Concise and Logically Consistent Conformal Sets for Neuro-Symbolic Concept-Based Models

In three linesCOCOCO, a post-hoc framework, integrates Conformal Prediction into neuro-symbolic concept-based models (NeSy-CBMs) to improve reliability. It conformalizes concepts and labels jointly via a single deduction-abduction revision step, ensuring consistency, coverage, and conciseness with distribution-free guarantees. Tested on 8 datasets.

Summary generated by Claude — human-verified