Distributionally Robust Optimization with Probabilistic Group
This addresses robustness issues in ML models for applications where group labeling is uncertain, offering a more flexible approach than prior methods.
The paper tackles the problem of machine learning models learning spurious correlations by proposing PG-DRO, a framework that uses probabilistic group membership for distributionally robust optimization, achieving superior performance on image classification and NLP benchmarks.
Modern machine learning models may be susceptible to learning spurious correlations that hold on average but not for the atypical group of samples. To address the problem, previous approaches minimize the empirical worst-group risk. Despite the promise, they often assume that each sample belongs to one and only one group, which does not allow expressing the uncertainty in group labeling. In this paper, we propose a novel framework PG-DRO, which explores the idea of probabilistic group membership for distributionally robust optimization. Key to our framework, we consider soft group membership instead of hard group annotations. The group probabilities can be flexibly generated using either supervised learning or zero-shot approaches. Our framework accommodates samples with group membership ambiguity, offering stronger flexibility and generality than the prior art. We comprehensively evaluate PG-DRO on both image classification and natural language processing benchmarks, establishing superior performance