Multi-Objective Population Based Training

TL;DR

This paper introduces MO-PBT, a multi-objective extension of Population Based Training, designed to optimize conflicting hyperparameters efficiently across multiple objectives, outperforming existing methods in diverse scenarios.

Contribution

The paper presents MO-PBT, the first multi-objective version of Population Based Training, capable of handling multiple conflicting objectives in hyperparameter optimization.

Findings

MO-PBT outperforms random search and single-objective PBT.

MO-PBT surpasses the state-of-the-art MO-ASHA in experiments.

Effective in optimizing multiple conflicting objectives like accuracy and fairness.

Abstract

Population Based Training (PBT) is an efficient hyperparameter optimization algorithm. PBT is a single-objective algorithm, but many real-world hyperparameter optimization problems involve two or more conflicting objectives. In this work, we therefore introduce a multi-objective version of PBT, MO-PBT. Our experiments on diverse multi-objective hyperparameter optimization problems (Precision/Recall, Accuracy/Fairness, Accuracy/Adversarial Robustness) show that MO-PBT outperforms random search, single-objective PBT, and the state-of-the-art multi-objective hyperparameter optimization algorithm MO-ASHA.