FGSS-BO: Feedback-Guided Hypothesis Management for LLM-Based Bayesian Optimization

Key points

• Large language model (LLM)-guided Bayesian optimization (BO) uses past observations to guide search, but proposed directions and their revision can remain implicit.1
• An LLM reflector proposes and revises hypotheses; a deterministic curator validates them and maintains identities and test records.1
• On two retrospective protocol/media tasks with nearest-neighbor response oracles, it attains higher sample mean final best and search AUC than the matched LLAMBO baseline and evaluated non-LLM baselines.1
• Six direct-training hyperparameter-optimization tasks show task-dependent transfer, strongest on diabetes SVM; exact-grid and reaction controls examine other settings.1

Sources (1)

• [1]FGSS-BO: Feedback-Guided Hypothesis Management for LLM-Based Bayesian Optimization
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 15, 04:42 AM

  We propose Feedback-Guided Search State (FGSS), a structured hypothesis-management interface connecting language-based guidance to observed evidence, executable parameter interventions, and comparative support updates.

  Large language model (LLM)-guided Bayesian optimization (BO) uses past observations to guide search, but proposed directions and their revision can remain implicit.

Extractive summary: sentences quoted from the sources.