Preference-based Multi-Objective Bayesian Optimization with Gradients

Joshua Hang Sai Ip, Ankush Chakrabarty, Hideyuki Masui, Ali Mesbah, Diego Romeres

Keywords: Machine Learning, Bayesian Learning, Human-in-the-loop Machine Learning

TL;DR: We introduce a novel algorithm for multi-objective Bayesian Optimization that combines utility-based approaches with multi-gradient descent to approach Pareto optimality.

Abstract: We propose PUB-MOBO for personalized multi-objective Bayesian Optimization. PUB-MOBO combines utility-based MOBO with local multi-gradient descent to refine user-preferred solutions to be near-Pareto-optimal. Unlike traditional methods, PUB-MOBO does not require estimating the entire Pareto-front, making it more efficient. Experimental results on synthetic and real-world benchmarks show that PUB-MOBO consistently outperforms existing methods in terms of proximity to the Pareto-front and utility regret.

Submission Number: 108