Publication

Redundant features can hurt robustness to distribution shift

Related publications (27)

A story of two transitions: From adhesive to abrasive wear and from ductile to brittle regime

Jean-François Molinari, Sacha Zenon Wattel

Atomistic simulations performed with a family of model potential with tunable hardness have proven to be a great tool for advancing the understanding of wear processes at the asperity level. They have been instrumental in finding a critical length scale, w ...

2024

Making Computer Vision Models Robust and Adaptive

Shuqing Teresa Yeo

Visual perception is indispensable for many real-world applications. However, perception models deployed in the real world will encounter numerous and unpredictable distribution shifts, for example, changes in geographic locations, motion blur, and adverse ...

EPFL2023

Learning Robust and Adaptive Representations: from Interactions, for Interactions

Yuejiang Liu

Interactions are ubiquitous in our world, spanning from social interactions between human individuals to physical interactions between robots and objects to mechanistic interactions among different components of an intelligent system. Despite their prevale ...

EPFL2023

Federated Learning under Covariate Shifts with Generalization Guarantees

Volkan Cevher, Thomas Michaelsen Pethick, Grigorios Chrysos, Fanghui Liu

This paper addresses intra-client and inter-client covariate shifts in federated learning (FL) with a focus on the overall generalization performance. To handle covariate shifts, we formulate a new global model training paradigm and propose Federated Impor ...

2023

Rapid Network Adaptation: Learning to Adapt Neural Networks Using Test-Time Feedback

Amir Roshan Zamir, Shuqing Teresa Yeo, Oguzhan Fatih Kar, Zahra Sodagar

We propose a method for adapting neural networks to distribution shifts at test-time. In contrast to training-time robustness mechanisms that attempt to anticipate and counter the shift, we create a closed-loop system and make use of test-time feedback sig ...

Ieee Computer Soc2023

Towards Robust Vision Transformer

Shaokai Ye, Yuan He

Recent advances on Vision Transformer (ViT) and its improved variants have shown that self-attention-based networks surpass traditional Convolutional Neural Networks (CNNs) in most vision tasks. However, existing ViTs focus on the standard accuracy and com ...

IEEE COMPUTER SOC2022

OptTTA: Learnable Test-Time Augmentation for Source-Free Medical Image Segmentation Under Domain Shift

Jean-Philippe Thiran, Guillaume Marc Georges Vray, Seyedbehzad Bozorgtabar, Devavrat Tomar

As distribution shifts are inescapable in realistic clinical scenarios due to inconsistencies in imaging protocols, scanner vendors, and across different centers, well-trained deep models incur a domain generalization problem in unseen environments. Despit ...

PMLR2022

Feature distribution learning by passive exposure

Gizay Ceylan, David Pascucci

Humans can rapidly estimate the statistical properties of groups of stimuli, including their average and variability. But recent studies of so-called Feature Distribution Learning (FDL) have shown that observers can quickly learn even more complex aspects ...

2022

Robustness and invariance properties of image classifiers

Apostolos Modas

Deep neural networks have achieved impressive results in many image classification tasks. However, since their performance is usually measured in controlled settings, it is important to ensure that their decisions remain correct when deployed in noisy envi ...

EPFL2022

System Support for Robust Distributed Learning

Arsany Hany Abdelmessih Guirguis

Machine learning (ML) applications are ubiquitous. They run in different environments such as datacenters, the cloud, and even on edge devices. Despite where they run, distributing ML training seems the only way to attain scalable, high-quality learning. B ...

EPFL2022

Continual Test-Time Domain Adaptation

Qin Wang, Olga Fink

Test-time domain adaptation aims to adapt a source pretrained model to a target domain without using any source data. Existing works mainly consider the case where the target domain is static. However, real-world machine perception systems are running in n ...

IEEE COMPUTER SOC2022

Robust Generalization despite Distribution Shift via Minimum Discriminating Information

Daniel Kuhn, Tobias Sutter, Andreas Krause

Training models that perform well under distribution shifts is a central challenge in machine learning. In this paper, we introduce a modeling framework where, in addition to training data, we have partial structural knowledge of the shifted test distribut ...

2021

Multiple testing with test statistics following heavy-tailed distributions

Zhiwen Jiang

In multiple testing problems where the components come from a mixture model of noise and true effect, we seek to first test for the existence of the non-zero components, and then identify the true alternatives under a fixed significance level

\alpha

. Two ...

EPFL2021

Robustness via Cross-Domain Ensembles

Amir Roshan Zamir, Shuqing Teresa Yeo, Oguzhan Fatih Kar

We present a method for making neural network predictions robust to shifts from the training data distribution. The proposed method is based on making predictions via a diverse set of cues (called 'middle domains') and ensembling them into one strong predi ...

IEEE2021

Fractional extreme distributions

Thomas Simon, Lotfi Boudabsa

We consider three classes of linear differential equations on distribution functions, with a fractional order alpha is an element of [0; 1]. The integer case alpha = 1 corresponds to the three classical extreme families. In general, we show that there is a ...

2020

Impact of a companion and of chromospheric emission on the shape of chromosome maps for globular clusters

Carmela Lardo

Context. Globular clusters (GCs) host multiple populations of stars that are well-separated in a photometric diagram - the chromosome map - built from specific Hubble Space Telescope (HST) filters. Stars from different populations feature at various locati ...

EDP SCIENCES S A2020

Scalable Stochastic Optimization: Scenario Reduction with Guarantees

Kilian Schindler

Stochastic optimization is a popular modeling paradigm for decision-making under uncertainty and has a wide spectrum of applications in management science, economics and engineering. However, the stochastic optimization models one faces in practice are int ...

EPFL2020

On Vacuous and Non-Vacuous Generalization Bounds for Deep Neural Networks

Konstantinos Pitas

Deep neural networks have been empirically successful in a variety of tasks, however their theoretical understanding is still poor. In particular, modern deep neural networks have many more parameters than training data. Thus, in principle they should over ...

EPFL2020

From Data to Decisions: Distributionally Robust Optimization is Optimal

Peyman Mohajerin Esfahani, Daniel Kuhn

We study stochastic programs where the decision-maker cannot observe the distribution of the exogenous uncertainties but has access to a finite set of independent samples from this distribution. In this setting, the goal is to find a procedure that transfo ...

2021

The probability of intransitivity in dice and close elections

Jan Hazla

We study the phenomenon of intransitivity in models of dice and voting. First, we follow a recent thread of research for n-sided dice with pairwise ordering induced by the probability, relative to 1/2, that a throw from one die is higher than the other. We ...

2020

Related publications (27)

\alpha

. Two ...

EPFL2021