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

Download:

Current browse context:

stat.ML

Change to browse by:

References & Citations

Bookmark

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

Statistics > Machine Learning

Title: Bayesian Optimization for Policy Search via Online-Offline Experimentation

Abstract: Online field experiments are the gold-standard way of evaluating changes to real-world interactive machine learning systems. Yet our ability to explore complex, multi-dimensional policy spaces - such as those found in recommendation and ranking problems - is often constrained by the limited number of experiments that can be run simultaneously. To alleviate these constraints, we augment online experiments with an offline simulator and apply multi-task Bayesian optimization to tune live machine learning systems. We describe practical issues that arise in these types of applications, including biases that arise from using a simulator and assumptions for the multi-task kernel. We measure empirical learning curves which show substantial gains from including data from biased offline experiments, and show how these learning curves are consistent with theoretical results for multi-task Gaussian process generalization. We find that improved kernel inference is a significant driver of multi-task generalization. Finally, we show several examples of Bayesian optimization efficiently tuning a live machine learning system by combining offline and online experiments.
Subjects: Machine Learning (stat.ML); Machine Learning (cs.LG)
Cite as: arXiv:1904.01049 [stat.ML]
  (or arXiv:1904.01049v2 [stat.ML] for this version)

Submission history

From: Ben Letham [view email]
[v1] Mon, 1 Apr 2019 18:19:11 GMT (1967kb,D)
[v2] Mon, 29 Apr 2019 16:38:06 GMT (1967kb,D)

Link back to: arXiv, form interface, contact.