Improving Pareto Front Learning via Multi-Head HyperNetwork

Explore the intricacies of multi-objective optimization and Pareto front learning in this 54-minute seminar presented by Le Duy Dung, Assistant Professor at VinUniversity. Delve into the challenges of existing Pareto front learning methods and discover a novel Multi-head HyperNetwork (MHN) architecture designed to improve the quality of obtained Pareto fronts. Learn how this approach generates multiple Pareto solutions from diverse trade-off preferences and maximizes hypervolume value to enhance performance. Gain insights into the application of this method across various machine learning tasks and its significant advantages over baseline approaches in producing high-quality Pareto fronts.