paper_id
stringlengths
9
13
title
stringlengths
15
153
paper_url
stringlengths
41
45
authors
listlengths
1
21
type
stringclasses
3 values
primary_area
stringclasses
0 values
abstract
large_stringlengths
310
2.43k
keywords
listlengths
1
32
TL;DR
large_stringlengths
23
250
submission_number
int64
1
11.7k
arxiv_id
stringlengths
10
10
arxiv_id_source
stringclasses
2 values
WN1TaGjVC9U
A generative nonparametric Bayesian model for whole genomes
https://openreview.net/forum?id=WN1TaGjVC9U
[ "Alan Nawzad Amin", "Eli N Weinstein", "Debora Susan Marks" ]
Poster
null
Generative probabilistic modeling of biological sequences has widespread existing and potential use across biology and biomedicine, particularly given advances in high-throughput sequencing, synthesis and editing. However, we still lack methods with nucleotide resolution that are tractable at the scale of whole genomes...
[ "Bayesian nonparametrics", "genomics", "hypothesis testing" ]
We develop a scalable nonparametric Bayesian model of genomic sequences with theoretical guarantees.
7,445
null
null
Tzkev89HeLZ
On Empirical Risk Minimization with Dependent and Heavy-Tailed Data
https://openreview.net/forum?id=Tzkev89HeLZ
[ "Abhishek Roy", "Krishna Balasubramanian", "Murat A Erdogdu" ]
Poster
null
In this work, we establish risk bounds for Empirical Risk Minimization (ERM) with both dependent and heavy-tailed data-generating processes. We do so by extending the seminal works~\cite{pmlr-v35-mendelson14, mendelson2018learning} on the analysis of ERM with heavy-tailed but independent and identically distributed obs...
[ "non-iid learning", "risk bounds", "empirical risk minimization", "concentration inequalities", "small-ball method" ]
This paper develops risk bounds for empirical risk minimization with (polynomially) heavy-tailed and strictly stationary exponentially $\beta$-mixing data generating model.
7,443
2109.02224
title_snapshot
tqQ-8MuSqm
Scaling Up Exact Neural Network Compression by ReLU Stability
https://openreview.net/forum?id=tqQ-8MuSqm
[ "Thiago Serra", "Xin Yu", "Abhinav Kumar", "Srikumar Ramalingam" ]
Poster
null
We can compress a rectifier network while exactly preserving its underlying functionality with respect to a given input domain if some of its neurons are stable. However, current approaches to determine the stability of neurons with Rectified Linear Unit (ReLU) activations require solving or finding a good approximatio...
[ "Scaling", "Neural Networks", "Lossless Compression", "ReLU" ]
We scale up the lossless compression in ReLU neural network based on solving a single optimization problem to identify all stable neurons.
7,441
2102.07804
title_snapshot
q4Dln9kWFA0
Heterogeneous Multi-player Multi-armed Bandits: Closing the Gap and Generalization
https://openreview.net/forum?id=q4Dln9kWFA0
[ "Chengshuai Shi", "Wei Xiong", "Cong Shen", "Jing Yang" ]
Poster
null
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...
[ "Multi-agent System", "Multi-armed Bandits", "Decentralized Learning" ]
This work closed the regret gap from centralized performance in decentralized heterogeneous multi-player multi-armed bandits, and extended the study from the linear reward function to general reward functions.
7,433
2110.14622
title_snapshot
b83ibRX55T
Towards Gradient-based Bilevel Optimization with Non-convex Followers and Beyond
https://openreview.net/forum?id=b83ibRX55T
[ "Risheng Liu", "Yaohua Liu", "Shangzhi Zeng", "Jin Zhang" ]
Spotlight
null
In recent years, Bi-Level Optimization (BLO) techniques have received extensive attentions from both learning and vision communities. A variety of BLO models in complex and practical tasks are of non-convex follower structure in nature (a.k.a., without Lower-Level Convexity, LLC for short). However, this challenging cl...
[ "Bi-level programming", "gradient-based method", "asymptotic convergence", "few-shot classification", "data hyper-cleaning" ]
null
7,432
2110.00455
title_snapshot
f9mSLa07Ncc
Learning latent causal graphs via mixture oracles
https://openreview.net/forum?id=f9mSLa07Ncc
[ "Bohdan Kivva", "Goutham Rajendran", "Pradeep Kumar Ravikumar", "Bryon Aragam" ]
Poster
null
We study the problem of reconstructing a causal graphical model from data in the presence of latent variables. The main problem of interest is recovering the causal structure over the latent variables while allowing for general, potentially nonlinear dependencies. In many practical problems, the dependence between raw ...
[ "causal graphical models", "latent variables", "mixture model", "algorithms" ]
Theoretical guarantees and efficient algorithms for learning causal graphical models with latent variables.
7,425
2106.15563
title_snapshot
5af9FHClUZu
Fast Projection onto the Capped Simplex with Applications to Sparse Regression in Bioinformatics
https://openreview.net/forum?id=5af9FHClUZu
[ "Andersen Ang", "Jianzhu Ma", "Nianjun Liu", "Kun Huang", "Yijie Wang" ]
Poster
null
We consider the problem of projecting a vector onto the so-called k-capped simplex, which is a hyper-cube cut by a hyperplane. For an n-dimensional input vector with bounded elements, we found that a simple algorithm based on Newton's method is able to solve the projection problem to high precision with a complexity ro...
[ "Capped Simplex", "Projection", "Newton's Method", "Sparse Regression via Boolean Relaxation", "Bioinformatics", "GWAS" ]
We find that solving the projection onto the capped simplex by Newton's method is FAST.
7,422
2110.08471
title_snapshot
YV3uoawS5KK
Averaging on the Bures-Wasserstein manifold: dimension-free convergence of gradient descent
https://openreview.net/forum?id=YV3uoawS5KK
[ "Jason Altschuler", "Sinho Chewi", "Patrik Robert Gerber", "Austin J Stromme" ]
Spotlight
null
We study first-order optimization algorithms for computing the barycenter of Gaussian distributions with respect to the optimal transport metric. Although the objective is geodesically non-convex, Riemannian gradient descent empirically converges rapidly, in fact faster than off-the-shelf methods such as Euclidean grad...
[ "Bures-Wasserstein barycenter", "dimension-free convergence", "entropic regularization", "first-order optimization", "geometric median", "non-convex optimization", "Riemannian optimization" ]
We improve state-of-the-art convergence guarantees for Riemannian gradient descent for computing geometric averages of Gaussians.
7,421
2106.08502
title_snapshot
Ke9lCi1vGF
Escaping Saddle Points with Compressed SGD
https://openreview.net/forum?id=Ke9lCi1vGF
[ "Dmitrii Avdiukhin", "Grigory Yaroslavtsev" ]
Poster
null
Stochastic gradient descent (SGD) is a prevalent optimization technique for large-scale distributed machine learning. While SGD computation can be efficiently divided between multiple machines, communication typically becomes a bottleneck in the distributed setting. Gradient compression methods can be used to alleviate...
[ "optimization", "distributed optimization", "nonconvex optimization", "machine learning", "gradient descent", "saddle points" ]
SGD with compressor converges to a second-order stationary point with improved total communication
7,419
2105.10090
title_snapshot
YadmOcMC9aa
Reinforcement Learning with Latent Flow
https://openreview.net/forum?id=YadmOcMC9aa
[ "Wenling Shang", "Xiaofei Wang", "Aravind Srinivas", "Aravind Rajeswaran", "Yang Gao", "Pieter Abbeel", "Michael Laskin" ]
Poster
null
Temporal information is essential to learning effective policies with Reinforcement Learning (RL). However, current state-of-the-art RL algorithms either assume that such information is given as part of the state space or, when learning from pixels, use the simple heuristic of frame-stacking to implicitly capture temp...
[ "Machine Learning", "Reinforcement Learning", "Latent Flow" ]
We introduce FLARE, a pixel-based RL algorithm which utilizes late fusion with latent flow to improve performance on the DeepMind control and Atari pixel-based RL benchmarks.
7,413
2101.01857
title_snapshot
bm1Mrc3WHSe
Rank Overspecified Robust Matrix Recovery: Subgradient Method and Exact Recovery
https://openreview.net/forum?id=bm1Mrc3WHSe
[ "Lijun Ding", "Liwei Jiang", "Yudong Chen", "Qing Qu", "Zhihui Zhu" ]
Poster
null
We study the robust recovery of a low-rank matrix from sparsely and grossly corrupted Gaussian measurements, with no prior knowledge on the intrinsic rank. We consider the robust matrix factorization approach. We employ a robust $\ell_1$ loss function and deal with the challenge of the unknown rank by using an overspec...
[ "Low rank", "Robust recovery", "Subgradient", "Rank overspecification" ]
null
7,412
2109.11154
title_snapshot
XnIYa2OG2sr
An Exact Characterization of the Generalization Error for the Gibbs Algorithm
https://openreview.net/forum?id=XnIYa2OG2sr
[ "Gholamali Aminian", "Yuheng Bu", "Laura Toni", "Miguel R. D. Rodrigues", "Gregory Wornell" ]
Poster
null
Various approaches have been developed to upper bound the generalization error of a supervised learning algorithm. However, existing bounds are often loose and lack of guarantees. As a result, they may fail to characterize the exact generalization ability of a learning algorithm. Our main contribution is an exact chara...
