LAGO: A Local-Global Optimization Framework Combining Trust Region Methods and Bayesian Optimization
Researchers have developed LAGO, a novel framework that combines Bayesian Optimization with gradient-based trust region methods for optimizing expensive-to-evaluate functions. This approach adaptively balances global exploration with local refinement, proposing candidate points from both strategies and selecting the next evaluation based on predicted improvement. LAGO aims to enhance Bayesian Optimization by efficiently refining solutions in promising areas while maintaining exploratory behavior when local steps are less competitive, and it incorporates a mechanism to reduce numerical instability. AI
IMPACT Introduces a new optimization technique that could improve the efficiency of training complex AI models.