Learning robust models that adapt to shifting data distributions

Learning the Robustness Mechanism with Bilevel Optimization

Machine Learning

Summary

Machine learning models often struggle when the data they see changes in unexpected ways. This paper introduces a way to teach models to be more robust by automatically learning how to handle these changes using a special optimization process. Instead of guessing how to make the model tough against tricky data, the method learns the best settings from extra data. The authors show that this approach works well both in theory and practice, especially when different groups of data change in various ways at the same time.

What this means in practice

Authors

Yiyang Shen, Qihang Lin, Weiran Wang

Abstract

We propose a distributionally robust learning framework where parameters defining the robustness mechanism are learned from held-out data instead of extensively tuned. Using bilevel optimization with both upper and lower level minimax problems, we create two instances of our framework to tackle setups with and without group labels in the training set. Theoretically, we provide sample complexity analysis for our robustness mechanism learning paradigm, showing that it achieves generalization guarantees comparable to exhaustive grid search while being more computationally efficient. Empirically, we evaluate our framework under a challenging setup when both intra-group and inter-group test distribution shifts occur at the same time, thereby demonstrating the efficacy and scalability of our method.