Russian Federation
Russian Federation
Russian Federation
Modern deep neural networks demonstrate outstanding results in various fields, from computer vision to natural language processing. However, their complex internal structure often makes the decision-making process opaque, which creates a "black box" problem. This is a major obstacle to the implementation of such systems in critical areas such as medicine, autonomous driving, and law, which require not only high accuracy, but also an understanding of decision-making logic, trust, and the ability to audit. In response to this challenge, Explicable AI (XAI) is actively developing, which aims to create methods for interpreting and explaining artificial intelligence solutions. This article provides an overview and comparative analysis of key XAI methods, including LIME, SHAP, and Grad-CAM. Their theoretical foundations, fields of application and practical significance for ensuring the reliability and security of intelligent systems are considered.
Explicable AI, interpretability, deep learning, black box, LIME, SHAP, Grad-CAM, computer vision, trust in AI
1. Ob'edinenie glubokogo obucheniya i ob'yasnimogo II dlya neinvazivnogo prognozirovaniya mutaciy EGFR i KRAS v NSCLC: novyy radiogenomnyy podhod / F. Shariati, V. Pavlov, S. Fedyashina, N. Serebrennikov // V Mezhdunarodnaya konferenciya po neyronnym setyam i neyrotehnologiyam (NeuroNT'2024) : Sbornik dokladov konferencii, Sankt-Peterburg, 20 iyunya 2024 goda. – Sankt-Peterburg: Federal'noe gosudarstvennoe avtonomnoe obrazovatel'noe uchrezhdenie vysshego obrazovaniya Sankt-Peterburgskiy gosudarstvennyy elektrotehnicheskiy universitet LETI im. V.I. Ul'yanova (Lenina), 2024. – S. 41-44. – EDN RWFMZV.
2. Argumentative XAI: A Survey / K. Cyras, A. Rago, E. Albini [et al.] // IJCAI International Joint Conference on Artificial Intelligence : 30, Virtual, Online, 19–27 avgusta 2021 goda. – Virtual, Online, 2021. – P. 4392-4399. – EDN RDGNIP. DOI: https://doi.org/10.24963/ijcai.2021/600
3. Caroprese, L. Argumentation approaches for explanaible AI in medical informatics / L. Caroprese, E. Vocaturo, E. Zumpano // Intelligent Systems with Applications. – 2022. – Vol. 16. – P. 200109. – DOIhttps://doi.org/10.1016/j.iswa.2022.200109. – EDN TQUGES.
