Causally Inspired Regularization Enables Domain General Representations

TL;DR

This paper introduces a causally inspired regularization framework that improves the identification of domain-general features, leading to better transfer performance across diverse domains without prior knowledge of spurious features.

Contribution

The paper proposes a novel regularization method based on causal graph principles to identify domain-general features without prior spurious feature knowledge.

Findings

Effective on synthetic and real-world datasets

Outperforms state-of-the-art methods in transfer accuracy

Applicable without prior spurious feature information

Abstract

Given a causal graph representing the data-generating process shared across different domains/distributions, enforcing sufficient graph-implied conditional independencies can identify domain-general (non-spurious) feature representations. For the standard input-output predictive setting, we categorize the set of graphs considered in the literature into two distinct groups: (i) those in which the empirical risk minimizer across training domains gives domain-general representations and (ii) those where it does not. For the latter case (ii), we propose a novel framework with regularizations, which we demonstrate are sufficient for identifying domain-general feature representations without a priori knowledge (or proxies) of the spurious features. Empirically, our proposed method is effective for both (semi) synthetic and real-world data, outperforming other state-of-the-art methods in average and worst-domain transfer accuracy.