Explainable Benchmarking for Iterative Optimization Heuristics

Genetic and Evolutionary Computation Conference Companion (GECCO Companion), pp. 79–80

Abstract

This paper summarizes our recent work published in ACM Transactions on Evolutionary Learning and Optimization (2024), entitled "Explainable Benchmarking for Iterative Optimization Heuristics". We introduce IOHxplainer, a novel software library designed to systematically and explainably analyze the performance impact of various components and hyperparameters in modular or parametrized optimization heuristics. By leveraging explainable AI techniques such as SHAP values and machine learning models, IOHxplainer allows for an in-depth analysis of the contributions and interactions of algorithmic components across diverse benchmark scenarios.

Authors 4

1. Bas van Stein Aachen

   Leiden University

   Affiliation as printed

   Leiden University, Leiden, Netherlands

2. Diederick Vermetten Aachen

   Leiden University

   Affiliation as printed

   Leiden University, Leiden, Netherlands

3. Anna V. Kononova Aachen

   Leiden University

   Affiliation as printed

   Leiden University, Leiden, Netherlands

4. Thomas Bäck Aachen

   Leiden University

   Affiliation as printed

   Leiden University, Leiden, Netherlands