Title
Computing attribution scores for classification in explainable AI
Abstract
Attribution scores are used in XAI to quantitatively measure the strength of explanations for results obtained from Machine Learning models. Those explanations are commonly high-score feature values appearing in the input. Several scores have been proposed and investigated. We concentrate mostly on model-agnostic and local scores, that is, the ML model is used as an input/output relation, and the explanations are for individual inputs. In addition to describing some of them, we address the computational challenges they offer. Depending on the classifiers, some hardness and tractability results will be presented, together with an application to neural-network classifiers. We also discuss the relevance of logical reasoning with explanations.
Bio
Leopoldo Bertossi is an Emeritus Professor of Computer Science at Carleton University, Canada; and an Adjunct Professor at Western University, London, Canada. He is a Senior Researcher at the Millennium Institute of Foundational Research on Data (IMFD, Chile). He has a PhD in Mathematics from the Catholic University of Chile (PUC, Chile). His areas of research are Explainable AI, Knowledge Representation, Data Management, and Computational Logic.
Speaker is presenting online.
Practical
The Tuesdays seminar series is devoted to various topics relevant for the Integreat centre´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.
For those unable to attend in person: https://uio.zoom.us/j/65632027408
Point of contacts:
- Organiser and moderator: Martin Giese, University of Oslo
- Integreat Administrative Leader Maria Dikova, University of Oslo