理解(interpret)表示用可被认知(understandable)的说法去解释(explain)或呈现(present)。在机器学习的场景中,可解释性(interpretability)就表示模型能够使用人类可认知的说法进行解释和呈现。[Finale Doshi-Velez]
机器学习模型被许多人称为“黑盒”。这意味着虽然我们可以从中获得准确的预测,但我们无法清楚地解释或识别这些预测背后的逻辑。但是我们如何从模型中提取重要的见解呢?要记住哪些事项以及我们需要实现哪些功能或工具?这些是在提出模型可解释性问题时会想到的重要问题。
本文整理了可解释机器学习相关领域最新的论文,书籍、资源、博客等,分享给需要朋友。
资源整理自网络,源地址:https://github.com/wangyongjie-ntu/Awesome-explainable-AI
所有资源下载地址,见源地址。
本资源含了近年来热门的可解释人工智能(XAI)的前沿研究。从下图我们可以看到可解释/可解释AI的趋势。关于这个主题的出版物正在蓬勃发展。
下图展示了XAI的几个用例。在这里,根据这个数字将出版物分成几个类别。
研究性论文
The elephant in the interpretability room: Why use attention as explanation when we have saliency methods, EMNLP Workshop 2020
Explainable Machine Learning in Deployment, FAT 2020
A brief survey of visualization methods for deep learning models from the perspective of Explainable AI, Information Visualization 2020
Explaining Explanations in AI, ACM FAT 2019
Machine learning interpretability: A survey on methods and metrics, Electronics, 2019
A Survey on Explainable Artificial Intelligence (XAI): Towards Medical XAI, IEEE TNNLS 2020
Interpretable machine learning: definitions, methods, and applications, Arxiv preprint 2019
Visual Analytics in Deep Learning: An Interrogative Survey for the Next Frontiers, IEEE Transactions on Visualization and Computer Graphics, 2019
Explainable Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI, Information Fusion, 2019
Evaluating Explanation Without Ground Truth in Interpretable Machine Learning, Arxiv preprint 2019
A survey of methods for explaining black box models, ACM Computing Surveys, 2018
Explaining Explanations: An O