Interpretable and Efficient Counterfactual Generation for Real-Time User Interaction
Keywords: Explainable AI, Generative AI, Human-Machine interaction
Abstract: Among the various forms of post-hoc explanations for black-box models, counterfactuals stand out for their intuitiveness and effectiveness. However, longstanding challenges in counterfactual explanations involve the efficiency of the search process, the likelihood of generated instances, their interpretability, and in some cases, the validity of the explanations themselves. In this work we introduce a generative framework designed to address all of these issues. Notably, this is the first framework capable of generating interpretable counterfactual images in real-time, making it suitable for human-in-the-loop classification and decision-making. Our method leverages a disentangled regularized autoencoder to achieve two complementary goals: generating high-quality instances and promoting label disentanglement to provide full control over the decision boundary. This allows the model to sidestep expensive gradient-based optimizations by directly generating counterfactuals based on the adversarial distribution. A user study conducted on a challenging human-machine classification task demonstrates the effectiveness of the approach in improving human performance, highlighting the critical role of counterfactual explanations in achieving this advantage.
Primary Area: interpretability and explainable AI
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 11578
Loading