Beyond Accuracy: A Multi-Objective Approach to Machine Learning
Intervenant(s) : Yaochu JIN (University of Surrey, Guildford, U.K.)
Résumé: Machine learning is inherently a multi-objective optimization problem, and its multiple objectives are typically conflicting to each other. Instead of converting the multiple objectives into a single one using hyperparameters, like commonly done in traditional machine learning, this talk presents a Pareto-based approach to multi-objective machine learning, where the aim is to achieve multiple models representing tradeoffs between accuracy and complexity, accuracy and interpretability, or accuracy and diversity. Examples of multi-objective generation of interpretable fuzzy systems, extraction of interpretable symbolic rules from trained neural networks, multi-objective feature extraction, and multi-objective federated learning will be given. The talk is concluded by a brief summary and an outline of remaining challenges.
Plus d'information : http://lfi.lip6.fr/seminaires
Ce séminaire est organisé conjointement avec le Chapitre Français de l’IEEE Computational Intelligence Society (http://ieee-ci.lip6.fr/)
Christophe.Marsala (at) nulllip6.fr