Timezone: »

 
Poster
Scalable Thompson Sampling using Sparse Gaussian Process Models
Sattar Vakili · Henry Moss · Artem Artemev · Vincent Dutordoir · Victor Picheny

Tue Dec 07 08:30 AM -- 10:00 AM (PST) @
in Poster Session 1 »

Thompson Sampling (TS) from Gaussian Process (GP) models is a powerful tool for the optimization of black-box functions. Although TS enjoys strong theoretical guarantees and convincing empirical performance, it incurs a large computational overhead that scales polynomially with the optimization budget. Recently, scalable TS methods based on sparse GP models have been proposed to increase the scope of TS, enabling its application to problems that are sufficiently multi-modal, noisy or combinatorial to require more than a few hundred evaluations to be solved. However, the approximation error introduced by sparse GPs invalidates all existing regret bounds. In this work, we perform a theoretical and empirical analysis of scalable TS. We provide theoretical guarantees and show that the drastic reduction in computational complexity of scalable TS can be enjoyed without loss in the regret performance over the standard TS. These conceptual claims are validated for practical implementations of scalable TS on synthetic benchmarks and as part of a real-world high-throughput molecular design task.

Keywords: Optimization · Kernel Methods

Author Information

Sattar Vakili (MediaTek Research)
Henry Moss (Secondmind)

I am a Senior Machine Learning Researcher at Secondmind (formerly PROWLER.io). I leverage information-theoretic arguments to provide efficient, reliable and scalable Bayesian optimisation for problems inspired by science and the automotive industry.

Artem Artemev (Imperial College London)
Vincent Dutordoir (University of Cambridge)
Victor Picheny (Prowler)

More from the Same Authors