Robust Generalization despite Distribution Shift via Minimum Discriminating Information
arXiv:2106.04443
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.
arXiv comment23 pages, 4 figures
References in corpus (7)
- Distributionally Robust Logistic Regression
- Invariant Models for Causal Transfer Learning
- Regularization via Mass Transportation
- Learning from Logged Implicit Exploration Data
- A unified view of entropy-regularized Markov decision processes
- A Theory of Regularized Markov Decision Processes
- Distributionally Robust Optimization with Markovian Data