Bayesian Optimization in High Dimensions via Random Embeddings
Bayesian optimization techniques have been successfully applied to robotics, planning, sensor placement, recommendation, advertising, intelligent user interfaces and automatic algorithm configuration. Despite these successes, the approach is restricted to problems of moderate dimension, and several...
| Main Authors: | , , , , |
|---|---|
| Format: | Conference item |
| Published: |
2013
|
| _version_ | 1826276483392339968 |
|---|---|
| author | Wang, Z Zoghi, M Hutter, F Matheson, D de Freitas, N |
| author_facet | Wang, Z Zoghi, M Hutter, F Matheson, D de Freitas, N |
| author_sort | Wang, Z |
| collection | OXFORD |
| description | Bayesian optimization techniques have been successfully applied to robotics, planning, sensor placement, recommendation, advertising, intelligent user interfaces and automatic algorithm configuration. Despite these successes, the approach is restricted to problems of moderate dimension, and several workshops on Bayesian optimization have identified its scaling to high dimensions as one of the holy grails of the field. In this paper, we introduce a novel random embedding idea to attack this problem. The resulting Random EMbedding Bayesian Optimization (REMBO) algorithm is very simple and applies to domains with both categorical and continuous variables. The experiments demonstrate that REMBO can effectively solve high-dimensional problems, including automatic parameter configuration of a popular mixed integer linear programming solver. |
| first_indexed | 2024-03-06T23:14:38Z |
| format | Conference item |
| id | oxford-uuid:66b2e122-50db-4064-8b4b-be3a8ebfae3f |
| institution | University of Oxford |
| last_indexed | 2024-03-06T23:14:38Z |
| publishDate | 2013 |
| record_format | dspace |
| spelling | oxford-uuid:66b2e122-50db-4064-8b4b-be3a8ebfae3f2022-03-26T18:33:34ZBayesian Optimization in High Dimensions via Random EmbeddingsConference itemhttp://purl.org/coar/resource_type/c_5794uuid:66b2e122-50db-4064-8b4b-be3a8ebfae3fDepartment of Computer Science2013Wang, ZZoghi, MHutter, FMatheson, Dde Freitas, NBayesian optimization techniques have been successfully applied to robotics, planning, sensor placement, recommendation, advertising, intelligent user interfaces and automatic algorithm configuration. Despite these successes, the approach is restricted to problems of moderate dimension, and several workshops on Bayesian optimization have identified its scaling to high dimensions as one of the holy grails of the field. In this paper, we introduce a novel random embedding idea to attack this problem. The resulting Random EMbedding Bayesian Optimization (REMBO) algorithm is very simple and applies to domains with both categorical and continuous variables. The experiments demonstrate that REMBO can effectively solve high-dimensional problems, including automatic parameter configuration of a popular mixed integer linear programming solver. |
| spellingShingle | Wang, Z Zoghi, M Hutter, F Matheson, D de Freitas, N Bayesian Optimization in High Dimensions via Random Embeddings |
| title | Bayesian Optimization in High Dimensions via Random Embeddings |
| title_full | Bayesian Optimization in High Dimensions via Random Embeddings |
| title_fullStr | Bayesian Optimization in High Dimensions via Random Embeddings |
| title_full_unstemmed | Bayesian Optimization in High Dimensions via Random Embeddings |
| title_short | Bayesian Optimization in High Dimensions via Random Embeddings |
| title_sort | bayesian optimization in high dimensions via random embeddings |
| work_keys_str_mv | AT wangz bayesianoptimizationinhighdimensionsviarandomembeddings AT zoghim bayesianoptimizationinhighdimensionsviarandomembeddings AT hutterf bayesianoptimizationinhighdimensionsviarandomembeddings AT mathesond bayesianoptimizationinhighdimensionsviarandomembeddings AT defreitasn bayesianoptimizationinhighdimensionsviarandomembeddings |