Event Details

STOR-i Masterclass, Associate Professor Peter Frazier, Cornell University - Bayesian Optimization

Bayesian Optimization

Bayesian optimization is a powerful tool for optimizing time-consuming-to-evaluate non-convex derivative-free objective functions. It is widely used for tuning hyper parameters in deep neural networks, and has wider applications including engineering design, materials and drug discovery,and tuning e-commerce systems and online markets with A/B testing. It uses a machine learning predictive method, often Gaussian process regression,to model the objective function and an acquisition function to guide the choice of points at which to evaluate the objective. We give an introduction to these methods, describing Gaussian process regression and acquisition functions including expected improvement, Thompson sampling, and knowledge gradient. Practice-oriented discussions will be supported with labs in pure python and Facebook's BoTorch BayesOpt library. We then briefly touch on more advanced techniques: parallel evaluations, constraints, high-dimensional inputs, optimization with preference learning, and grey-box Bayesian optimization including multi-information-source optimization, multi-fidelity optimization and composite functions.

Monday 16 March

10:00-12:00 Lecture1 Bowland North SR6

14:30-16:30 Lecture2 FST Training Rooms (SAT A76)

Tuesday 17 March

10:00-12:00 Lecture3 Furness LT3