STREGO: A Scalable Implementation of a Trust-Region Framework for Noisy Bayesian Optimization

Name: STREGO: A Scalable Implementation of a Trust-Region Framework for Noisy Bayesian Optimization
Uploaded: 2026-10-18T12:52:24.171210+00:00
Duration: 543 s
Description: STREGO is a scalable Bayesian optimization method that uses a trust-region framework to handle noisy function evaluations while maintaining convergence guarantees. It achieves competitive performance in high dimensions by combining a global phase that considers uncertainty reduction across the whole domain with a local phase that focuses only on points near the current best solution.

STREGO is a scalable Bayesian optimization method that uses a trust-region framework to handle noisy function evaluations while maintaining convergence guarantees. It achieves competitive performance in high dimensions by combining a global phase that considers uncertainty reduction across the whole domain with a local phase that focuses only on points near the current best solution.

Open the full episode