Accounting for Unobserved Confounding in Domain Generalization
This addresses domain generalization for healthcare applications, but it is incremental as it builds on existing invariance principles with a new relaxation.
The paper tackles the problem of learning robust prediction models across multiple datasets despite unobserved confounders, by proposing a method that uses a regularization term to encourage partial equality of error derivatives, and demonstrates empirical performance on healthcare data from image, speech, and tabular modalities.
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.