Diverse Prototypical Ensembles Improve Robustness to Subpopulation Shift

TL;DR

This paper introduces Diverse Prototypical Ensembles (DPEs), a novel ensemble method that improves robustness to subpopulation shifts without requiring group annotations, outperforming existing methods on multiple real-world datasets.

Contribution

The paper proposes DPEs, an ensemble of diverse classifiers with prototypical layers, to adaptively handle subpopulation shifts without relying on group labels or assumptions.

Findings

DPEs outperform state-of-the-art methods in worst-group accuracy.

DPEs demonstrate robustness across nine diverse real-world datasets.

The approach does not require group annotations or prior assumptions.

Abstract

The subpopulationtion shift, characterized by a disparity in subpopulation distributibetween theween the training and target datasets, can significantly degrade the performance of machine learning models. Current solutions to subpopulation shift involve modifying empirical risk minimization with re-weighting strategies to improve generalization. This strategy relies on assumptions about the number and nature of subpopulations and annotations on group membership, which are unavailable for many real-world datasets. Instead, we propose using an ensemble of diverse classifiers to adaptively capture risk associated with subpopulations. Given a feature extractor network, we replace its standard linear classification layer with a mixture of prototypical classifiers, where each member is trained to classify the data while focusing on different features and samples from other members. In empirical evaluation on nine real-world datasets, covering diverse domains and kinds of subpopulation shift, our method of Diverse Prototypical Ensembles (DPEs) often outperforms the prior state-of-the-art in worst-group accuracy. The code is available at https://github.com/minhto2802/dpe4subpop