analiza aiqbench
This paper examines how LLMs handle the exploration-exploitation trade-off in black-box optimization, a problem that Bayesian optimization handles with explicit acquisition functions. Its main claim is that splitting a single prompt into a strategy agent that sets interpretable criteria weights and a generation agent that proposes candidates reduces the instability and premature convergence seen in single-agent setups. Readers should treat the reported gains as preliminary, since the abstract gives no benchmark specifics, and should check the full results before comparing against established Bayesian optimization baselines.
Multi-Agent LLMs for Adaptive Acquisition in Bayesian Optimization
arXiv:2603.28959v2 Announce Type: replace-cross Abstract: The exploration-exploitation trade-off is central to sequential decision-making and black-box optimization, yet how Large Language Models (LLMs) reason about and manage this trade-off remains poorly understood. Unlike Bayesian Optimization, where exploration and exploitation are explicitly encoded through acquisition functions, LLM-based optimization relies on implicit, prompt-based reasoning over historical evaluations, making search behavior difficult to analyze or control. In this work, we present a metric-level study of LLM-mediated search policy learning, studying how LLMs construct and adapt exploration-exploitation strategies under multiple operational definitions of exploration, including informativeness, diversity, and representativeness. We show that single-agent LLM approaches, which jointly perform strategy selection and candidate generation within a single prompt, suffer from cognitive overload, leading to unstable search dynamics and premature convergence. To address this limitation, we propose a multi-agent framework that decomposes exploration-exploitation control into strategic policy mediation and tactical candidate generation. A strategy agent assigns interpretable weights to multiple search criteria, while a generation agent produces candidates conditioned on the resulting search policy defined as weights. This decomposition renders exploration-exploitation decisions explicit, observable, and adjustable. Empirical results across various continuous optimization benchmarks indicate that separating strategic control from candidate generation substantially improves the effectiveness of LLM-mediated search.
Abstract:The exploration-exploitation trade-off is central to sequential decision-making and black-box optimization, yet how Large Language Models (LLMs) reason about and manage this trade-off remains poorly understood. Unlike Bayesian Optimization, where exploration and exploitation are explicitly encoded through acquisition functions, LLM-based optimization relies on implicit, prompt-based reasoning over historical evaluations, making search behavior difficult to analyze or control. In this work, we present a metric-level study of LLM-mediated search policy learning, studying how LLMs construct and adapt exploration-exploitation strategies under multiple operational definitions of exploration, including informativeness, diversity, and representativeness. We show that single-agent LLM approaches, which jointly perform strategy selection and candidate generation within a single prompt, suffer from cognitive overload, leading to unstable search dynamics and premature convergence. To address this limitation, we propose a multi-agent framework that decomposes exploration-exploitation control into strategic policy mediation and tactical candidate generation. A strategy agent assigns interpretable weights to multiple search criteria, while a generation agent produces candidates conditioned on the resulting search policy defined as weights. This decomposition renders exploration-exploitation decisions explicit, observable, and adjustable. Empirical results across various continuous optimization benchmarks indicate that separating strategic control from candidate generation substantially improves the effectiveness of LLM-mediated search.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2603.28959 [cs.LG]
(or arXiv:2603.28959v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2603.28959 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 291 cuvinte
Subiecte conexe
Agentic AI Reasoning and Learned Sampling for Task-Aware Projection Selection in Sparse-View CT: Predicting Hemodynamics
arXiv · 31 oct. 2026
Amortized Acquisition Optimization for Bayesian Optimization with Variational Mutual Information
arXiv · 6 oct. 2026
ILP-BO: Integer Linear Programming-Based Black-Box Optimization
arXiv · 30 sept. 2026
Groundcover picks up Kubernetes optimization startup Wand in its first acquisition
SiliconANGLE · 24 sept. 2026