Comparing Solvability Patterns of Algorithms across Diverse Problem Landscapes
A. Nikolikj, T. Eftimov
The Genetic and Evolutionary Computation Conference GECCO 2024
Melbourne, Australia, 14-18 July, 2024
In the field of black-box optimization, it is crucial to understand why certain algorithm excels in some problem instances while struggling with others. Recent efforts have introduced the concept of an algorithm footprint, which identifies both easy and challenging problem instances for an algorithm. However, a major challenge persists—the lack of comparability among different algorithm footprints for effective benchmarking. This paper addresses this issue by introducing a practical solution using multi-target regression models (MTR). These models can predict the performance of multiple algorithms simultaneously based on problem landscape features, creating a shared space for comparing various algorithm footprints. This approach overcomes the current challenge of non-comparability among algorithm footprints, providing a structured framework for evaluating algorithm behavior. By establishing a common landscape feature set and using a single performance prediction model, not only can algorithm footprints be compared, but the explanations for algorithm behavior can also be analyzed systematically. We apply the methodology to a portfolio of three different algorithms, exposing their similarities and differences on the Black-Box Optimization Benchmarking (BBOB) suite.
BIBTEX copied to Clipboard