New benchmark analysis reveals how transformations affect hMPA optimizer

A new paper on arXiv (2607.22862) by Grzegorz Sroka and Sławomir T. Wierzchoń tackles a longstanding issue in benchmark testing: the standard CEC 2017 benchmark applies bias, shift, and rotation simultaneously, making it impossible to isolate each transformation's effect on algorithmic behavior. The authors introduce a parameterized implementation that controls these transformations independently while preserving the original functions and transformation data. Using this framework, they diagnose the hybrid Marine Predators Algorithm (hMPA), whose candidate mechanism depends on numerical objective values and coordinate-wise reconstruction.

To analyze results, the team adapted DSC and extended DSC (eDSC) from algorithm-level comparison to configuration-level diagnosis, enabling the first exhaustive analysis of all eight bias-shift-rotation configurations. Experiments covered 29 functions, dimensions 10-100, 30 independent runs, and a budget of 10,000*dim evaluations. Key findings: shift generally worsens objective values; isolated rotation causes no systematic deterioration; isolated bias has little effect on final-solution distributions. Shift-rotation combinations differ most consistently from the control. While rankings vary across functions preventing a consistent ordering, the framework provides a reproducible diagnostic protocol extensible to other continuous optimizers and related CEC benchmarks (2024, 2025).