This page organizes papers related to artificial intelligence published around the world. This page is summarized using Google Gemini and is operated on a non-profit basis. The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.
Chih-Yu Chang, Milad Azvar, Chinedum Okwudire, Raed Al Kontar
LLINBO: LLM-in-the-Loop BO
Outline
Bayesian optimization (BO) is a widely used sequential decision-making tool for high-cost black-box function optimization. Recently, large-scale language models (LLMs) have demonstrated remarkable adaptability in low-data environments and have become a promising tool for black-box optimization by leveraging contextual knowledge to suggest high-quality query points. However, relying solely on LLMs as optimization agents poses risks due to the lack of explicit surrogate modeling and calibrated uncertainty, as well as inherently opaque internal mechanisms. This structural opacity makes it difficult to characterize or control the exploration-exploitation tradeoff, thereby compromising theoretical addressability and reliability. To address this, we propose LLINBO: LLM-in-the-Loop BO, a hybrid framework for BO that combines LLMs with statistical surrogate experts (e.g., Gaussian Processes (GP)). The core philosophy is to leverage the contextual inference strengths of LLMs for early exploration while relying on principled statistical models for efficient exploitation. Specifically, we introduce three mechanisms that enable this collaboration and establish theoretical guarantees. We conclude the paper with a practical proof-of-concept in the context of 3D printing. The result reproduction code can be found at https://github.com/UMDataScienceLab/LLM-in-the-Loop-BO에서 .
Takeaways, Limitations
•
We present a novel Bayesian optimization framework that combines the contextual inference capabilities of LLM with the efficient utilization of statistical models.
•
Leverage LLM for early exploration and rely on statistical models for efficient utilization.
•
Presenting a proof of concept for a real-world problem in 3D printing.
•
Addressing the issues of LLM's lack of uncertainty and opacity of its internal mechanisms
•
Presenting a mechanism that provides theoretical guarantees
•
Specific Limitations is not specified in the paper.