[ "Gibbs algorithm", "generalization error", "information-theoretic bounds", "PAC-Bayesian bounds" ]
Our main contribution is an exact characterization of the expected generalization error of the Gibbs algorithm using symmetrized KL information between the input training samples and the output hypothesis.
7,405
null
null
mekyxmlLJNd
Effective Meta-Regularization by Kernelized Proximal Regularization
https://openreview.net/forum?id=mekyxmlLJNd
[ "Weisen Jiang", "James Kwok", "Yu Zhang" ]
Poster
null
We study the problem of meta-learning, which has proved to be advantageous to accelerate learning new tasks with a few samples. The recent approaches based on deep kernels achieve the state-of-the-art performance. However, the regularizers in their base learners are not learnable. In this paper, we propose an algorithm...
[ "meta-learning" ]
null
7,404
null
null
48uzkHOKMfz
Accelerating Robotic Reinforcement Learning via Parameterized Action Primitives
https://openreview.net/forum?id=48uzkHOKMfz
[ "Murtaza Dalal", "Deepak Pathak", "Ruslan Salakhutdinov" ]
Poster
null
Despite the potential of reinforcement learning (RL) for building general-purpose robotic systems, training RL agents to solve robotics tasks still remains challenging due to the difficulty of exploration in purely continuous action spaces. Addressing this problem is an active area of research with the majority of foc...
[ "reinforcement learning", "robotic manipulation", "motion primitives", "hierarchical RL" ]
We show that a simple redefinition of an RL agent's underlying action space can substantially improve performance on complex robotic control tasks.
7,402
2110.15360
title_snapshot
WigDnV-_Gq
BernNet: Learning Arbitrary Graph Spectral Filters via Bernstein Approximation
https://openreview.net/forum?id=WigDnV-_Gq
[ "Mingguo He", "Zhewei Wei", "Zengfeng Huang", "Hongteng Xu" ]
Poster
null
Many representative graph neural networks, $e.g.$, GPR-GNN and ChebNet, approximate graph convolutions with graph spectral filters. However, existing work either applies predefined filter weights or learns them without necessary constraints, which may lead to oversimplified or ill-posed filters. To overcome these issue...
[ "Graph Neural Networks", "Bernstein Polynomial Approximation", "Arbitrary Spectral Filters" ]
We propose BernNet, a graph neural network that designs and learns an arbitrary spectral filter via Bernstein polynomial approximation.
7,400
2106.10994
title_snapshot
jFMzBeLyTc0
Class-Disentanglement and Applications in Adversarial Detection and Defense
https://openreview.net/forum?id=jFMzBeLyTc0
[ "Kaiwen Yang", "Tianyi Zhou", "Yonggang Zhang", "Xinmei Tian", "Dacheng Tao" ]
Poster
null
What is the minimum necessary information required by a neural net $D(\cdot)$ from an image $x$ to accurately predict its class? Extracting such information in the input space from $x$ can allocate the areas $D(\cdot)$ mainly attending to and shed novel insights to the detection and defense of adversarial attacks. In t...
[ "class disentanglement", "interpretable machine learning", "adversarial detection" ]
We propose a simple VAE+classifier structure to separate the class information from an image by decomposing it into two part.
7,396
null
null
kQDPhAZHYi
S$^3$: Sign-Sparse-Shift Reparametrization for Effective Training of Low-bit Shift Networks
https://openreview.net/forum?id=kQDPhAZHYi
[ "Xinlin Li", "Bang Liu", "Yaoliang Yu", "Wulong Liu", "Chunjing Xu", "Vahid Partovi Nia" ]
Poster
null
Shift neural networks reduce computation complexity by removing expensive multiplication operations and quantizing continuous weights into low-bit discrete values, which are fast and energy-efficient compared to conventional neural networks. However, existing shift networks are sensitive to the weight initialization an...
[ "Convolution Neural Network", "CNN", "Edge Computing", "Edge", "Quantization", "Low-bit", "Multiplication-free", "Bit-shift" ]
S$^3$ re-parameterization for efficient training of low-bit shift networks combating vanishing gradient problem and weight sign freezing problem, 3-bit multiplication-free network compete full-precision model on ImageNet.
7,392
2107.03453
title_snapshot
aoXERVeC7cC
Selective Sampling for Online Best-arm Identification
https://openreview.net/forum?id=aoXERVeC7cC
[ "Romain Camilleri", "Zhihan Xiong", "Maryam Fazel", "Lalit K Jain", "Kevin Jamieson" ]
Poster
null
This work considers the problem of selective-sampling for best-arm identification. Given a set of potential options $\mathcal{Z}\subset\mathbb{R}^d$, a learner aims to compute with probability greater than $1-\delta$, $\arg\max_{z\in \mathcal{Z}} z^{\top}\theta_{\ast}$ where $\theta_{\ast}$ is unknown. At each time ste...
[ "active learning", "bandits" ]
null
7,388
2110.14864
title_snapshot
sRojdWhXJx
Revitalizing CNN Attention via Transformers in Self-Supervised Visual Representation Learning
https://openreview.net/forum?id=sRojdWhXJx
[ "Chongjian GE", "Youwei Liang", "Yibing Song", "Jianbo Jiao", "Jue Wang", "Ping Luo" ]
Poster
null
Studies on self-supervised visual representation learning (SSL) improve encoder backbones to discriminate training samples without labels. While CNN encoders via SSL achieve comparable recognition performance to those via supervised learning, their network attention is under-explored for further improvement. Motivated ...
[ "Self-Supervised Visual Representation Learning", "Vision Transformers" ]
We revitalize CNN encoder attentions via transformer in self-supervised visual representation learning
7,387
2110.05340
title_judge
oAjn5-AgSd
Local Signal Adaptivity: Provable Feature Learning in Neural Networks Beyond Kernels
https://openreview.net/forum?id=oAjn5-AgSd
[ "Stefani Karp", "Ezra Winston", "Yuanzhi Li", "Aarti Singh" ]
Poster
null
Neural networks have been shown to outperform kernel methods in practice (including neural tangent kernels). Most theoretical explanations of this performance gap focus on learning a complex hypothesis class; in some cases, it is unclear whether this hypothesis class captures realistic data. In this work, we propose a ...
[ "deep learning theory", "neural networks", "kernels" ]
We propose an explanation for the gap between neural networks and their corresponding neural tangent kernels based on the ability of neural networks to find a sparse, localized signal in the presence of noise.
7,384
null
null
Z2ZWIvNeVUl
On the Stochastic Stability of Deep Markov Models
https://openreview.net/forum?id=Z2ZWIvNeVUl
[ "Jan Drgona", "Sayak Mukherjee", "Jiaxin Zhang", "Frank Y Liu", "Mahantesh Halappanavar" ]
Poster
null
Deep Markov models (DMM) are generative models which are scalable and expressive generalization of Markov models for representation, learning, and inference problems. However, the fundamental stochastic stability guarantees of such models have not been thoroughly investigated. In this paper, we present a novel stabilit...
[ "markov models", "deep neural networks", "stochastic stability", "dynamical systems" ]
The paper presents a novel stability analysis method for deep Markov models and provide sufficient conditions of DMM's stochastic stability.
7,382
2111.04601
title_snapshot
wRFj6EKvpl
How Data Augmentation affects Optimization for Linear Regression
https://openreview.net/forum?id=wRFj6EKvpl
[ "Boris Hanin", "Yi Sun" ]
Poster
null
Though data augmentation has rapidly emerged as a key tool for optimization in modern machine learning, a clear picture of how augmentation schedules affect optimization and interact with optimization hyperparameters such as learning rate is nascent. In the spirit of classical convex optimization and recent work on imp...
[ "data augmentation", "stochastic optimization", "convex optimization" ]
We fully characterize the impact of data augmentation on optimization in the case of linear regression with MSE loss.
7,379
2010.11171
title_snapshot
rA9HFxFT7th
Sageflow: Robust Federated Learning against Both Stragglers and Adversaries
https://openreview.net/forum?id=rA9HFxFT7th
[ "Jungwuk Park", "Dong-Jun Han", "Minseok Choi", "Jaekyun Moon" ]
Poster
null
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...
[ "Federated Learning", "Stragglers", "Adversaries" ]
We propose a new federated learning algorithm that handles both stragglers and adversaries simultaneously, via staleness-aware grouping with entropy-based filtering and loss-weighted averaging.
7,375
null
null
wJXWzCsGlZw
Practical, Provably-Correct Interactive Learning in the Realizable Setting: The Power of True Believers
https://openreview.net/forum?id=wJXWzCsGlZw
[ "JULIAN KATZ-SAMUELS", "Blake Mason", "Kevin Jamieson", "Rob Nowak" ]
Poster
null
We consider interactive learning in the realizable setting and develop a general framework to handle problems ranging from best arm identification to active classification. We begin our investigation with the observation that agnostic algorithms \emph{cannot} be minimax-optimal in the realizable setting. Hence, we desi...
[ "Active Learning", "Active Classification", "Multi-Armed Bandits" ]
We consider interactive learning in the realizable setting and design novel computationally efficient algorithms for general function classes that match the minimax lower bound up to logarithmic factors.
