Gaussian Process Bandits for AI System Optimization with Selectively Verified Feedback
Abstract
Gaussian process bandits exploit similarities among configurations to optimize prompts, retrieval systems, and agent workflows. In these applications, however, inexpensive model-based evaluations can disagree with trusted task outcomes. We study selective verification when an action is chosen before a diagnostic is observed and trusted feedback is informative on an unknown set of diagnostic states. A finite reliability class and a finite RKHS signal dictionary describe these two sources of uncertainty. We construct a fixed audit design that combines reliability coverage with reward separation and establish a finite-time regret bound that accounts for a hard audit budget and the rounds needed to obtain informative observations. A safe adaptive extension mixes this robust design with predictable Chernoff allocations while preserving a uniform identification floor and exact expected acquisition-cost accounting. For signed-coordinate kernels, structured and diffuse reliability classes require different audit scales despite sharing the same reward geometry. Synthetic experiments distinguish these certified designs from an adaptive posterior heuristic and GP controls. Safe adaptive allocation reduces aggregate regret relative to the fixed design under the exact dictionary, but its finite-budget error certificate is numerically loose. Conditional-feedback errors can remove the heuristic's advantage and target perturbations can reverse it. Separate AI-task replays assess feedback correction and fitted-dictionary policies without establishing end-to-end validation of a learned selective-verification method.
Related
- Improved Regret Analysis for Parallel Gaussian Process Bandit Optimization
- Exploration strategies for Gaussian process bandit optimization with inference regret guarantees
- ISBO: Scalable Spatio-Temporal Bayesian Optimization with Log Gaussian Cox Process Models via the INLA-SPDE Approach
- Convergence Rates of Gaussian Process Thompson Sampling for Efficient Global Optimization
- Henstock--Kurzweil Gauge Integral in the Non--Gaussian Regime: A Machine--Verified Construction
BibTeX
@misc{wu2026gaussian,
title = {{Gaussian Process Bandits for AI System Optimization with Selectively Verified Feedback}},
author = {Shuang Wu},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.17760},
eprint = {2610.17760},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.17760},
}
RIS
TY - UNPB
TI - Gaussian Process Bandits for AI System Optimization with Selectively Verified Feedback
AU - Shuang Wu
PY - 2026
DA - 2026/10/13
PB - arXiv
DO - 10.48550/arxiv.2610.17760
UR - https://arxiv.org/abs/2610.17760
AB - Gaussian process bandits exploit similarities among configurations to optimize prompts, retrieval systems, and agent workflows. In these applications, however, inexpensive model-based evaluations can disagree with trusted task outcomes. We study selective verification when an action is chosen before a diagnostic is observed and trusted feedback is informative on an unknown set of diagnostic states. A finite reliability class and a finite RKHS signal dictionary describe these two sources of uncertainty. We construct a fixed audit design that combines reliability coverage with reward separation and establish a finite-time regret bound that accounts for a hard audit budget and the rounds needed to obtain informative observations. A safe adaptive extension mixes this robust design with predictable Chernoff allocations while preserving a uniform identification floor and exact expected acquisition-cost accounting. For signed-coordinate kernels, structured and diffuse reliability classes require different audit scales despite sharing the same reward geometry. Synthetic experiments distinguish these certified designs from an adaptive posterior heuristic and GP controls. Safe adaptive allocation reduces aggregate regret relative to the fixed design under the exact dictionary, but its finite-budget error certificate is numerically loose. Conditional-feedback errors can remove the heuristic's advantage and target perturbations can reverse it. Separate AI-task replays assess feedback correction and fitted-dictionary policies without establishing end-to-end validation of a learned selective-verification method.
ER -