DASH is an LLM-enhanced framework for automated Bayesian optimization that decouples surrogate selection from acquisition adaptation. Surrogates are chosen using predictive reliability, uncertainty calibration, and ranking consistency, while a two-stage controller reallocates quotas among acquisition functions and lets an LLM make the final choice from a BO shortlist. The framework also adds knowledge-guided warm starts and structured memory. Across four chemical optimization tasks, the paper reports a 12.51% improvement in trajectory-level Acceleration Factor and a 5.00% improvement in endpoint Enhancement Factor over the strongest AutoBO baseline. Ablations and contamination checks are also reported.
No heat snapshots are available in the last 24 hours.