Robust Generalization despite Distribution Shift via Minimum Discriminating Information

TL;DR

This paper proposes a framework combining prior structural knowledge and distributionally robust optimization to improve model generalization under distribution shifts, with applications in biased data classification and off-policy evaluation.

Contribution

It introduces a novel approach using minimum discriminating information and large deviation bounds to handle distribution shifts with limited data.

Findings

Explicit generalization bounds derived for shifted distributions

Effective in biased data classification scenarios

Applicable to off-policy evaluation in MDPs

Abstract

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 distribution. We employ the principle of minimum discriminating information to embed the available prior knowledge, and use distributionally robust optimization to account for uncertainty due to the limited samples. By leveraging large deviation results, we obtain explicit generalization bounds with respect to the unknown shifted distribution. Lastly, we demonstrate the versatility of our framework by demonstrating it on two rather distinct applications: (1) training classifiers on systematically biased data and (2) off-policy evaluation in Markov Decision Processes.