We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Statistics > Machine Learning

Title: A sequential Monte Carlo approach to Thompson sampling for Bayesian optimization

Abstract: Bayesian optimization through Gaussian process regression is an effective method of optimizing an unknown function for which every measurement is expensive. It approximates the objective function and then recommends a new measurement point to try out. This recommendation is usually selected by optimizing a given acquisition function. After a sufficient number of measurements, a recommendation about the maximum is made. However, a key realization is that the maximum of a Gaussian process is not a deterministic point, but a random variable with a distribution of its own. This distribution cannot be calculated analytically. Our main contribution is an algorithm, inspired by sequential Monte Carlo samplers, that approximates this maximum distribution. Subsequently, by taking samples from this distribution, we enable Thompson sampling to be applied to (armed-bandit) optimization problems with a continuous input space. All this is done without requiring the optimization of a nonlinear acquisition function. Experiments have shown that the resulting optimization method has a competitive performance at keeping the cumulative regret limited.
Subjects: Machine Learning (stat.ML); Systems and Control (eess.SY)
Cite as: arXiv:1604.00169 [stat.ML]
  (or arXiv:1604.00169v3 [stat.ML] for this version)

Submission history

From: Hildo Bijl [view email]
[v1] Fri, 1 Apr 2016 09:07:26 GMT (1149kb,D)
[v2] Wed, 6 Jul 2016 10:44:59 GMT (1118kb,D)
[v3] Tue, 16 May 2017 16:26:48 GMT (636kb,D)

Link back to: arXiv, form interface, contact.