On Counterfactual Data Augmentation Under Confounding

TL;DR

This paper analyzes the impact of confounding biases on classifiers and proposes a causal, counterfactual data augmentation method to improve generalization, validated on image datasets like MNIST and CelebA.

Contribution

It introduces a causal perspective on counterfactual data augmentation and presents a simple algorithm for generating counterfactual images to mitigate confounding biases.

Findings

Improves classifier robustness against confounding biases

Enhances generalization on MNIST variants and CelebA datasets

Complements state-of-the-art methods effectively

Abstract

Counterfactual data augmentation has recently emerged as a method to mitigate confounding biases in the training data. These biases, such as spurious correlations, arise due to various observed and unobserved confounding variables in the data generation process. In this paper, we formally analyze how confounding biases impact downstream classifiers and present a causal viewpoint to the solutions based on counterfactual data augmentation. We explore how removing confounding biases serves as a means to learn invariant features, ultimately aiding in generalization beyond the observed data distribution. Additionally, we present a straightforward yet powerful algorithm for generating counterfactual images, which effectively mitigates the influence of confounding effects on downstream classifiers. Through experiments on MNIST variants and the CelebA datasets, we demonstrate how our simple augmentation method helps existing state-of-the-art methods achieve good results.