7,368
2111.04915
title_snapshot
DvxH_RCnSj3
Implicit Task-Driven Probability Discrepancy Measure for Unsupervised Domain Adaptation
https://openreview.net/forum?id=DvxH_RCnSj3
[ "Mao Li", "Kaiqi Jiang", "Xinhua Zhang" ]
Poster
null
Probability discrepancy measure is a fundamental construct for numerous machine learning models such as weakly supervised learning and generative modeling. However, most measures overlook the fact that the distributions are not the end-product of learning, but are the basis of downstream predictor. Therefore it is im...
[ "probability discrepancy measure", "unsupervised domain adaptation" ]
Warping the probability discrepancy measure towards the end tasks can significantly improve unsupervised domain adaptation.
7,364
null
null
Xs-vglI4EBi
On the Convergence Theory of Debiased Model-Agnostic Meta-Reinforcement Learning
https://openreview.net/forum?id=Xs-vglI4EBi
[ "Alireza Fallah", "Kristian Georgiev", "Aryan Mokhtari", "Asuman E. Ozdaglar" ]
Poster
null
We consider Model-Agnostic Meta-Learning (MAML) methods for Reinforcement Learning (RL) problems, where the goal is to find a policy using data from several tasks represented by Markov Decision Processes (MDPs) that can be updated by one step of \textit{stochastic} policy gradient for the realized MDP. In particular, u...
[ "meta-learning theory", "reinforcement learning theory", "optimization" ]
We resolve the bias issue in the update of original Model-Agnostic Meta-Learning (MAML) method for the reinforcement learning problem and provide convergence guarantees for our method.
7,363
2002.05135
title_snapshot
xRLT28nnlFV
On Robust Optimal Transport: Computational Complexity and Barycenter Computation
https://openreview.net/forum?id=xRLT28nnlFV
[ "Khang Le", "Huy Nguyen", "Quang Minh Nguyen", "Tung Pham", "Hung Bui", "Nhat Ho" ]
Poster
null
We consider robust variants of the standard optimal transport, named robust optimal transport, where marginal constraints are relaxed via Kullback-Leibler divergence. We show that Sinkhorn-based algorithms can approximate the optimal cost of robust optimal transport in $\widetilde{\mathcal{O}}(\frac{n^2}{\varepsilon})$...
[ "optimal transport", "optimization", "complexity", "robustness" ]
We provide computational complexity for a robust variant of optimal transport and the corresponding barycenter problem.
7,359
2102.06857
title_snapshot
ZgUZmeV1Mtu
Few-Round Learning for Federated Learning
https://openreview.net/forum?id=ZgUZmeV1Mtu
[ "Younghyun Park", "Dong-Jun Han", "Do-Yeon Kim", "Jun Seo", "Jaekyun Moon" ]
Poster
null
In federated learning (FL), a number of distributed clients targeting the same task collaborate to train a single global model without sharing their data. The learning process typically starts from a randomly initialized or some pretrained model. In this paper, we aim at designing an initial model based on which an arb...
[ "Federated Learning" ]
We propose a meta-training algorithm to design an initial model based on which an arbitrary group of clients can obtain a global model for its own purpose, within only a few rounds of FL
7,357
null
null
NE0YlkgRo9x
A single gradient step finds adversarial examples on random two-layers neural networks
https://openreview.net/forum?id=NE0YlkgRo9x
[ "Sebastien Bubeck", "Yeshwanth Cherapanamjeri", "Gauthier Gidel", "Remi Tachet des Combes" ]
Spotlight
null
Daniely and Schacham recently showed that gradient descent finds adversarial examples on random undercomplete two-layers ReLU neural networks. The term “undercomplete” refers to the fact that their proof only holds when the number of neurons is a vanishing fraction of the ambient dimension. We extend their result to th...
[ "adversarial examples", "random neural networks", "deep-learning theory" ]
We prove that a single gradient step finds adversarial examples on random two-layers neural networks.
7,355
2104.03863
title_snapshot
qb0qTdxPWzY
List-Decodable Mean Estimation in Nearly-PCA Time
https://openreview.net/forum?id=qb0qTdxPWzY
[ "Ilias Diakonikolas", "Daniel Kane", "Daniel Kongsgaard", "Jerry Li", "Kevin Tian" ]
Spotlight
null
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...
[ "robust statistics", "learning theory", "mixture models", "semidefinite programming", "list-decodable learning" ]
We give a state-of-the-art algorithm for list-decodable mean estimation, the robust generalization of learning mixture models, attaining optimal error in polylogarithmic calls to approximate PCA.
7,354
2011.09973
title_snapshot
YxxzNLfXBz
Conservative Data Sharing for Multi-Task Offline Reinforcement Learning
https://openreview.net/forum?id=YxxzNLfXBz
[ "Tianhe Yu", "Aviral Kumar", "Yevgen Chebotar", "Karol Hausman", "Sergey Levine", "Chelsea Finn" ]
Poster
null
Offline reinforcement learning (RL) algorithms have shown promising results in domains where abundant pre-collected data is available. However, prior methods focus on solving individual problems from scratch with an offline dataset without considering how an offline RL agent can acquire multiple skills. We argue that a...
[ "offline reinforcement learning", "multi-task reinforcement learning", "deep reinforcement learning" ]
null
7,353
2109.08128
title_snapshot
Tv0O_cAdKtW
On sensitivity of meta-learning to support data
https://openreview.net/forum?id=Tv0O_cAdKtW
[ "Mayank Agarwal", "Mikhail Yurochkin", "Yuekai Sun" ]
Poster
null
Meta-learning algorithms are widely used for few-shot learning. For example, image recognition systems that readily adapt to unseen classes after seeing only a few labeled examples. Despite their success, we show that modern meta-learning algorithms are extremely sensitive to the data used for adaptation, i.e. support ...
[ "meta-learning", "sensitivity", "robustness" ]
We demonstrate that meta-learning algorithms applied to few-shot learning are extremely sensitive to the support data.
7,352
2110.13953
title_snapshot
09-zkOYoVof
Generalization of Model-Agnostic Meta-Learning Algorithms: Recurring and Unseen Tasks
https://openreview.net/forum?id=09-zkOYoVof
[ "Alireza Fallah", "Aryan Mokhtari", "Asuman E. Ozdaglar" ]
Poster
null
In this paper, we study the generalization properties of Model-Agnostic Meta-Learning (MAML) algorithms for supervised learning problems. We focus on the setting in which we train the MAML model over $m$ tasks, each with $n$ data points, and characterize its generalization error from two points of view: First, we assum...
[ "meta-learning theory", "generalization bounds", "algorithmic stability", "convex optimization" ]
null
7,347
2102.03832
title_snapshot
YlM3tey8Z5I
Self-Interpretable Model with Transformation Equivariant Interpretation
https://openreview.net/forum?id=YlM3tey8Z5I
[ "Yipei Wang", "Xiaoqian Wang" ]
Poster
null
With the proliferation of machine learning applications in the real world, the demand for explaining machine learning predictions continues to grow especially in high-stakes fields. Recent studies have found that interpretation methods can be sensitive and unreliable, where the interpretations can be disturbed by pertu...
[ "interpretable machine learning", "transformation equivariance", "computer vision" ]
We propose a self-interpretable model that has transformation-equivariant interpretations, and comparable expressive power as benchmark black-box model.
7,346
null
null
ChWy1anEuow
Risk Bounds for Over-parameterized Maximum Margin Classification on Sub-Gaussian Mixtures
https://openreview.net/forum?id=ChWy1anEuow
[ "Yuan Cao", "Quanquan Gu", "Misha Belkin" ]
Poster
null
Modern machine learning systems such as deep neural networks are often highly over-parameterized so that they can fit the noisy training data exactly, yet they can still achieve small test errors in practice. In this paper, we study this "benign overfitting" phenomenon of the maximum margin classifier for linear classi...
[ "maximum margin classification", "over-parameterization", "benign overfitting" ]
null
7,340
2104.13628
title_snapshot
KCd-3Pz8VjM
Automatic Unsupervised Outlier Model Selection
https://openreview.net/forum?id=KCd-3Pz8VjM
[ "Yue Zhao", "Ryan Rossi", "Leman Akoglu" ]
Poster
null
Given an unsupervised outlier detection task on a new dataset, how can we automatically select a good outlier detection algorithm and its hyperparameter(s) (collectively called a model)? In this work, we tackle the unsupervised outlier model selection (UOMS) problem, and propose MetaOD, a principled, data-driven approa...
[ "outlier detection", "anomaly detection", "unsupervised learning", "model selection", "automated machine learning" ]
Automatic Unsupervised Outlier Model Selection
7,338
null
null
pu6loAVvBZb
Robust Regression Revisited: Acceleration and Improved Estimation Rates
https://openreview.net/forum?id=pu6loAVvBZb
[ "Arun Jambulapati", "Jerry Li", "Tselil Schramm", "Kevin Tian" ]
Poster
null
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{...
[ "robust statistics", "stochastic optimization", "linear regression", "acceleration" ]
We present nearly-linear time algorithms for statistical regression problems with improved runtime or estimation guarantees over the prior state-of-the-art.
7,337
2106.11938
title_snapshot
2ybxtABV2Og
BNS: Building Network Structures Dynamically for Continual Learning
https://openreview.net/forum?id=2ybxtABV2Og
[ "Qi Qin", "Wenpeng Hu", "Han Peng", "Dongyan Zhao", "Bing Liu" ]
Poster
null
Continual learning (CL) of a sequence of tasks is often accompanied with the catastrophic forgetting(CF) problem. Existing research has achieved remarkable results in overcoming CF, especially for task continual learning. However, limited work has been done to achieve another important goal of CL,knowledge transfer.In ...
