GLASSES: Relieving the myopia of Bayesian optimisation
We present glasses: Global optimisation with Look-Ahead through Stochastic Simulation and Expected-loss Search. The majority of global optimisation approaches in use are myopic, in only considering the impact of the next function value; the non-myopic approaches that do exist are able to consider on...
| Main Authors: | , , |
|---|---|
| Format: | Conference item |
| Published: |
Journal of Machine Learning Research
2016
|
| _version_ | 1826269010278219776 |
|---|---|
| author | Gonzalez, J Osborne, M Lawrence, N |
| author_facet | Gonzalez, J Osborne, M Lawrence, N |
| author_sort | Gonzalez, J |
| collection | OXFORD |
| description | We present glasses: Global optimisation with Look-Ahead through Stochastic Simulation and Expected-loss Search. The majority of global optimisation approaches in use are myopic, in only considering the impact of the next function value; the non-myopic approaches that do exist are able to consider only a handful of future evaluations. Our novel algorithm, glasses, permits the consideration of dozens of evaluations into the future. This is done by approximating the ideal look-ahead loss function, which is expensive to evaluate, by a cheaper alternative in which the future steps of the algorithm are simulated beforehand. An Expectation Propagation algorithm is used to compute the expected value of the loss. We show that the far-horizon planning thus enabled leads to substantive performance gains in empirical tests. |
| first_indexed | 2024-03-06T21:18:20Z |
| format | Conference item |
| id | oxford-uuid:40919a8e-fcd4-4ee2-80cf-9a9b77df23c5 |
| institution | University of Oxford |
| last_indexed | 2024-03-06T21:18:20Z |
| publishDate | 2016 |
| publisher | Journal of Machine Learning Research |
| record_format | dspace |
| spelling | oxford-uuid:40919a8e-fcd4-4ee2-80cf-9a9b77df23c52022-03-26T14:38:42ZGLASSES: Relieving the myopia of Bayesian optimisationConference itemhttp://purl.org/coar/resource_type/c_5794uuid:40919a8e-fcd4-4ee2-80cf-9a9b77df23c5Symplectic Elements at OxfordJournal of Machine Learning Research2016Gonzalez, JOsborne, MLawrence, NWe present glasses: Global optimisation with Look-Ahead through Stochastic Simulation and Expected-loss Search. The majority of global optimisation approaches in use are myopic, in only considering the impact of the next function value; the non-myopic approaches that do exist are able to consider only a handful of future evaluations. Our novel algorithm, glasses, permits the consideration of dozens of evaluations into the future. This is done by approximating the ideal look-ahead loss function, which is expensive to evaluate, by a cheaper alternative in which the future steps of the algorithm are simulated beforehand. An Expectation Propagation algorithm is used to compute the expected value of the loss. We show that the far-horizon planning thus enabled leads to substantive performance gains in empirical tests. |
| spellingShingle | Gonzalez, J Osborne, M Lawrence, N GLASSES: Relieving the myopia of Bayesian optimisation |
| title | GLASSES: Relieving the myopia of Bayesian optimisation |
| title_full | GLASSES: Relieving the myopia of Bayesian optimisation |
| title_fullStr | GLASSES: Relieving the myopia of Bayesian optimisation |
| title_full_unstemmed | GLASSES: Relieving the myopia of Bayesian optimisation |
| title_short | GLASSES: Relieving the myopia of Bayesian optimisation |
| title_sort | glasses relieving the myopia of bayesian optimisation |
| work_keys_str_mv | AT gonzalezj glassesrelievingthemyopiaofbayesianoptimisation AT osbornem glassesrelievingthemyopiaofbayesianoptimisation AT lawrencen glassesrelievingthemyopiaofbayesianoptimisation |