Interpretable image classification based on antifactual data

Zhenyu Lu

doi:10.1016/j.patcog.2025.112772

Interpretable image classification based on antifactual data

2025-11-19

0

PRE

AI

DOI：10.1016/j.patcog.2025.112772

Original

Original request for help

Save

Abstract

En 中文

• The proposed method can simultaneously improve both the interpretability and the performance of deep neural networks in image classification. • The antifactual data is efficiently constructed to decouple the intrinsic features containing categorical semantic information from other extracted features. • The proposed Interpretable Classification Accuracy is a novel metric to reflect the interpretability by evaluating the dependence of classification results on intrinsic features.

Interpretable image classification based on antifactual data

Abstract

Journal

Organization