Toward practical and plausible counterfactual explanation through latent adjustment in disentangled space

Seung-Hyup Na, Woo-Jeoung Nam, Seong-Whan Lee

Abstract: Highlights•This method generates practical and plausible counterfactuals for all classes.•Closeness to input and to counterfactual data manifold are in trade-offs.•Counterfactuals are generated through minimal perturbation of class information.•Semantic information is linearly changeable in a disentangled space.•Novel latent adjustment algorithms in this space are proposed.