Background information

ReX is a causal explainability tool for image classifiers. ReX is black-box, that is, agnostic to the internal structure of the classifier. We assume that we can modify the inputs and send them to the classifier, observing the output. ReX outperforms other tools on single explanations, non-contiguous explanations (for partially obscured images), and multiple explanations.

Assumptions

ReX works on the assumption that if we can intervene on the inputs to a model and observe changes in its outputs, we can use this information to reason about the way the DNN makes its decisions.

Presentations about ReX

• Attacking your black box classifier with ReX

• Causal Explanations For Image Classifiers

Papers

1. Causal Explanations for Image Classifiers. Under review. This paper introduces the tool ReX.

2. Multiple Different Black Box Explanations for Image Classifiers. In ECAI 2025.

3. 3D ReX: Causal Explanations in 3D Neuroimaging Classification. Presented at Imageomics-AAAI-25. 3D explanations for neuroimaging.

4. Explanations for Occluded Images. In ICCV’21. This paper introduces causality for image classifier explanations. Note: the tool is called DC-Causal in this paper.

5. Explaining Image Classifiers using Statistical Fault Localization. In ECCV’20. The first paper on ReX. Note: the tool is called DeepCover in this paper.

6. Explaining Negative Classifications of AI Models in Tumor Diagnosis. In UAI 2025. This uses ReX explanations in its algorithm.