learning High-Precision Bounding Box for Rotated Object Detection via Kullback-Leibler Divergence
Xue Yang', Xiaojiang Yang', Jirui Yang", Qi Ming', Wentao Wang', Qi Tian", Junchi Yan
Department of Computer Science and Engineering,
MoE Key Lab Artificial Intelligence, Al Institute, Shanghai Jiao Tong University
Univers... | 51,659 | |||
oiq92o1EFg1 | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 1,549 | LLC: Accurate, Multi-purpose Learnt Low-dimensional Binary Codes | Learning binary representations of instances and classes is a classical problem with several high potential applications. In modern settings, the compression of high-dimensional neural representations to low-dimensional binary codes is a challenging task and often require large bit-codes to be accurate. In this work, w... | [
"Aditya Kusupati",
"Matthew Wallingford",
"Vivek Ramanujan",
"Raghav Somani",
"Jae Sung Park",
"Krishna Pillutla",
"Prateek Jain",
"Sham M. Kakade",
"Ali Farhadi"
] | [
"Classification",
"Efficient Classification",
"Error Correcting Output Codes",
"Binary Codes",
"Retrieval",
"Large-scale Classification"
] | NeurIPS 2021 Poster | Accept (Poster) | The problem considered in this paper, namely, of learning compression schemes for high dimensional neural networks, is a very natural and important problem. The paper achieves new, state-of-the-art results with new techniques for this natural problem. While the proposed technique lacks theoretical grounding, the empiri... | 5 | [{"review_id": "xZqa7SV1lu", "reviewer": "Reviewer_vExU", "summary": "Low-dimensional binary codes are important for a variety of large-scale ml tasks, especially retrieval. This paper proposed a method called LLC for learning semantially meaningful low-dimensional biary codes. Compare to the existing literature, LLC c... | LLC: Accurate, Multi-purpose Learnt Low-dimensional Binary Codes
Aditya Kusupati",
Matthew Wallingford", Vivek Ramanujan', Raghav Somani .ae Sung Parki, Krishna Pillutlal, Prateek Jain", Sham Kakadet and Ali Farhadi
UUniversity of Washington, 'Google Research India
(kusupati pecpuppti,cc2244,,11 raccx244,aaaa,,rreae ra... | 56,137 | |||
oaIa8kTLjmP | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 11,633 | Asymptotics of the Bootstrap via Stability with Applications to Inference with Model Selection | One of the most commonly used methods for forming confidence intervals is the empirical bootstrap, which is especially expedient when the limiting distribution of the estimator is unknown. However, despite its ubiquitous role in machine learning, its theoretical properties are still not well understood. Recent developm... | [
"Morgane Austern",
"Vasilis Syrgkanis"
] | [
"bootstrap",
"asymptotics",
"confidence intervals",
"model selection"
] | NeurIPS 2021 Poster | Accept (Poster) | There has been a tremendous amount of discussion about this paper. Several points of criticism have been raised by Reviewer XTTq, mostly concerning the used stability condition, consistency in high dimensions and the generality of the specific kernel used and the examples in sections 4/5. On the other hand, all other r... | 4 | [{"review_id": "ibNL-scP5ih", "reviewer": "Reviewer_iG7z", "summary": "The authors derive general stability conditions under which the empirical bootstrap estimator is consistent and quantify the speed of convergence. These results are illustrated for two-sample kernel tests after kernel selection and the empirical ris... | Asymptoties of the Bootstrap via Stability with Applications to Inference with Model Selection
Morgane Austern
Vasilis Syrgkanis Microsoft Research vasy@microsoot..cmm
Microsoft Research morgane, aesteeeeeetttteemmmm.m.... com
Abstract
One of the most commonly used methods for forming confidence intervals is the empiri... | 40,211 | |||
ogjTzvtqbtK | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 8,285 | Deformable Butterfly: A Highly Structured and Sparse Linear Transform | We introduce a new kind of linear transform named Deformable Butterfly (DeBut) that generalizes the conventional butterfly matrices and can be adapted to various input-output dimensions. It inherits the fine-to-coarse-grained learnable hierarchy of traditional butterflies and when deployed to neural networks, the promi... | [
"Rui Lin",
"Jie Ran",
"King Hung Chiu",
"Graziano Chesi",
"Ngai Wong"
] | [
"Deformable Butterfly",
"Linear transform",
"Model compression"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper continues the line of work attempting to reduce the complexity (in terms of number of parameters and train/inference time) of linear transformations in deep networks, using products of linear transformations inspired by the ‘Butterfly’ structure. From what I understand, the main contribution here compared to... | 4 | [{"review_id": "wNoVlj3h_Y2", "reviewer": "Reviewer_X4Ja", "summary": "The paper proposes a learned decomposition of convolutional/linear layers' weight matrices, inspired by the butterfly transformation. Apart from the proposed structured linear transform, the paper proposes an ALS method for initializing decomposed l... | Deformable Butterfly: A Highly Structured and Sparse Linear Transform
Rui Lin li+
Jie Ran
King Hung Chiu² Grazinao Chesi!
Ngai Wong 1M
Department of Electrical and Electronic Engineering.
The University of Hong Kong, Hong Kong
Emial Address: {linrui, jieran, chesi, nwong) hku hk
2 United Microelectronies Centre (Hong K... | 42,978 | |||
oVEGzC7ieOB | neurips | 2,021 | main | NeurIPS.cc/2021/Conference | 7,573 | Constrained Two-step Look-Ahead Bayesian Optimization | Recent advances in computationally efficient non-myopic Bayesian optimization offer improved query efficiency over traditional myopic methods like expected improvement, with only a modest increase in computational cost. These advances have been largely limited to unconstrained BO methods with only a few exceptions whic... | [
"Duke Zhang",
"Xiangyu Zhang",
"Peter I. Frazier"
] | [
"Bayesian optimization",
"Non-myopic",
"Gaussian processes",
"Derivative-free optimization"
] | NeurIPS 2021 Poster | Accept (Poster) | This paper presents a two-step lookahead Bayesian optimization (BO) strategy in the presence of inequality constraints, which builds on a previous work on an unconstrained version of the two-step lookahead BO and overcomes the weakness of an earlier multi-step lookahead BO. The paper is generally well-written and give... | 4 | [{"review_id": "j612djtZOA", "reviewer": "Reviewer_F2Td", "summary": "This work suggests a two-step lookahead Bayesian optimization strategy in the presence of inequality constraints, which is non-myopic and computationally efficient. The previous work [1] applies unreliable derivative-free optimization of acquisition ... | Two-step lookahead Bayesian optimization with inequality constraints
Yunxiang Zhang Cornell University y22547@cornel1.cm edu
Xiangyu Zhang Comell University x25560cornel1.edd
Peter I. Frazier Coreell University
pf98@cornel1. edu
Abstract
Recent advances in computationally efficient non-myopic Bayesian optimization offe... | 45,164 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.