Meta-learning for scalable multi-objective Bayesian optimization

Jiarong Pan

PhD at Bosch Center for Artificial Intelligence

Abstract: Many real-world applications consider multiple objectives, potentially competing ones. For instance, for a model deciding whether to grant or deny loans, ensuring accurate while fair decisions is critical. Multi-objective Bayesian optimization (MOBO) is a sample-efficient technique for optimizing an expensive black-box function across multiple objectives. However, many MOBO methods cannot scale with the number of the objectives and observations due to their underlying probabilistic models and performance metrics, limiting its practicality. Moreover, few MOBO techniques consider the case using available historical data from similar tasks to improve the sample-efficiency. This research aims to explore advanced strategies in MOBO, focusing on the applications with more number of objectives and data, where traditional methods fall short due to computational constraints.

Keywords: Multi-objective optimization,Bayesian optimization,meta-learning
Scientific area: Optimization
Visiting period: 1st April to 30th June 2024
Visiting Lab: Automated Machine Learning Lab, TUE

Bio: Jiarong Pan is a second year PhD candidate at Bosch Center for Artificial Intelligence. The goal of his research is to find sample-efficient solutions for optimization problems using machine learning. His research topic includes Bayesian optimization, multi-objective optimization and meta-learning.