[ "continual learning" ]
A reinforcement learning based continual learning method is proposed.
7,331
null
null
ZDMqRGSksHs
Adaptive Sampling for Minimax Fair Classification
https://openreview.net/forum?id=ZDMqRGSksHs
[ "Shubhanshu Shekhar", "Greg Fields", "Mohammad Ghavamzadeh", "Tara Javidi" ]
Poster
null
Machine learning models trained on uncurated datasets can often end up adversely affecting inputs belonging to underrepresented groups. To address this issue, we consider the problem of adaptively constructing training sets which allow us to learn classifiers that are fair in a {\em minimax} sense. We first propose an ...
[ "Fairness", "deep learning", "active sampling" ]
We propose and analyze an algorithm to adaptively build datasets to ensure minimax fairness among specified subpopulations.
7,327
2103.00755
title_snapshot
MlFcgL2AP4d
Near-Optimal Offline Reinforcement Learning via Double Variance Reduction
https://openreview.net/forum?id=MlFcgL2AP4d
[ "Ming Yin", "Yu Bai", "Yu-Xiang Wang" ]
Poster
null
We consider the problem of offline reinforcement learning (RL) --- a well-motivated setting of RL that aims at policy optimization using only historical data. Despite its wide applicability, theoretical understandings of offline RL, such as its optimal sample complexity, remain largely open even in basic settings such...
[ "Theory", "Reinforcement Learning Theory", "Markov Decision Process Theory" ]
null
7,323
2102.01748
title_snapshot
6dUJPrPPUau
Few-Shot Data-Driven Algorithms for Low Rank Approximation
https://openreview.net/forum?id=6dUJPrPPUau
[ "Piotr Indyk", "Tal Wagner", "David Woodruff" ]
Poster
null
Recently, data-driven and learning-based algorithms for low rank matrix approximation were shown to outperform classical data-oblivious algorithms by wide margins in terms of accuracy. Those algorithms are based on the optimization of sparse sketching matrices, which lead to large savings in time and memory during tes...
[ "low rank approximation", "numerical linear algebra", "learning-based algorithms", "svd", "matrix sketching" ]
We provide new, time and sample efficient, and interpretable algorithms for data-driven low rank matrix approximation. We provide both theoretical results and empirical evaluation for our algorithms.
7,305
null
null
mjyMGFL8N2
Provable Guarantees for Self-Supervised Deep Learning with Spectral Contrastive Loss
https://openreview.net/forum?id=mjyMGFL8N2
[ "Jeff Z. HaoChen", "Colin Wei", "Adrien Gaidon", "Tengyu Ma" ]
Oral
null
Recent works in self-supervised learning have advanced the state-of-the-art by relying on the contrastive learning paradigm, which learns representations by pushing positive pairs, or similar examples from the same class, closer together while keeping negative pairs far apart. Despite the empirical successes, theoretic...
[ "theory", "deep learning theory", "unsupervised learning theory", "representation learning theory" ]
We propose a novel theoretical framework for studying self-supervised learning algorithms.
7,303
2106.04156
title_snapshot
x8k1nAoGu1U
Fast Doubly-Adaptive MCMC to Estimate the Gibbs Partition Function with Weak Mixing Time Bounds
https://openreview.net/forum?id=x8k1nAoGu1U
[ "Shahrzad Haddadan", "Yue Zhuang", "Cyrus Cousins", "Eli Upfal" ]
Poster
null
We present a novel method for reducing the computational complexity of rigorously estimating the partition functions of Gibbs (or Boltzmann) distributions, which arise ubiquitously in probabilistic graphical models. A major obstacle to applying the Gibbs distribution in practice is the need to estimate their partition ...
[ "MCMC", "Gibbs distribution", "partition functions" ]
we propose a doubly adaptive algorithms for estimating partition function of Gibbs directions and show theoretically and through experiments that it beats the state of the art
7,296
2111.07372
title_snapshot
FMPuzXV1fR
Breaking the centralized barrier for cross-device federated learning
https://openreview.net/forum?id=FMPuzXV1fR
[ "Sai Praneeth Karimireddy", "Martin Jaggi", "Satyen Kale", "Mehryar Mohri", "Sashank J. Reddi", "Sebastian U Stich", "Ananda Theertha Suresh" ]
Poster
null
Federated learning (FL) is a challenging setting for optimization due to the heterogeneity of the data across different clients which gives rise to the client drift phenomenon. In fact, obtaining an algorithm for FL which is uniformly better than simple centralized training has been a major open problem thus far. In th...
[ "Federated Learning", "Non-convex Optimization", "Distributed Optimization", "Communication Complexity" ]
New framework which i) adapts arbitary centralized algorithms to the federated setting, and ii) obtains the first rates which are uniformly better than centralized training.
7,292
null
null
_pmQOVi3gHx
Linear and Kernel Classification in the Streaming Model: Improved Bounds for Heavy Hitters
https://openreview.net/forum?id=_pmQOVi3gHx
[ "Arvind V. Mahankali", "David Woodruff" ]
Poster
null
We study linear and kernel classification in the streaming model. For linear classification, we improve upon the algorithm of (Tai, et al. 2018), which solves the $\ell_1$ point query problem on the optimal weight vector $w_* \in \mathbb{R}^d$ in sublinear space. We first give an algorithm solving the more difficult $\...
[ "Linear classification", "heavy hitters", "kernel classification", "streaming algorithms", "sketching", "tensors" ]
null
7,285
null
null
kcI3T5qe1jr
Controllable and Compositional Generation with Latent-Space Energy-Based Models
https://openreview.net/forum?id=kcI3T5qe1jr
[ "Weili Nie", "Arash Vahdat", "Anima Anandkumar" ]
Poster
null
Controllable generation is one of the key requirements for successful adoption of deep generative models in real-world applications, but it still remains as a great challenge. In particular, the compositional ability to generate novel concept combinations is out of reach for most current models. In this work, we use en...
[ "Controllable generation", "compositional generation", "image synthesis", "energy-based models", "deep generative models", "neural ODEs" ]
We introduce an EBM in the latent space of existing generative models and a new ODE sampling method for controllable and compositional generation.
7,281
2110.10873
title_snapshot
817F5yuNAf1
Automatic and Harmless Regularization with Constrained and Lexicographic Optimization: A Dynamic Barrier Approach
https://openreview.net/forum?id=817F5yuNAf1
[ "Chengyue Gong", "Xingchao Liu", "qiang liu" ]
Poster
null
Many machine learning tasks have to make a trade-off between two loss functions, typically the main data-fitness loss and an auxiliary loss. The most widely used approach is to optimize the linear combination of the objectives, which, however, requires manual tuning of the combination coefficient and is theoretically ...
[ "constrained optimization", "lexicographic optimization", "multi-objective optimization", "pareto set", "multi-task learning" ]
null
7,276
null
null
8bbevt2MKPX
Continuous-time edge modelling using non-parametric point processes
https://openreview.net/forum?id=8bbevt2MKPX
[ "Xuhui Fan", "Bin Li", "Feng Zhou", "Scott A Sisson" ]
Poster
null
The mutually-exciting Hawkes process (ME-HP) is a natural choice to model reciprocity, which is an important attribute of continuous-time edge (dyadic) data. However, existing ways of implementing the ME-HP for such data are either inflexible, as the exogenous (background) rate functions are typically constant and the ...
[ "Continuous-time Edges", "Sigmoidal Gaussian Process", "Variational Inference" ]
We use Sigmoidal Gaussian Process modulated point processes to model continuous-time edges.
7,274
null
null
a7APmM4B9d
Decision Transformer: Reinforcement Learning via Sequence Modeling
https://openreview.net/forum?id=a7APmM4B9d
[ "Lili Chen", "Kevin Lu", "Aravind Rajeswaran", "Kimin Lee", "Aditya Grover", "Michael Laskin", "Pieter Abbeel", "Aravind Srinivas", "Igor Mordatch" ]
Poster
null
We introduce a framework that abstracts Reinforcement Learning (RL) as a sequence modeling problem. This allows us to draw upon the simplicity and scalability of the Transformer architecture, and associated advances in language modeling such as GPT-x and BERT. In particular, we present Decision Transformer, an architec...
[ "transformers", "reinforcement learning", "deep learning", "generative modeling" ]
Transformers can do offline RL successfully.
7,270
2106.01345
title_snapshot
7PkfLkyLMRM
Mitigating Covariate Shift in Imitation Learning via Offline Data With Partial Coverage
https://openreview.net/forum?id=7PkfLkyLMRM
[ "Jonathan Daniel Chang", "Masatoshi Uehara", "Dhruv Sreenivas", "Rahul Kidambi", "Wen Sun" ]
Poster
null
This paper studies offline Imitation Learning (IL) where an agent learns to imitate an expert demonstrator without additional online environment interactions. Instead, the learner is presented with a static offline dataset of state-action-next state triples from a potentially less proficient behavior policy. We introdu...
[ "Offline Imitation Learning", "Imitation Learning" ]
We present an algorithmic framework to mitigate the covariate shift issue in imitation learning using offline data with partial coverage and the principle of pessimism in the face of uncertainty.
