B0623
Title: Multi-fidelity Bayesian optimization in high-dimensional settings
Authors: Hendriico Merila - University of Southampton (United Kingdom) [presenting]
Abstract: High-dimensional computer models are found commonly in many modern applications. Oftentimes, that model is optimised and its minimum or maximum value is found. Unfortunately, this is a very challenging task for many reasons, one of them being the curse of dimensionality. Bayesian optimisation is often used for such scenarios. The focus is on noisy computer models whose accuracy is controlled by choosing the amount of computational budget allocated to each run. A number of techniques are used for mathematical optimisation and are adopted into the Bayesian optimisation framework. This allows defining a novel algorithm for multi-fidelity Bayesian optimisation that is appropriate for noisy, high-dimensional computer models.
