Diverse Prototypical Ensembles Improve Robustness to Subpopulation Shift
This addresses subpopulation shift in machine learning, which degrades model performance when training and target distributions differ, offering a solution without needing group annotations, though it is incremental as it builds on ensemble and prototypical methods.
The paper tackles subpopulation shift by proposing Diverse Prototypical Ensembles, which replace a linear classification layer with a mixture of prototypical classifiers to adaptively capture subpopulation risks without requiring group annotations. In evaluations on nine real-world datasets, the method often outperforms prior state-of-the-art in worst-group accuracy.
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