7,265
2106.03207
title_judge
-zgb2v8vV_w
Adaptive Risk Minimization: Learning to Adapt to Domain Shift
https://openreview.net/forum?id=-zgb2v8vV_w
[ "Marvin Mengxin Zhang", "Henrik Marklund", "Nikita Dhawan", "Abhishek Gupta", "Sergey Levine", "Chelsea Finn" ]
Poster
null
A fundamental assumption of most machine learning algorithms is that the training and test data are drawn from the same underlying distribution. However, this assumption is violated in almost all practical applications: machine learning systems are regularly tested under distribution shift, due to changing temporal cor...
[ "distribution shift", "domain generalization", "test time adaptation" ]
null
7,260
2007.02931
title_snapshot
ebIORrYImx
Practical Near Neighbor Search via Group Testing
https://openreview.net/forum?id=ebIORrYImx
[ "Joshua Engels", "Benjamin Coleman", "Anshumali Shrivastava" ]
Spotlight
null
We present a new algorithm for the approximate near neighbor problem that combines classical ideas from group testing with locality-sensitive hashing (LSH). We reduce the near neighbor search problem to a group testing problem by designating neighbors as "positives," non-neighbors as "negatives," and approximate member...
[ "group testing", "locality sensitive hashing", "near neighbor search", "index" ]
We combine group testing with locality sensitive hashing to develop a near neighbor search algorithm with 10x faster query time on high-dimensional datasets.
7,254
2106.11565
title_snapshot
twz1QqzU0Hp
A No-go Theorem for Robust Acceleration in the Hyperbolic Plane
https://openreview.net/forum?id=twz1QqzU0Hp
[ "Linus Hamilton", "Ankur Moitra" ]
Poster
null
In recent years there has been significant effort to adapt the key tools and ideas in convex optimization to the Riemannian setting. One key challenge has remained: Is there a Nesterov-like accelerated gradient method for geodesically convex functions on a Riemannian manifold? Recent work has given partial answers and ...
[ "geodesic convexity", "acceleration", "lower bounds" ]
We prove that in a noisy setting, there is no analogue of accelerated gradient descent for geodesically convex functions on the hyperbolic plane.
7,252
null
null
UYI6Sk_3Nox
Low-dimensional Structure in the Space of Language Representations is Reflected in Brain Responses
https://openreview.net/forum?id=UYI6Sk_3Nox
[ "Richard Antonello", "Javier S. Turek", "Vy A. Vo", "Alexander Huth" ]
Poster
null
How related are the representations learned by neural language models, translation models, and language tagging tasks? We answer this question by adapting an encoder-decoder transfer learning method from computer vision to investigate the structure among 100 different feature spaces extracted from hidden representatio...
[ "fMRI Encoding Models", "Language Representations", "Natural Language Processing" ]
We show that language representations from NLP models have low-dimensional structure and that this structure is reflected in brain responses to those representations.
7,244
2106.05426
title_snapshot
ii5mGEbRo93
Logarithmic Regret in Feature-based Dynamic Pricing
https://openreview.net/forum?id=ii5mGEbRo93
[ "Jianyu Xu", "Yu-Xiang Wang" ]
Spotlight
null
Feature-based dynamic pricing is an increasingly popular model of setting prices for highly differentiated products with applications in digital marketing, online sales, real estate and so on. The problem was formally studied as an online learning problem [Javanmard & Nazerzadeh, 2019] where a seller needs to propose p...
[ "dynamic pricing", "online learning", "adversarial features", "optimal regret", "affine invariant", "distribution-free." ]
We present algorithms that guarantees logarithmic (minimax) regrets in both stochastic and adversarial feature-based dynamic pricing problems with market noises.
7,237
2102.10221
title_snapshot
XGSQfOVxVp4
On the Variance of the Fisher Information for Deep Learning
https://openreview.net/forum?id=XGSQfOVxVp4
[ "Alexander Soen", "Ke Sun" ]
Poster
null
In the realm of deep learning, the Fisher information matrix (FIM) gives novel insights and useful tools to characterize the loss landscape, perform second-order optimization, and build geometric learning theories. The exact FIM is either unavailable in closed form or too expensive to compute. In practice, it is almost...
[ "Fisher information", "natural gradient", "Cramer-Rao Lower Bound", "deep learning" ]
We explore the variance of the Fisher information matrix in the context of deep learning.
7,229
2107.04205
title_snapshot
NCDMYD2y5kK
Deep Extrapolation for Attribute-Enhanced Generation
https://openreview.net/forum?id=NCDMYD2y5kK
[ "Alvin Chan", "Ali Madani", "Ben Krause", "Nikhil Naik" ]
Poster
null
Attribute extrapolation in sample generation is challenging for deep neural networks operating beyond the training distribution. We formulate a new task for extrapolation in sequence generation, focusing on natural language and proteins, and propose GENhance, a generative framework that enhances attributes through a le...
[ "extrapolation", "generative modeling", "controllable generation", "protein design" ]
How do we generate sequences that extrapolate beyond the training distribution?
7,205
2107.02968
title_snapshot
fqfHJqNy_uY
Rates of Estimation of Optimal Transport Maps using Plug-in Estimators via Barycentric Projections
https://openreview.net/forum?id=fqfHJqNy_uY
[ "NABARUN DEB", "Promit Ghosal", "Bodhisattva Sen" ]
Poster
null
Optimal transport maps between two probability distributions $\mu$ and $\nu$ on $\R^d$ have found extensive applications in both machine learning and statistics. In practice, these maps need to be estimated from data sampled according to $\mu$ and $\nu$. Plug-in estimators are perhaps most popular in estimating transpo...
[ "Discrete-discrete optimal transport", "Kantorovich relaxation", "Legendre-Fenchel dual", "Semi-discrete Optimal Transport", "Wasserstein barycenter", "Wasserstein Distance." ]
We provide a new stability estimate for barycentric projections under minimal smoothness assumptions which we use to prove rates of convergence for general plug-in estimators of optimal transport maps.
7,202
2107.01718
title_snapshot
LJjC6DmSkgT
Continual Learning via Local Module Composition
https://openreview.net/forum?id=LJjC6DmSkgT
[ "Oleksiy Ostapenko", "Pau Rodriguez", "Massimo Caccia", "Laurent Charlin" ]
Poster
null
Modularity is a compelling solution to continual learning (CL), the problem of modeling sequences of related tasks. Learning and then composing modules to solve different tasks provides an abstraction to address the principal challenges of CL including catastrophic forgetting, backward and forward transfer across tasks...
[ "continual learning", "modularity", "compositionality", "neverending learning", "lifelong learning", "plasticity", "plasticity-stability dilemma", "task incremental learning", "multitask learning", "routing", "dynamic architectures", "model growing", "out-of-distribution generalization", "...
We introduce an approach to modular Continual Learning where each module can “decide” about its relevancy given an input.
7,196
2111.07736
title_snapshot
JpDlWGTBHB
Probabilistic Attention for Interactive Segmentation
https://openreview.net/forum?id=JpDlWGTBHB
[ "Prasad Gabbur", "Manjot Bilkhu", "Javier Movellan" ]
Spotlight
null
We provide a probabilistic interpretation of attention and show that the standard dot-product attention in transformers is a special case of Maximum A Posteriori (MAP) inference. The proposed approach suggests the use of Expectation Maximization algorithms for on-line adaptation of key and value model parameters. This ...
[ "Attention", "Transformers", "Probabilistic model", "Gaussian mixture model", "Interactive segmentation", "Semantic segmentation" ]
A new perspective of attention as a probabilistic generative model with applications to interactive image segmentation.
7,195
2106.15338
title_snapshot
70kOIgjKhbA
When does Contrastive Learning Preserve Adversarial Robustness from Pretraining to Finetuning?
https://openreview.net/forum?id=70kOIgjKhbA
[ "Lijie Fan", "Sijia Liu", "Pin-Yu Chen", "Gaoyuan Zhang", "Chuang Gan" ]
Poster
null
Contrastive learning (CL) can learn generalizable feature representations and achieve state-of-the-art performance of downstream tasks by finetuning a linear classifier on top of it. However, as adversarial robustness becomes vital in image classification, it remains unclear whether or not CL is able to preserve robu...
[ "Adversarial robustness", "self-supervised learning", "pretraining and finetuning" ]
This work revisits and advances contrastive learning principles through the lens of adversarial robustness, aiming to improve robustness transferability in the self-supervised pretraining + supervised finetuning paradigm
7,194
2111.01124
title_snapshot
_Rtm4rYnIIL
MobILE: Model-Based Imitation Learning From Observation Alone
https://openreview.net/forum?id=_Rtm4rYnIIL
[ "Rahul Kidambi", "Jonathan Daniel Chang", "Wen Sun" ]
Poster
null
This paper studies Imitation Learning from Observations alone (ILFO) where the learner is presented with expert demonstrations that consist only of states visited by an expert (without access to actions taken by the expert). We present a provably efficient model-based framework MobILE to solve the ILFO problem. MobILE ...
[ "Imitation Learning from Observation Alone", "Imitation Learning", "Imitation Learning Theory" ]
We present an efficient model-based algorithm with strategic exploration for solving the imitation learning from observations alone problem.
7,191
2102.10769
title_snapshot
tjwQaOI9tdy
Symbolic Regression via Deep Reinforcement Learning Enhanced Genetic Programming Seeding
https://openreview.net/forum?id=tjwQaOI9tdy
[ "Terrell N. Mundhenk", "Mikel Landajuela", "Ruben Glatt", "Claudio P. Santiago", "Daniel faissol", "Brenden K. Petersen" ]
Poster
null
Symbolic regression is the process of identifying mathematical expressions that fit observed output from a black-box process. It is a discrete optimization problem generally believed to be NP-hard. Prior approaches to solving the problem include neural-guided search (e.g. using reinforcement learning) and genetic progr...
