The study focuses on the role of Large Language Models (LLMs) in accelerating Bayesian optimization in molecular space. It emphasizes the importance of automation in material discovery, with Bayesian optimization being a crucial component. The researchers evaluate LLMs as fixed feature extractors for Bayesian optimization and utilize parameter-efficient fine-tuning methods and Bayesian neural networks. The findings show that LLMs can be beneficial for Bayesian optimization over molecules, but only if they have been pre-trained or fine-tuned with domain-specific data.
Publication date: 8 Feb 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2402.05015
