用马氏距离解读神经网络线性层，让模型决策更透明。
Interpreting Neural Networks through Mahalanobis Distance
- 将神经网络线性层与马氏距离关联，提供新解释视角。
- 理论证明该方法可提升模型鲁棒性与泛化能力。
- 适合关注模型可解释性与透明度的研究者。
本文提出一个理论框架，将神经网络的线性层与马氏距离相连接，为神经网络可解释性提供了新视角。以往研究主要关注激活函数对性能的优化，而本文通过统计距离度量重新解读这些函数，拓展了神经网络研究中较少涉及的领域。该框架为构建更可解释的神经网络模型奠定了基础，对需要透明性的应用场景尤为重要。尽管本工作为纯理论研究，未包含实证数据，但所提出的基于距离的解释方法有望增强模型鲁棒性、改善泛化能力，并提供更直观的决策解释。
原文摘要 · Abstract (English)
This paper introduces a theoretical framework that connects neural network linear layers with the Mahalanobis distance, offering a new perspective on neural network interpretability. While previous studies have explored activation functions primarily for performance optimization, our work interprets these functions through statistical distance measures, a less explored area in neural network research. By establishing this connection, we provide a foundation for developing more interpretable neural network models, which is crucial for applications requiring transparency. Although this work is theoretical and does not include empirical data, the proposed distance-based interpretation has the potential to enhance model robustness, improve generalization, and provide more intuitive explanations of neural network decisions.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。