[ "symbolic regression", "genetic programming", "neural-guided search", "reinforcement learning" ]
We use a hybrid genetic programming and neural-guided search approach to solve symbolic regression.
7,188
2111.00053
title_judge
GEm4o9A6Jfb
PLUR: A Unifying, Graph-Based View of Program Learning, Understanding, and Repair
https://openreview.net/forum?id=GEm4o9A6Jfb
[ "Zimin Chen", "Vincent Josua Hellendoorn", "Pascal Lamblin", "Petros Maniatis", "Pierre-Antoine Manzagol", "Daniel Tarlow", "Subhodeep Moitra" ]
Spotlight
null
Machine learning for understanding and editing source code has recently attracted significant interest, with many developments in new models, new code representations, and new tasks. This proliferation can appear disparate and disconnected, making each approach seemingly unique and incompatible, thus obscuring the core...
[ "learning for code", "program understanding", "program repair", "relation-aware transformers", "graph-based deep learning" ]
A single graph-based architecture can be applied to 16 seemingly different ML4Code tasks and achieves great results.
7,185
null
null
gbcsmD3Iznu
End-to-End Weak Supervision
https://openreview.net/forum?id=gbcsmD3Iznu
[ "Salva Rühling Cachay", "Benedikt Boecking", "Artur Dubrawski" ]
Poster
null
Aggregating multiple sources of weak supervision (WS) can ease the data-labeling bottleneck prevalent in many machine learning applications, by replacing the tedious manual collection of ground truth labels. Current state of the art approaches that do not use any labeled training data, however, require two separate mo...
[ "weak supervision", "deep learning", "data programming" ]
A neural end-to-end system that learns exclusively from multiple sources of weak supervision
7,184
2107.02233
title_snapshot
ELU8Bu1Z9w1
Autonomous Reinforcement Learning via Subgoal Curricula
https://openreview.net/forum?id=ELU8Bu1Z9w1
[ "Archit Sharma", "Abhishek Gupta", "Sergey Levine", "Karol Hausman", "Chelsea Finn" ]
Poster
null
Reinforcement learning (RL) promises to enable autonomous acquisition of complex behaviors for diverse agents. However, the success of current reinforcement learning algorithms is predicated on an often under-emphasised requirement -- each trial needs to start from a fixed initial state distribution. Unfortunately, res...
[ "reinforcement learning", "curriculum", "autonomous learning", "reset-free reinforcement learning" ]
null
7,181
2107.12931
title_snapshot
vRwnHlAgK5x
Coupled Segmentation and Edge Learning via Dynamic Graph Propagation
https://openreview.net/forum?id=vRwnHlAgK5x
[ "Zhiding Yu", "Rui Huang", "Wonmin Byeon", "Sifei Liu", "Guilin Liu", "Thomas Breuel", "Anima Anandkumar", "Jan Kautz" ]
Poster
null
Image segmentation and edge detection are both central problems in perceptual grouping. It is therefore interesting to study how these two tasks can be coupled to benefit each other. Indeed, segmentation can be easily transformed into contour edges to guide edge learning. However, the converse is nontrivial since gener...
[ "Semantic Segmentation", "Semantic Edge Detection", "Structured Prediction", "Propagation Network" ]
A coupled learning framework for joint semantic segmentation and semantic edge detection
7,178
null
null
zO6Q8q2AmbV
On Component Interactions in Two-Stage Recommender Systems
https://openreview.net/forum?id=zO6Q8q2AmbV
[ "Jiri Hron", "Karl Krauth", "Michael Jordan", "Niki Kilbertus" ]
Poster
null
Thanks to their scalability, two-stage recommenders are used by many of today's largest online platforms, including YouTube, LinkedIn, and Pinterest. These systems produce recommendations in two steps: (i) multiple nominators—tuned for low prediction latency—preselect a small subset of candidates from the whole item po...
[ "recommender systems", "mixture of experts", "bandits", "scalability" ]
Candidate generators in two-stage systems can be viewed as experts for their item subset, and thus trained jointly using Mixture-of-Experts algorithms.
7,173
2106.14979
title_snapshot
W9oywyjO8VN
Tractable Regularization of Probabilistic Circuits
https://openreview.net/forum?id=W9oywyjO8VN
[ "Anji Liu", "Guy Van den Broeck" ]
Spotlight
null
Probabilistic Circuits (PCs) are a promising avenue for probabilistic modeling. They combine advantages of probabilistic graphical models (PGMs) with those of neural networks (NNs). Crucially, however, they are tractable probabilistic models, supporting efficient and exact computation of many probabilistic inference qu...
[ "Probabilistic Circuits", "Tractable Probabilistic Models", "Parameter Regularization", "Overfitting" ]
We proposed tractable regularization techniques for Probabilistic Circuits.
7,172
2106.02264
title_snapshot
BbikqBWZTGB
NeRV: Neural Representations for Videos
https://openreview.net/forum?id=BbikqBWZTGB
[ "Hao Chen", "Bo He", "Hanyu Wang", "Yixuan Ren", "Ser-Nam Lim", "Abhinav Shrivastava" ]
Poster
null
We propose a novel neural representation for videos (NeRV) which encodes videos in neural networks. Unlike conventional representations that treat videos as frame sequences, we represent videos as neural networks taking frame index as input. Given a frame index, NeRV outputs the corresponding RGB image. Video encoding...
[ "neural representation", "implicit representation", "video compression", "video denoising" ]
null
7,169
2110.13903
title_snapshot
apK65PUH0l9
Detecting Errors and Estimating Accuracy on Unlabeled Data with Self-training Ensembles
https://openreview.net/forum?id=apK65PUH0l9
[ "Jiefeng Chen", "Frederick Liu", "Besim Avci", "Xi Wu", "Yingyu Liang", "Somesh Jha" ]
Poster
null
When a deep learning model is deployed in the wild, it can encounter test data drawn from distributions different from the training data distribution and suffer drop in performance. For safe deployment, it is essential to estimate the accuracy of the pre-trained model on the test data. However, the labels for the test ...
[ "unsupervised accuracy estimation", "error detection", "self-training ensembles" ]
Propose a principled and practically effective framework for unsupervised accuracy estimation and error detection tasks with theoretical analysis and state-of-the-art performance
7,162
2106.15728
title_snapshot
TFEFvU0ZV6Q
Baby Intuitions Benchmark (BIB): Discerning the goals, preferences, and actions of others
https://openreview.net/forum?id=TFEFvU0ZV6Q
[ "Kanishk Gandhi", "Gala Stojnic", "Brenden M. Lake", "Moira Rose Dillon" ]
Poster
null
To achieve human-like common sense about everyday life, machine learning systems must understand and reason about the goals, preferences, and actions of other agents in the environment. By the end of their first year of life, human infants intuitively achieve such common sense, and these cognitive achievements lay the ...
[ "Cognitive Psychology", "Common Sense", "Reasoning", "Cognitive Development", "Intuitive Psychology" ]
We present the Baby Intuitions Benchmark (BIB) that challenges machines to understand and reason about the goals, preferences, and actions of other agents.
7,160
2102.11938
title_snapshot
9SD2Rb3NiWu
A Compositional Atlas of Tractable Circuit Operations for Probabilistic Inference
https://openreview.net/forum?id=9SD2Rb3NiWu
[ "antonio vergari", "YooJung Choi", "Anji Liu", "Stefano Teso", "Guy Van den Broeck" ]
Oral
null
Circuit representations are becoming the lingua franca to express and reason about tractable generative and discriminative models. In this paper, we show how complex inference scenarios for these models that commonly arise in machine learning---from computing the expectations of decision tree ensembles to information-...
[ "probabilistic reasoning", "tractable inference", "probabilistic circuits", "sum-product networks" ]
We systematically characterize a tractable model class for an inference scenario by building a modular pipeline of atomic operations and thus distilling an efficient algorithm for it
7,157
2102.06137
title_judge
vAMh-dcNMcR
Consistent Non-Parametric Methods for Maximizing Robustness
https://openreview.net/forum?id=vAMh-dcNMcR
[ "Robi Bhattacharjee", "Kamalika Chaudhuri" ]
Poster
null
Learning classifiers that are robust to adversarial examples has received a great deal of recent attention. A major drawback of the standard robust learning framework is the imposition of an artificial robustness radius $r$ that applies to all inputs, and ignores the fact that data may be highly heterogeneous. In parti...
[ "non-parametric classifiers", "adversarial examples", "robustness", "large sample limit" ]
We propose a new notion of adaptive robustness and examine conditions for non-parametric methods to converge in this setting.
7,156
2102.09086
title_snapshot
NEQYGJr1qL3
Scalable Neural Data Server: A Data Recommender for Transfer Learning
https://openreview.net/forum?id=NEQYGJr1qL3
[ "Tianshi Cao", "Sasha Doubov", "David Acuna", "Sanja Fidler" ]
Poster
null
Absence of large-scale labeled data in the practitioner's target domain can be a bottleneck to applying machine learning algorithms in practice. Transfer learning is a popular strategy for leveraging additional data to improve the downstream performance, but finding the most relevant data to transfer from can be challe...
