Answers are grounded in this paper and cite their sources.
This preprint presents a recursive explainable artificial intelligence (R-XAI) approach inspired by the dual-process framework of System 1 and System 2. The proposed architecture combines a fast sub-symbolic perception layer with a slower analytical and neuro-fuzzy reasoning layer, linked through a salience-based switching mechanism. The approach is designed to provide visual, logical, and textual explanations while maintaining predictive performance and enabling verification of model decisions. The proposed methodology is experimentally evaluated on the task of automated detection and interpretation of coronary stenoses using the CADiCA-XR dataset. The reported evaluation achieved AUC = 0.92, Dice = 0.84, and a median attention-map IoU of 0.68. The work also considers meta-explainability, robustness of explanations, human-interpretable reasoning, and the application of recursive XAI architectures in high-risk decision-support systems. This publication is a preprint based on the author’s Master’s thesis and is made publicly available for scholarly communication, citation, and further development of the proposed R-XAI methodology. Русский: В препринте представлена концепция рекурсивного объяснимого искусственного интеллекта (R-XAI), основанная на взаимодействии быстрого субсимвольного и медленного аналитико-логического контуров. Методология апробирована на задаче автоматизированного выявления и объяснения коронарных стенозов.
No takes yet. Share an insight, caveat, or question.
Yuri Vladislavovich Trofimov (2026) studied this question.