Explaining Image Classifiers Using Contrastive Counterfactuals in Generative Latent Spaces
Published in Arxiv, 2022
Recommended citation: Alipour, K., Lahiri, A., Adeli, E., Salimi, B., Pazzani, M. (2022). Improving users' Explaining Image Classifiers Using Contrastive Counterfactuals in Generative Latent Spaces. https://arxiv.org/abs/2206.05257
The paper proposes a novel method to generate causal and interpretable counterfactual explanations for image classifiers using pretrained generative models, allowing for transparency in black-box algorithms without the need for re-training or conditioning, and provides examples of the method’s effectiveness in identifying how different attributes influence classifier outputs in face attribute classification.