Diffusion Models for Counterfactual Explanations

Counterfactual explanations have shown promising results as a post-hoc framework to make image classifiers more explainable. In this paper, we propose DiME, method allowing the generation of counterfactual images using recent diffusion models. By leveraging guided generative process, our proposed methodology shows how use gradients target classifier generate input instances. Further, analyze current approaches evaluate spurious correlations and extend evaluation measurements by proposing new metric: Correlation Difference. Our experimental validations show that algorithm surpasses previous state-of-the-art on 5 out 6 metrics CelebA.