Keynote talk by Prof. Michael Emmerich
Title: Multiobjective Optimization and Decision Analysis from an Explainable AI Perspective
Abstract: It is essential in the modern practice of prediction to add explanations to predictions and results found by optimization. Reasons are that the decision-maker holds responsibility for decisions, users want to be informed about the reasons for decisions, and in application domains, people want to not only make optimal decisions but also learn about their domain and gain trust in a solution.
Surprisingly, in the literature of explainable AI, the focus is almost entirely on prediction and the important AI field of optimization is often neglected. However, in practice, many decision-making is based rather on optimizations rather than solely on predictions.
In multicriteria optimization, however, the concept of explanations is often discussed under other technical terms, for instance, innovation or sensitivity analysis. Moreover, predictions are frequently used as objective functions or constraint functions and it could be interesting how to make use of explanations of predictions in optimization.
In my talk, I will give an overview of these aspects and perspectives of how explanations and explainability are related to multicriteria optimization, existing theories, and topics that would qualify as viable topics for future research. I will give examples from drug discovery and building architectural design, and epidemic management, among others.