重庆理工大学学报2024，Vol.38Issue(15)：188-196,9.DOI:10.3969/j.issn.1674-8425(z).2024.08.022
神经网络的拓扑解释综述
A review on topological interpretation of neural networks
摘要
Abstract
As neural network technology finds extensive applications in critical fields such as medical diagnosis and financial risk assessment,the demand for transparency and interpretability in decision-making processes has increasingly grown.Although numerous studies have explored the interpretability of neural networks from various perspectives,current methods have yet to fully elucidate their decision mechanisms,which limits their deployment in scenarios requiring high reliability and interpretability.This paper systematically reviews the application of topological methods in neural network interpretability research,providing a detailed analysis of the strengths and limitations of these methods in revealing the inner workings of neural networks.The study specifically examines the role of topological tools in analyzing the feature space and parameter space of neural networks and summarizes the challenges and future directions faced by related research in practical applications.This review offers valuable insights for further enhancing the transparency and interpretability of neural networks.关键词
神经网络可解释性/拓扑数据分析/持续同调/Mapper算法Key words
neural network interpretability/topological data analysis/persistent homology/Mapper algorithm分类
信息技术与安全科学引用本文复制引用
何宇楠,阳蕾,王佳慧..神经网络的拓扑解释综述[J].重庆理工大学学报,2024,38(15):188-196,9.基金项目
重庆理工大学科研启动基金项目 （）
重庆市教委科学技术研究项目(KJQN202101108) （KJQN202101108）
