Accounting for Unobserved Confounding in Domain Generalization

TL;DR

This paper proposes a novel method to improve domain generalization by accounting for unobserved confounders through a relaxed invariance principle and distributionally robust optimization, demonstrated on healthcare data across multiple modalities.

Contribution

It introduces a new invariance property for causal solutions under unobserved confounding and connects it with a distributionally robust optimization framework.

Findings

Improved robustness on healthcare datasets across multiple modalities.

Effective handling of unobserved confounders in domain generalization.

Enhanced model stability and generalization performance.

Abstract

This paper investigates the problem of learning robust, generalizable prediction models from a combination of multiple datasets and qualitative assumptions about the underlying data-generating model. Part of the challenge of learning robust models lies in the influence of unobserved confounders that void many of the invariances and principles of minimum error presently used for this problem. Our approach is to define a different invariance property of causal solutions in the presence of unobserved confounders which, through a relaxation of this invariance, can be connected with an explicit distributionally robust optimization problem over a set of affine combination of data distributions. Concretely, our objective takes the form of a standard loss, plus a regularization term that encourages partial equality of error derivatives with respect to model parameters. We demonstrate the empirical performance of our approach on healthcare data from different modalities, including image, speech and tabular data.