Скачать или смотреть IEEE CIS Webinar: Evolving Metaheuristics with LLMs The LLaMEA Approach to Automated

Automated Algorithm Design (AAD) is accelerating as large language models join evolutionary search to invent optimisation heuristics on demand. This webinar unveils LLaMEA, a framework that lets LLMs mutate and evaluate code, discovering algorithms that rival CMA-ES, TuRBO and hand-designed heuristics for continuous, combinatorial and Bayesian optimisation. We will dissect prompt engineering, heavy-tailed mutation control, in-the-loop hyper-parameter tuning, and BLADE, a capability-focused benchmark for LLM-driven design. Practical insights on cost management and industrial deployment are provided. The session closes with a summary, take-away messages and future research directions. Attendees will gain actionable guidelines, open-source tools, and perspective on the evolving AAD landscape.

Biography:

Niki van Stein is Assistant Professor of Explainable AI and Evolutionary Computing at Leiden University, leading the XAI research group within the Natural Computing Cluster. Her work unites evolutionary search, large language models and transparent machine learning, yielding projects like the BLADE benchmarking suite and LLaMEA, an LLM-guided evolutionary algorithm that designs optimisation heuristics. She has authored more than ninety peer-reviewed publications, is active in industry projects on predictive maintenance, engineering design and fair AI. Niki promotes reproducible science, open-source software and interdisciplinary education. She earned a PhD from Leiden University in 2018.