High Throughput Bayesian Optimization for Large-Scale Data Applications

Explore a comprehensive seminar talk on High Throughput Bayesian Optimisation delivered by Mr. Victor Picheny at the INI Seminar Room. Learn how Bayesian optimisation can be adapted for high data volume scenarios, particularly relevant to automotive industry applications. Discover innovative approaches developed at Secondmind that address the computational challenges of traditional Gaussian Process models, which typically become impractical with datasets exceeding thousands of points. Understand the implementation of sparse variational GP models and Thompson sampling strategies that enable Bayesian optimisation to maintain both theoretical validity and practical effectiveness when handling thousands to millions of observations. The 59-minute presentation demonstrates how to overcome the limitations of small data regimes while maintaining the powerful optimization capabilities that Bayesian methods are known for.