Tree-Structured Synergy of Large Language Models and Bayesian Optimization for Efficient CASH
Conference on Neural Information Processing Systems (NeurIPS), 2026. Poster.
Overview
LB-MCTS combines large language models and Bayesian optimization within a shared Monte Carlo tree search framework for combined algorithm selection and hyperparameter optimization (CASH). A reliability-aware policy balances semantic proposals from an LLM with quantitative proposals from Bayesian optimization as the search progresses.