Benchmarking the Hooke-Jeeves Method, MTS-LS1, and BSrr on the Large-scale BBOB Function Set

TL;DR

This paper benchmarks three black-box optimization methods on large-scale BBOB functions, revealing that BSrr performs exceptionally well on separable functions, while the Hooke-Jeeves method outperforms MTS-LS1 on unimodal problems.

Contribution

It provides a comparative analysis of Hooke-Jeeves, MTS-LS1, and BSrr on large-scale BBOB functions, highlighting the effectiveness of BSrr and the influence of asymmetry.

Findings

BSrr shows state-of-the-art performance on separable large-scale functions.

Asymmetry significantly affects MTS-LS1's performance.

Hooke-Jeeves outperforms MTS-LS1 on unimodal separable functions.

Abstract

This paper investigates the performance of three black-box optimizers exploiting separability on the 24 large-scale BBOB functions, including the Hooke-Jeeves method, MTS-LS1, and BSrr. Although BSrr was not specially designed for large-scale optimization, the results show that BSrr has a state-of-the-art performance on the five separable large-scale BBOB functions. The results show that the asymmetry significantly influences the performance of MTS-LS1. The results also show that the Hooke-Jeeves method performs better than MTS-LS1 on unimodal separable BBOB functions.