Researchers test multi-solution generation for evolutionary diversity optimisation
New study explores generating multiple solutions per generation to boost diversity in optimisation.
A new paper on arXiv (2606.21161) from researchers Adel Nikfarjam, Jakob Bossek, Aneta Neumann, and Frank Neumann explores a key question in evolutionary diversity optimisation (EDO): can generating multiple solutions per generation outperform the standard single-solution approach? Traditional EDO builds a diverse set of high-quality solutions by iteratively creating one new solution and removing the one that contributes least to diversity. This gradual process ensures diversity increases steadily, but it may be slow. The study asks whether producing several new solutions simultaneously could yield faster convergence to a more diverse population.
The major hurdle is survival selection: when multiple solutions are changed at once, each solution's contribution to diversity becomes interdependent, making standard selection methods (e.g., truncation or tournament selection) ineffective. The authors propose and test alternative survival strategies tailored for this multi-solution-per-generation setting. Their results, to be presented at the 2026 IEEE Congress on Evolutionary Computation (CEC), could lead to more efficient algorithms for problems requiring diverse, high-quality solution sets—such as engineering design, drug discovery, or AI model configuration.
- Traditional EDO creates only one new solution per generation to maintain steady diversity increase.
- The paper investigates whether generating multiple solutions per generation speeds up diversity optimisation.
- Conventional survival selection fails when multiple solutions change simultaneously; the authors propose new methods to handle interdependence.
Why It Matters
Faster diversity optimisation could dramatically improve solution quality in engineering, drug discovery, and AI.