##基于信息瓶颈的深度学习可解释性增强
Jincheng Zhang
阅读操作
确认中在文库中上传 PDF 后可生成中文音频讲解。
摘要与影响
This paper explores a novel approach to enhancing the interpretability of deep learning models by leveraging the concept of information bottlenecks. Traditional methods of explaining deep learning models often struggle to pinpoint the specific regions or features driving a model's decision-making process. This research introduces a framework that identifies crucial information bottlenecks within the model's architecture during forward propagation. By analyzing the flow of information through these bottlenecks – characterized by significant reductions in information dimensionality – we aim to provide a more granular and understandable explanation of the model's internal logic. The core idea is that these bottlenecks represent points where the model is forced to compress information, thereby highlighting the most salient features and pathways contributing to the final prediction. This work provides a theoretical foundation and a methodological approach for interpreting deep learning models based on the principles of information theory and the concept of information bottlenecks. The results demonstrate the potential of this approach to improve model transparency and trustworthiness.
逐年被引趋势
暂无年度引用数据
关键指标
同类平均 = 1
同领域 · 同年份 · 同类型
Google Scholar 与 OpenAlex 的被引统计范围不同,数值存在差异属正常。
AI 辅助阅读
依据:摘要
可就本文提问;依据不足时会说明。
学术脉络
学科主题
计算机 / AIExplainable Artificial Intelligence (XAI)
Advanced Neural Network Applications · Adversarial Robustness in Machine Learning