Robustness Analysis of Counterfactual Explanations from Generative Models

This archive contains the source code of empirical study described in the paper "Robustness Evaluation of Counterfactual Explanations from Generative Models". Our highlights are the following: Systematic evaluation of counterfactual explanation robustness across task com-plexity. A taxonomy categorizing generative model-based counterfactual methods. High initial validity does not ensure counterfactual robustness against modelchanges. Most methods for generating counterfactual explanations account for proximitythrough a hyperparameter, disregarding true information about the model’s de-cision boundary.