Explaining Classifiers with Causal Concept Effect (CaCE)

TL;DR

This paper introduces CaCE, a causal measure for understanding neural network decisions based on human concepts, using VAEs to estimate causal effects and avoid confounding errors.

Contribution

It proposes CaCE as a causal explanation method for neural networks and demonstrates how VAEs can effectively estimate these causal effects in complex datasets.

Findings

VAE-CaCE accurately estimates true concept causal effects.

CaCE outperforms correlation-based methods in avoiding confounding.

Effective on high-dimensional image datasets.

Abstract

How can we understand classification decisions made by deep neural networks? Many existing explainability methods rely solely on correlations and fail to account for confounding, which may result in potentially misleading explanations. To overcome this problem, we define the Causal Concept Effect (CaCE) as the causal effect of (the presence or absence of) a human-interpretable concept on a deep neural net's predictions. We show that the CaCE measure can avoid errors stemming from confounding. Estimating CaCE is difficult in situations where we cannot easily simulate the do-operator. To mitigate this problem, we use a generative model, specifically a Variational AutoEncoder (VAE), to measure VAE-CaCE. In an extensive experimental analysis, we show that the VAE-CaCE is able to estimate the true concept causal effect, compared to baselines for a number of datasets including high dimensional images.