Causally Inspired Regularization Enables Domain General Representations

arXiv:2404.16277v1 Announce Type: new
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 …

abstract arxiv causal cs.lg data domain domains feature general graph graphs identify input-output literature predictive process regularization set standard stat.ml type