Extracting Rules from Machine Learning Models in Angluin’s Style
Abstract
We first study Angluin’s exact learning framework and Valiant's probably approximately correct (PAC) learning framework, from computational learning theory. We discuss how results from the exact framework can be transferred to the PAC framework. Then, we see an overview of recent approaches to extract simpler abstractions of complex neural networks in Angluin's style. The aim of constructing such abstractions is to obtain high level information from machine learning models, which can be useful to interpret their behavior, detect harmful biases, among others. Finally, we discuss algorithms for learning logical theories in Angluin’s framework, in particular, those for learning rules in Horn logic.
Associate Professor Ana Ozaki, Department of Informatics, University of Oslo
Ozaki has background in computer science, specialized in the field of knowledge representation and reasoning and in machine learning theory. Her research focuses on learning logical theories formulated in description logic and related formalisms for knowledge representation. She serves the AI community as a program committee member for NORA and as a steering committee member of KR and DL. Ozaki is a member of the editorial boards of the Journal of Machine Learning Research and the Journal of Web Semantics. Ozaki has recently worked as Program Committee Chair for the 7th International Joint Conference on Rules and Reasoning and the 36th International Description Logic Workshop.
Relevant publications and resources
Sophie Blum, Raoul Koudijs, Ana Ozaki, Samia Touileb, "Learning Horn envelopes via queries from language models", International Journal of Approximate Reasoning, 2023.
DoI: https://www.sciencedirect.com/science/article/pii/S0888613X23001573
Practical
The Tuesdays seminars series are devoted to various topics relevant for the Integreat´s research focus. Presenters from the Integreat community and beyond have 40 minutes to present, followed by a group discussion.
Seminars are open for attendance for everybody.