[ "Transfer Learning", "Computer Vision", "Data Recommendation" ]
We present a data recommendation system for transfer learning that scales to arbitrary number of data sources.
7,153
2206.09386
title_snapshot
k505ekjMzww
Residual Pathway Priors for Soft Equivariance Constraints
https://openreview.net/forum?id=k505ekjMzww
[ "Marc Anton Finzi", "Gregory Benton", "Andrew Gordon Wilson" ]
Poster
null
Models such as convolutional neural networks restrict the hypothesis space to a set of functions satisfying equivariance constraints, and improve generalization in problems by capturing relevant symmetries. However, symmetries are often only partially respected, preventing models with restriction biases from fitting th...
[ "symmetry", "equivariance", "group equivariance", "priors" ]
We introduce Residual Pathway Priors for converting hard architectural constraints to soft priors, enabling exploitation of approximate symmetries.
7,151
2112.01388
title_snapshot
Jhp38rtUTV
UCB-based Algorithms for Multinomial Logistic Regression Bandits
https://openreview.net/forum?id=Jhp38rtUTV
[ "Sanae Amani", "Christos Thrampoulidis" ]
Poster
null
Out of the rich family of generalized linear bandits, perhaps the most well studied ones are logistic bandits that are used in problems with binary rewards: for instance, when the learner aims to maximize the profit over a user that can select one of two possible outcomes (e.g., `click' vs `no-click'). Despite remarkab...
[ "Generalized Linear Bandits", "Logistic Bandits", "Multinomial Logit (MNL)", "Upper Confidence Bound" ]
We study problems with more than possible outcomes selected by the user and use multinomial logit to model the probability of each possible outcome. We propose an algorithm with sublinear regret with small dependency on problem-dependent constants.
7,149
2103.11489
title_snapshot
LAwuz_L9U9j
Representation Learning for Event-based Visuomotor Policies
https://openreview.net/forum?id=LAwuz_L9U9j
[ "Sai Vemprala", "Sami Mian", "Ashish Kapoor" ]
Spotlight
null
Event-based cameras are dynamic vision sensors that provide asynchronous measurements of changes in per-pixel brightness at a microsecond level. This makes them significantly faster than conventional frame-based cameras, and an appealing choice for high-speed robot navigation. While an interesting sensor modality, this...
[ "Event cameras", "Representation Learning", "Reinforcement Learning", "Variational Autoencoder" ]
We present methods for representation learning and reinforcement learning directly from asynchronous event camera, and demonstrate advantages over frame-based techniques using an obstacle avoidance task.
7,148
2103.00806
title_snapshot
4S4nbt-rD6
Bridging the Gap Between Practice and PAC-Bayes Theory in Few-Shot Meta-Learning
https://openreview.net/forum?id=4S4nbt-rD6
[ "Nan Ding", "Xi Chen", "Tomer Levinboim", "Sebastian Goodman", "Radu Soricut" ]
Poster
null
Despite recent advances in its theoretical understanding, there still remains a significant gap in the ability of existing PAC-Bayesian theories on meta-learning to explain performance improvements in the few-shot learning setting, where the number of training examples in the target tasks is severely limited. This gap ...
[ "meta-learning", "few-shot learning", "PAC-Bayesian theory" ]
Improve PAC-Bayesian bounds on few-shot meta-learning.
7,145
2105.14099
title_snapshot
hJOLFJIJ_zy
Dense Keypoints via Multiview Supervision
https://openreview.net/forum?id=hJOLFJIJ_zy
[ "Zhixuan Yu", "Haozheng Yu", "Long Sha", "Sujoy Ganguly", "Hyun Soo Park" ]
Spotlight
null
This paper presents a new end-to-end semi-supervised framework to learn a dense keypoint detector using unlabeled multiview images. A key challenge lies in finding the exact correspondences between the dense keypoints in multiple views since the inverse of the keypoint mapping can be neither analytically derived nor dif...
[ "Dense keypoint estimation", "Multiview supervision", "Dense epipolar geometry", "Semi-supervised learning" ]
null
7,142
null
null
lzZX7E713nJ
Equivariant Manifold Flows
https://openreview.net/forum?id=lzZX7E713nJ
[ "Isay Katsman", "Aaron Lou", "Derek Lim", "Qingxuan Jiang", "Ser-Nam Lim", "Christopher De Sa" ]
Poster
null
Tractably modelling distributions over manifolds has long been an important goal in the natural sciences. Recent work has focused on developing general machine learning models to learn such distributions. However, for many applications these distributions must respect manifold symmetries—a trait which most previous mod...
[ "manifold", "normalizing flow", "equivariant", "invariant" ]
We construct manifold normalizing flows which are equivariant to isometric actions.
7,139
2107.08596
title_snapshot
t-7Jx48oaG
Analyzing the Generalization Capability of SGLD Using Properties of Gaussian Channels
https://openreview.net/forum?id=t-7Jx48oaG
[ "Hao Wang", "Yizhe Huang", "Rui Gao", "Flavio Calmon" ]
Poster
null
Optimization is a key component for training machine learning models and has a strong impact on their generalization. In this paper, we consider a particular optimization method---the stochastic gradient Langevin dynamics (SGLD) algorithm---and investigate the generalization of models trained by SGLD. We derive a new g...
[ "Information theory", "statistical learning theory" ]
null
7,137
null
null
DWvcqoRAQP8
Safe Policy Optimization with Local Generalized Linear Function Approximations
https://openreview.net/forum?id=DWvcqoRAQP8
[ "Akifumi Wachi", "Yunyue Wei", "Yanan Sui" ]
Poster
null
Safe exploration is a key to applying reinforcement learning (RL) in safety-critical systems. Existing safe exploration methods guaranteed safety under the assumption of regularity, and it has been difficult to apply them to large-scale real problems. We propose a novel algorithm, SPO-LF, that optimizes an agent's poli...
[ "Safe Reinforcement Learning", "Constrained Markov Decision Process" ]
Formulate a safe reinforcement learning problem where features are locally available upon observation, and propose an algorithm with theoretical guarantee on optimality and safety, which can be applied to large-scale problems.
7,133
2111.04894
title_snapshot
VD3TMzyxKK
Probabilistic Forecasting: A Level-Set Approach
https://openreview.net/forum?id=VD3TMzyxKK
[ "Hilaf Hasson", "Bernie Wang", "Tim Januschowski", "Jan Gasthaus" ]
Poster
null
Large-scale time series panels have become ubiquitous over the last years in areas such as retail, operational metrics, IoT, and medical domain (to name only a few). This has resulted in a need for forecasting techniques that effectively leverage all available data by learning across all time series in each panel. Amon...
[ "probabilistic forecasting", "quantile regression trees", "consistency" ]
null
7,132
null
null
SBNs7EULzqq
Non-Asymptotic Analysis for Two Time-scale TDC with General Smooth Function Approximation
https://openreview.net/forum?id=SBNs7EULzqq
[ "Yue Wang", "Shaofeng Zou", "Yi Zhou" ]
Poster
null
Temporal-difference learning with gradient correction (TDC) is a two time-scale algorithm for policy evaluation in reinforcement learning. This algorithm was initially proposed with linear function approximation, and was later extended to the one with general smooth function approximation. The asymptotic convergence fo...
[ "TDC", "finite-sample analysis", "tracking error bound", "Markovian noise", "non-convex" ]
This paper provides a non-asymptotic analysis for the TDC algorithm with general smooth function approximation.
7,131
2104.02836
title_snapshot
luWTh5Q63e
Predicting Event Memorability from Contextual Visual Semantics
https://openreview.net/forum?id=luWTh5Q63e
[ "Qianli Xu", "Fen Fang", "Ana Garcia del Molino", "Vigneshwaran Subbaraju", "Joo Hwee Lim" ]
Poster
null
Episodic event memory is a key component of human cognition. Predicting event memorability,i.e., to what extent an event is recalled, is a tough challenge in memory research and has profound implications for artificial intelligence. In this study, we investigate factors that affect event memorability according to a cue...
[ "Event memory", "image memorability", "visual semantics", "episodic memory", "lifelog" ]
A new dataset and baseline model for predicting event memorability from visual information and its context
7,129
null
null
MtvKv_BDVV
ATISS: Autoregressive Transformers for Indoor Scene Synthesis
https://openreview.net/forum?id=MtvKv_BDVV
[ "Despoina Paschalidou", "Amlan Kar", "Maria Shugrina", "Karsten Kreis", "Andreas Geiger", "Sanja Fidler" ]
Poster
null
The ability to synthesize realistic and diverse indoor furniture layouts automatically or based on partial input, unlocks many applications, from better interactive 3D tools to data synthesis for training and simulation. In this paper, we present ATISS, a novel autoregressive transformer architecture for creating diver...
[ "Indoor Scene Synthesis", "Layout Generation", "Autoregressive Set Generation", "Generative Models" ]
We propose an autoregressive transformer architecture for indoor scene synthesis that generates room layouts as unordered sets of objects and allows for a variety of interactive applications with versatile user input.
7,124
2110.03675
title_snapshot
RYcgfqmAOHh
Deep Learning with Label Differential Privacy
https://openreview.net/forum?id=RYcgfqmAOHh
[ "Badih Ghazi", "Noah Golowich", "Ravi Kumar", "Pasin Manurangsi", "Chiyuan Zhang" ]
Poster
null
The Randomized Response (RR) algorithm is a classical technique to improve robustness in survey aggregation, and has been widely adopted in applications with differential privacy guarantees. We propose a novel algorithm, Randomized Response with Prior (RRWithPrior), which can provide more accurate results while maintai...
