Gaussian Process Bandits with Paired Auxiliary Observations
Abstract
Bayesian optimization (BO) is widely used for expensive black-box optimization, yet observation noise makes each evaluation less informative and slows posterior learning. Standard BO learns from costly outcomes gathered during optimization. In many applications, however, low-cost auxiliary data are available beforehand. We introduce Auxiliary-Informed Gaussian Process Bandit (AI-GPB), a variance-reduction method that combines an offline estimate of the auxiliary mean with a correction coefficient learned from paired online observations. Notably, it exploits correlation between paired observation errors without requiring the auxiliary mean to approximate the objective. We establish finite-sample GP confidence bounds and regret guarantees, including explicit cumulative and simple regret rates for Matérn kernels. Numerical experiments across continuous BO benchmarks with varied landscapes and dimensions show consistent reductions in cumulative and simple regret relative to vanilla GP-UCB, even when within-pair correlation is moderate and the auxiliary mean is estimated from noisy offline data.
Related
- Near-Optimal Regret Bounds for UCB-Style Gaussian Process Bandits via Data Thinning
- Gaussian Process Bandits for AI System Optimization with Selectively Verified Feedback
- Cascading Bandits with Noisy Observations: Algorithms and Regret Bounds
- Exact Bayes Regret and Asymptotic Optimality in High-Dimensional Gaussian Bandits
- Adaptive Random Matrices in Gaussian Bandits: Spectral Universality and Selection-Induced Outliers
BibTeX
@misc{huang2026gaussian,
title = {{Gaussian Process Bandits with Paired Auxiliary Observations}},
author = {Weibo Huang and Qing Feng and Ruihao Zhu and Cheng Hua},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.24737},
eprint = {2610.24737},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.24737},
}
RIS
TY - UNPB
TI - Gaussian Process Bandits with Paired Auxiliary Observations
AU - Weibo Huang
AU - Qing Feng
AU - Ruihao Zhu
AU - Cheng Hua
PY - 2026
DA - 2026/10/17
PB - arXiv
DO - 10.48550/arxiv.2610.24737
UR - https://arxiv.org/abs/2610.24737
AB - Bayesian optimization (BO) is widely used for expensive black-box optimization, yet observation noise makes each evaluation less informative and slows posterior learning. Standard BO learns from costly outcomes gathered during optimization. In many applications, however, low-cost auxiliary data are available beforehand. We introduce Auxiliary-Informed Gaussian Process Bandit (AI-GPB), a variance-reduction method that combines an offline estimate of the auxiliary mean with a correction coefficient learned from paired online observations. Notably, it exploits correlation between paired observation errors without requiring the auxiliary mean to approximate the objective. We establish finite-sample GP confidence bounds and regret guarantees, including explicit cumulative and simple regret rates for Matérn kernels. Numerical experiments across continuous BO benchmarks with varied landscapes and dimensions show consistent reductions in cumulative and simple regret relative to vanilla GP-UCB, even when within-pair correlation is moderate and the auxiliary mean is estimated from noisy offline data.
ER -