On Quantitative Evaluations of Counterfactuals
This addresses a critical gap in explainable AI for researchers and practitioners by highlighting limitations in current evaluation methods, though it is incremental as it builds on existing work.
The paper tackles the problem of evaluating visual counterfactual examples in deep learning, finding that existing metrics often fail to distinguish good from bad counterfactuals on complex datasets and can be misled by tiny adversarial-like changes. It proposes two new metrics, the Label Variation Score and Oracle score, which are less vulnerable to such issues.
As counterfactual examples become increasingly popular for explaining decisions of deep learning models, it is essential to understand what properties quantitative evaluation metrics do capture and equally important what they do not capture. Currently, such understanding is lacking, potentially slowing down scientific progress. In this paper, we consolidate the work on evaluating visual counterfactual examples through an analysis and experiments. We find that while most metrics behave as intended for sufficiently simple datasets, some fail to tell the difference between good and bad counterfactuals when the complexity increases. We observe experimentally that metrics give good scores to tiny adversarial-like changes, wrongly identifying such changes as superior counterfactual examples. To mitigate this issue, we propose two new metrics, the Label Variation Score and the Oracle score, which are both less vulnerable to such tiny changes. We conclude that a proper quantitative evaluation of visual counterfactual examples should combine metrics to ensure that all aspects of good counterfactuals are quantified.