Variational Counterfactual Prediction under Runtime Domain Corruption

TL;DR

This paper introduces VEGAN, a novel variational causal effect model with adversarial domain adaptation, designed to improve counterfactual prediction accuracy under runtime domain corruption caused by distribution shifts and inaccessible variables.

Contribution

The paper proposes VEGAN, a two-stage adversarial domain adaptation framework that effectively handles runtime domain corruption in counterfactual prediction tasks, a challenge not addressed by prior methods.

Findings

VEGAN outperforms existing baselines on benchmark datasets.

The model effectively reduces distribution disparity between treated and control groups.

VEGAN maintains high prediction accuracy despite variable inaccessibility and distribution shifts.

Abstract

To date, various neural methods have been proposed for causal effect estimation based on observational data, where a default assumption is the same distribution and availability of variables at both training and inference (i.e., runtime) stages. However, distribution shift (i.e., domain shift) could happen during runtime, and bigger challenges arise from the impaired accessibility of variables. This is commonly caused by increasing privacy and ethical concerns, which can make arbitrary variables unavailable in the entire runtime data and imputation impractical. We term the co-occurrence of domain shift and inaccessible variables runtime domain corruption, which seriously impairs the generalizability of a trained counterfactual predictor. To counter runtime domain corruption, we subsume counterfactual prediction under the notion of domain adaptation. Specifically, we upper-bound the error w.r.t. the target domain (i.e., runtime covariates) by the sum of source domain error and inter-domain distribution distance. In addition, we build an adversarially unified variational causal effect model, named VEGAN, with a novel two-stage adversarial domain adaptation scheme to reduce the latent distribution disparity between treated and control groups first, and between training and runtime variables afterwards. We demonstrate that VEGAN outperforms other state-of-the-art baselines on individual-level treatment effect estimation in the presence of runtime domain corruption on benchmark datasets.