学术预告

“Post-Hoc Explanation-by-Example” in Theory and in Practice

发布日期:2023-08-21 发表者:陈治国 浏览次数:




报告时间:2023/08/23  上午10:00-11:00 

报告地点:第一综合楼B316会议室

报告题目:“Post-Hoc Explanation-by-Example” in Theory and in Practice

交流摘要:

The widespread use of machine learning has raised significant concerns over the ability to trust such systems due to the “black box” nature of their inference process. In this talk I will explain our work on explaining such systems using post-hoc explanation-by-example by way of the twin-system framework. In twin-systems, an opaque artificial neural network (ANN) is explained by “twinning” it with a more interpretable case-based reasoning (CBR) system, by mapping the feature weights from the form

报告人简介:

MIT博后Eoin M. Kenny (个人主页 https://eoinkenny.github.io/

Eoin M. Kenny is an explainable AI reseracher (XAI). Previously, he did my Ph.D. at University College Dublin, Ireland. There he worked on post-hoc explanation-by-example with his supervisor Mark Keane. Prior to that, for his undergrad, he completed my Bachelor of Music degree (and Master of Arts in Musicology & Performance) at the University of Maynooth. Currently, he is at MIT researching human-friendly explanations in the Interactive Robotics Group, with a focus on contrastive explanation, and deep reinforcement learning. He has published in many of the top AI/ML conferences and Journals as first author during the past few years, including ICLR, AAAI, IJCAI, and the AI Journal. His notable contributions have been the introduction of semi-factual explanations, the first inherently intrepretable (high performing) Deep RL system, the twin-system framework for post-hoc example-based explanation, and several high quality user evaluations (designed with cognitive psychologists).

Baidu
sogou