A LLM Evolutionary Algorithm for Automatically Generating Bayesian Optimization Algorithms

Explore an innovative evolutionary computation framework that leverages Large Language Models to automatically generate and optimize Bayesian Optimization algorithms in this 47-minute seminar. Discover how LLaMEA (Large Language Model Evolutionary Algorithm) uses LLMs as creative co-pilots to iteratively mutate and improve complete codebases, moving beyond traditional parameter tuning to generate entire optimization algorithms on demand. Learn about the specific application to Bayesian Optimization through LLaMEA-BO, which demonstrates how LLM-generated code, guided by adapted prompt engineering and mutation schedules, consistently discovers high-performing BO algorithms that scale effectively across different dimensions and optimization tasks. Examine how this approach generates competitive optimizers across continuous, combinatorial, and hyperparameter spaces, with practical insights into integration with AutoML toolchains and evaluation using the BLADE benchmarking suite. Gain access to open-source implementations and practical takeaways for applying LLM-driven algorithm generation in your own optimization challenges, presented by researchers Elena Raponi and Niki van Stein from Leiden University.