Automated Kernel Discovery Towards Understanding High-dimensional Bayesian Optimization

In three linesKernel Discovery, an LLM-driven evolutionary framework, automates Gaussian kernel design for high-dimensional Bayesian optimization. The method generates novel mathematical forms via LLM, converts them to validated code, and uses LOO-CRPS criterion to penalize overfitting. On 5 benchmarks, it achieves average rank 1.2 out of 17.

Summary generated by Claude — human-verified