BO-pro-c-tr reduces regret and overhead in large-scale Bayesian optimization
Original title: Large-Scale Bayesian Optimisation Using Uncertainty-Calibrated Product-of-Experts Gaussian Process Models with Trust Regions
Achieves 2.6% reduction in simple regret and 34.9% in computational overhead compared to independent local GP models.
The method uses product-of-experts Gaussian process with uncertainty calibration and one-to-many association strategy. It addresses limitations of independent local GPs in large-scale optimization. Empirical results show improvements over baselines.
- 2.6% average reduction in simple regret
- 34.9% reduction in computational overhead
- One-to-many association for trust regions integration
- Uses GP-pro-c surrogate model
From the original article
1 min read at arxiv.orgThe use of collections of independent local Gaussian processes (GP-ind) combined with trust regions is a widely adopted approach for large-scale and high-dimensional Bayesian optimisation (BO).
By Yean Hoon Ong · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.