Bilevel optimization by conditional Bayesian optimization
Explainable AI (XAI) can greatly enhance user trust and satisfaction in AI-assisted decision-making processes. Numerous explanation techniques (explainers) exist in the literature, and recent findings suggest that addressing multiple user needs requires employing a combination of these explainers. We refer to such combinations as explanation strategies. This paper introduces iSee - Intelligent Sharing of Explanation Experience, an interactive platform that facilitates the reuse of explanation strategies and promotes best practices in XAI by employing the Case-based Reasoning (CBR) paradigm. iSee uses an ontology-guided approach to effectively capture explanation requirements, while a behaviour tree-driven conversational chatbot captures user experiences of interacting with the explanations and provides feedback. In a case study, we illustrate the iSee CBR system capabilities by formalising a real-world radiograph fracture detection system and demonstrating how each interactive tools facilitate the CBR processes.
Bilevel optimization , Conditional Bayesian optimization , Stackelberg games , Gaussian process
Dogan, V. and Prestwich, S. (2023) 'Bilevel optimization by conditional Bayesian optimization', The 9th International Conference on Machine Learning, Optimization, and Data Science, Grasmere, UK. September 22-26, forthcoming publication
© the authors 2023