Title: Assessing the Real-World Applicability of Algorithm Selection Models

Abstract:

Algorithm Selection (AS) seeks to automatically determine the optimal optimization algorithm for specific problem instances by utilizing measurable problem traits and past performance metrics. This research examines the capacity of AS models to generalize across both artificial and genuine optimization environments. The study incorporates two prominent academic benchmark suites, BBOB and CEC, alongside two practical problem domains: robotics trajectory optimization and unmanned aerial vehicle path-planning. By conducting a systematic cross-benchmark assessment, we explore the transferability of AS models between different fields, pinpointing areas where generalization is effective or fails, and underscoring the difficulties encountered when deploying AS in authentic, context-specific scenarios. These results offer valuable perspectives on the resilience of existing AS methodologies and guide the creation of more dependable and widely usable AS systems for practical optimization challenges.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC