An Invariant Matching Property for Distribution Generalization under Intervened Response
This work addresses a practical limitation in causal models for distribution generalization, offering a solution for scenarios where response interventions occur, though it appears incremental as it builds on existing invariance principles.
The paper tackles the problem of distribution generalization when the response variable is intervened, which violates the standard invariance assumption. It proposes a novel invariance property using feature estimates as additional predictors, showing it enables generalization via a deterministic linear matching, with simulation results demonstrating its effectiveness across various intervention settings.
The task of distribution generalization concerns making reliable prediction of a response in unseen environments. The structural causal models are shown to be useful to model distribution changes through intervention. Motivated by the fundamental invariance principle, it is often assumed that the conditional distribution of the response given its predictors remains the same across environments. However, this assumption might be violated in practical settings when the response is intervened. In this work, we investigate a class of model with an intervened response. We identify a novel form of invariance by incorporating the estimates of certain features as additional predictors. Effectively, we show this invariance is equivalent to having a deterministic linear matching that makes the generalization possible. We provide an explicit characterization of the linear matching and present our simulation results under various intervention settings.