Unleashing LLMs in Bayesian Optimization: Preference-Guided Framework for Scientific Discovery

2026-05-18

AI Summary

This work addresses the limitations of Bayesian optimization in scientific experimentation—namely, slow cold-start convergence and poor scalability in high-dimensional spaces—by introducing LLM-guided Bayesian optimization (LGBO). LGBO uniquely integrates semantic preferences from large language models (LLMs) throughout the entire optimization process via a region-boosting preference mechanism that dynamically adjusts the surrogate model’s mean function, enabling controllable guidance while theoretically guaranteeing that worst-case performance is no worse than standard Bayesian optimization. Evaluated across multiple benchmarks in physics, chemistry, biology, and materials science, LGBO consistently outperforms existing methods. In wet-lab experiments optimizing Fe–Cr battery electrolytes, LGBO achieved 90% of the optimal performance in just six iterations, substantially accelerating the discovery process.

Abstract

Scientific discovery is increasingly constrained by costly experiments and limited resources, underscoring the need for efficient optimization in AI for science. Bayesian Optimization (BO), though widely adopted for balancing exploration and exploitation, often exhibits slow cold-start performance and poor scalability in high-dimensional settings, limiting its applicability in real-world scientific problems. To overcome these challenges, we propose LLM-Guided Bayesian Optimization (LGBO), the first LLM preference-guided BO framework that continuously integrates the semantic reasoning of large language models (LLMs) into the optimization loop. Unlike prior works that use LLMs only for warm-start initialization or candidate generation, LGBO introduces a region-lifted preference mechanism that embeds LLM-driven preferences into every iteration, shifting the surrogate mean in a stable and controllable way. Theoretically, we prove that LGBO does not perform significantly worse than standard BO in the worst case, while achieving significantly faster convergence when preferences align with the objective. Empirically, LGBO consistently outperforms existing methods across diverse dry benchmarks in physics, chemistry, biology, and materials science. Most notably, in a new wet-lab optimization of Fe-Cr battery electrolytes, LGBO attains \textbf{90\% of the best observed value within 6 iterations}, whereas standard BO and existing LLM-augmented baselines require more than 10. Together, these results suggest that LGBO offers a promising direction for integrating LLMs into scientific optimization workflows.

Problem

• Bayesian Optimization
• cold-start
• high-dimensional optimization
• scientific discovery
• scalability

Innovation

• LLM-guided optimization
• Bayesian optimization
• preference-guided framework
• scientific discovery
• region-lifted preference