Event Information:

• Wed

  19

  Feb

  2025

  SC Seminar: Gonçalo Daniel Esteves Marques

  11:00Room 32-349

  Gonçalo Daniel Esteves Marques, RPTU

  Title: Trust Region Bayesian Optimization for Global Optimization

  Abstract:

  Optimizing complex and computationally expensive systems is a significant challenge in science and engineering, particularly when the problem’s structure is unknown or impossible to simulate. This thesis examines Trust Region Bayesian Optimization (TuRBO), a novel framework designed to address the limitations of existing optimization techniques in high-dimensional and multimodal search spaces. Conventional approaches often struggle to balance global exploration and local refinement, especially in these challenging contexts.TuRBO employs trust region methods and Gaussian Process modeling to dynamically partition the search space into smaller, adaptive regions. This approach enables efficient navigation by balancing broad exploration with targeted refinement of promising solutions. By dynamically adjusting the trust regions based on optimization progress, TuRBO demonstrates a unique capability to handle the complexities of high-dimensional problems. This thesis provides a comprehensive study of the TuRBO framework, exploring its theoretical foundations, scalability, and adaptability. It reflects on the implications of this method for real-world applications, contributing to the development of more efficient and robust optimization strategies for fields such as machine learning, engineering design, and robotics.