← Back to all news
arXiv Statistics ML By Mickael Binois (ACUMES), Jeffrey Larson (ANL)

Adaptive Replication Strategies in Trust-Region-Based Bayesian Optimization of Stochastic Functions

Read the original on arXiv Statistics ML →

The paper presents OGPIT, a trust‑region Bayesian optimization method that uses Gaussian process models and adaptive replication to handle stochastic functions with high variance. By allocating repeated evaluations where most beneficial and incorporating cost‑aware acquisition modifications, the approach scales efficiently when many samples are needed to reduce noise. Numerical experiments demonstrate that adaptive replication improves computational efficiency while maintaining solution accuracy compared to baseline methods.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Statistics ML.

Related stories

arXiv Machine Learning
Jun 8

LAGO: A Local-Global Optimization Framework Combining Trust Region Methods and Bayesian Optimization

arXiv:2603. 02970v2 Announce Type: replace Abstract: We introduce LAGO, a LocAl-Global Optimization framework coupling Bayesian Optimization (BO) and gradient-based trust region local refinement through an adaptive competition mechanism for smooth expensive-to-evaluate objective functions with available gradients.

By Eliott Van Dieren, Tommaso Vanzan, Fabio Nobile
More like this →
arXiv Machine Learning
Sep 15

Bayesian Optimisation Using Product-of-Experts Gaussian Process Models with Uncertainty Calibration

The paper introduces BO-pro-c, a Bayesian optimisation algorithm that employs a product-of-experts Gaussian process (GP-pro-c) as its surrogate model. GP-pro-c combines multiple local GP experts to improve uncertainty quantification, reduce computational cost, and preserve global correlations, addressing the cubic complexity of single global GP models. Experiments show that BO-pro-c achieves competitive optimisation performance with a 0.9% lower simple regret and a 39.4% reduction in computational overhead compared to a single‑global‑GP baseline.

By Yean Hoon Ong
More like this →