Keynote at EvoStar 2025

At the EvoStar 2025 conference in Trieste, Italy, Tea Tušar gave a keynote titled “Ideals and Realities of Benchmarking in Evolutionary Multiobjective Optimization.” Her talk addressed a critical disconnect between the design of benchmarking experiments and the demands of real-world optimization problems. She emphasized that current practices often evaluate algorithms at a fixed number of solution evaluations, overlooking their anytime performance, which is an essential factor for selecting algorithms in time-constrained, real-world applications. Additionally, most studies ignore scalability across different problem dimensionalities, and the lack of a shared performance archive hampers fair comparisons with state-of-the-art methods. The speaker advocated for benchmarking that reflects real usage scenarios and warned against relying on overly simplistic or unrealistic problems that cause algorithm overfitting.