[ "differential privacy", "label differential privacy", "randomized response", "deep learning", "self-supervised learning" ]
null
7,123
2102.06062
title_snapshot
Tsp2PL7-GQ
Can You Learn an Algorithm? Generalizing from Easy to Hard Problems with Recurrent Networks
https://openreview.net/forum?id=Tsp2PL7-GQ
[ "Avi Schwarzschild", "Eitan Borgnia", "Arjun Gupta", "Furong Huang", "Uzi Vishkin", "Micah Goldblum", "Tom Goldstein" ]
Poster
null
Deep neural networks are powerful machines for visual pattern recognition, but reasoning tasks that are easy for humans may still be difficult for neural models. Humans possess the ability to extrapolate reasoning strategies learned on simple problems to solve harder examples, often by thinking for longer. For example,...
[ "Deep learning", "algorithms", "generalization", "recurrent networks", "prefix sums", "mazes", "chess" ]
Recurrent netowrks can learn processes that can generalize from easy training data to harder examples at test time by iterating more times.
7,113
2106.04537
title_snapshot
bDHBNVtB9XA
Learning Optimal Predictive Checklists
https://openreview.net/forum?id=bDHBNVtB9XA
[ "Haoran Zhang", "Quaid Morris", "Berk Ustun", "Marzyeh Ghassemi" ]
Poster
null
Checklists are simple decision aids that are often used to promote safety and reliability in clinical applications. In this paper, we present a method to learn checklists for clinical decision support. We represent predictive checklists as discrete linear classifiers with binary features and unit weights. We then learn...
[ "healthcare", "interpretability", "fairness", "integer programming", "discrete optimization", "classification" ]
We present an integer programming method to learn optimal checklists for classification tasks.
7,107
2112.01020
title_snapshot
KBnXrODoBW
Pay Attention to MLPs
https://openreview.net/forum?id=KBnXrODoBW
[ "Hanxiao Liu", "Zihang Dai", "David So", "Quoc V Le" ]
Poster
null
Transformers have become one of the most important architectural innovations in deep learning and have enabled many breakthroughs over the past few years. Here we propose a simple network architecture, gMLP, based solely on MLPs with gating, and show that it can perform as well as Transformers in key language and visio...
[ "Transformer", "Attention", "MLP" ]
A simple variant of MLP that works well for key applications that Transformers are good at: BERT for NLP and ViT for vision
7,105
2105.08050
title_snapshot
sxjpM-kvVv_
Center Smoothing: Certified Robustness for Networks with Structured Outputs
https://openreview.net/forum?id=sxjpM-kvVv_
[ "Aounon Kumar", "Tom Goldstein" ]
Poster
null
The study of provable adversarial robustness has mostly been limited to classification tasks and models with one-dimensional real-valued outputs. We extend the scope of certifiable robustness to problems with more general and structured outputs like sets, images, language, etc. We model the output space as a metric spa...
[ "Adversarial Robustness", "Certified Robustness", "Randomized Smoothing", "Structured Outputs" ]
null
7,104
2102.09701
title_snapshot
tJ_CO8orSI
Adjusting for Autocorrelated Errors in Neural Networks for Time Series
https://openreview.net/forum?id=tJ_CO8orSI
[ "Fan-Keng Sun", "Chris Lang", "Duane S Boning" ]
Poster
null
An increasing body of research focuses on using neural networks to model time series. A common assumption in training neural networks via maximum likelihood estimation on time series is that the errors across time steps are uncorrelated. However, errors are actually autocorrelated in many cases due to the temporality o...
[ "time series forecasting", "autocorrelated errors" ]
Learning autocorrelation coefficient jointly with model parameters to adjust the autocorrelated errors in neural networks for time series.
7,086
2101.12578
title_snapshot
a2Gr9gNFD-J
Characterizing possible failure modes in physics-informed neural networks
https://openreview.net/forum?id=a2Gr9gNFD-J
[ "Aditi Krishnapriyan", "Amir Gholami", "Shandian Zhe", "Robert Kirby", "Michael W. Mahoney" ]
Poster
null
Recent work in scientific machine learning has developed so-called physics-informed neural network (PINN) models. The typical approach is to incorporate physical domain knowledge as soft constraints on an empirical loss function and use existing machine learning methodologies to train the model. We demonstrate that, wh...
[ "scientific machine learning", "physics-informed neural networks", "regularization", "constrained optimization", "unconstrained optimization" ]
We characterize the challenges associated with incorporating fundamental physical laws into the machine learning process ("physics-informed neural networks"), and devise strategies to overcome their failure modes by changing the learning paradigm.
7,081
2109.01050
title_snapshot
w5fW0TNWPyc
Machine Learning for Variance Reduction in Online Experiments
https://openreview.net/forum?id=w5fW0TNWPyc
[ "Yongyi Guo", "Dominic Coey", "Mikael Konutgan", "Wenting Li", "Chris Schoener", "Matt Goldman" ]
Poster
null
We consider the problem of variance reduction in randomized controlled trials, through the use of covariates correlated with the outcome but independent of the treatment. We propose a machine learning regression-adjusted treatment effect estimator, which we call MLRATE. MLRATE uses machine learning predictors of the ou...
[ "experimentation", "variance reduction", "agnostic statistics", "debiased machine learning", "semiparametrics", "experiment splitting" ]
We show how to use supervised ML methods to substantially increase precision in experimental causal inference.
7,079
2106.07263
title_snapshot
DZKsFQyDB9
PatchGame: Learning to Signal Mid-level Patches in Referential Games
https://openreview.net/forum?id=DZKsFQyDB9
[ "Kamal Gupta", "Gowthami Somepalli", "Anubhav Gupta", "Vinoj Jayasundara", "Matthias Zwicker", "Abhinav Shrivastava" ]
Poster
null
We study a referential game (a type of signaling game) where two agents communicate with each other via a discrete bottleneck to achieve a common goal. In our referential game, the goal of the speaker is to compose a message or a symbolic representation of "important" image patches, while the task for the listener is t...
[ "emergent language", "referential games", "self-supervised learning", "mid-level patches" ]
Emergent communication via mid-level patches in a referential game played on a large-scale image dataset
7,078
2111.01785
title_snapshot
x2rdRAx3QF
Self-Consistent Models and Values
https://openreview.net/forum?id=x2rdRAx3QF
[ "Gregory Farquhar", "Kate Baumli", "Zita Marinho", "Angelos Filos", "Matteo Hessel", "Hado van Hasselt", "David Silver" ]
Poster
null
Learned models of the environment provide reinforcement learning (RL) agents with flexible ways of making predictions about the environment. Models enable planning, i.e. using more computation to improve value functions or policies, without requiring additional environment interactions. In this work, we investigate a w...
[ "reinforcement learning", "model-based reinforcement learning", "planning", "value equivalence" ]
Maybe we should train models and value functions to be jointly self-consistent.
7,070
2110.12840
title_snapshot
503UwCYEe5
Understanding How Encoder-Decoder Architectures Attend
https://openreview.net/forum?id=503UwCYEe5
[ "Kyle Aitken", "Vinay Venkatesh Ramasesh", "Yuan Cao", "Niru Maheswaranathan" ]
Poster
null
Encoder-decoder networks with attention have proven to be a powerful way to solve many sequence-to-sequence tasks. In these networks, attention aligns encoder and decoder states and is often used for visualizing network behavior. However, the mechanisms used by networks to generate appropriate attention matrices are st...
[ "Attention", "NLP" ]
We investigate the dynamics behind networks trained on sequence to sequence tasks with and without attention.
7,051
2110.15253
title_snapshot
gRlsFQMo_ze
Reverse engineering learned optimizers reveals known and novel mechanisms
https://openreview.net/forum?id=gRlsFQMo_ze
[ "Niru Maheswaranathan", "David Sussillo", "Luke Metz", "Ruoxi Sun", "Jascha Sohl-Dickstein" ]
Poster
null
Learned optimizers are parametric algorithms that can themselves be trained to solve optimization problems. In contrast to baseline optimizers (such as momentum or Adam) that use simple update rules derived from theoretical principles, learned optimizers use flexible, high-dimensional, nonlinear parameterizations. Alth...
[ "optimization", "learned optimizers", "reverse engineering", "RNNs" ]
We reverse engineer learned optimizers trained on a simple tasks and show that they learn interpretable and intuitive mechanisms.
7,045
2011.02159
title_snapshot
SPrVNsXnGd
Renyi Differential Privacy of The Subsampled Shuffle Model In Distributed Learning
https://openreview.net/forum?id=SPrVNsXnGd
[ "Antonious M. Girgis", "Deepesh Data", "Suhas Diggavi" ]
Poster
null
We study privacy in a distributed learning framework, where clients collaboratively build a learning model iteratively through interactions with a server from whom we need privacy. Motivated by stochastic optimization and the federated learning (FL) paradigm, we focus on the case where a small fraction of data samples ...
[ "Differential privacy", "Renyi divergence", "distributed learning", "privacy amplification via shuffling", "privacy composition." ]
We characterize the renyi differential privacy of the sampled shuffle model to provide tighter privacy composition for DP-SGD algorithm.
7,043
2107.08763
title_snapshot