Invariant Rationalization

TL;DR

This paper introduces an invariant rationalization method that uses a game-theoretic criterion to identify input features supporting predictions across diverse environments, reducing spurious correlations and improving interpretability.

Contribution

It proposes a novel invariant rationalization criterion based on game theory, enhancing robustness and interpretability over traditional mutual information approaches.

Findings

Reduces reliance on spurious correlations

Improves generalization across environments

Aligns better with human judgments

Abstract

Selective rationalization improves neural network interpretability by identifying a small subset of input features -- the rationale -- that best explains or supports the prediction. A typical rationalization criterion, i.e. maximum mutual information (MMI), finds the rationale that maximizes the prediction performance based only on the rationale. However, MMI can be problematic because it picks up spurious correlations between the input features and the output. Instead, we introduce a game-theoretic invariant rationalization criterion where the rationales are constrained to enable the same predictor to be optimal across different environments. We show both theoretically and empirically that the proposed rationales can rule out spurious correlations, generalize better to different test scenarios, and align better with human judgments. Our data and code are available.