Bayesian Optimization with a Prior for the Optimum
(2021) European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2021 In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 12977 LNAI. p.265-296- Abstract
While Bayesian Optimization (BO) is a very popular method for optimizing expensive black-box functions, it fails to leverage the experience of domain experts. This causes BO to waste function evaluations on bad design choices (e.g., machine learning hyperparameters) that the expert already knows to work poorly. To address this issue, we introduce Bayesian Optimization with a Prior for the Optimum (BOPrO). BOPrO allows users to inject their knowledge into the optimization process in the form of priors about which parts of the input space will yield the best performance, rather than BO’s standard priors over functions, which are much less intuitive for users. BOPrO then combines these priors with BO’s standard probabilistic model to form... (More)
While Bayesian Optimization (BO) is a very popular method for optimizing expensive black-box functions, it fails to leverage the experience of domain experts. This causes BO to waste function evaluations on bad design choices (e.g., machine learning hyperparameters) that the expert already knows to work poorly. To address this issue, we introduce Bayesian Optimization with a Prior for the Optimum (BOPrO). BOPrO allows users to inject their knowledge into the optimization process in the form of priors about which parts of the input space will yield the best performance, rather than BO’s standard priors over functions, which are much less intuitive for users. BOPrO then combines these priors with BO’s standard probabilistic model to form a pseudo-posterior used to select which points to evaluate next. We show that BOPrO is around 6.67 × faster than state-of-the-art methods on a common suite of benchmarks, and achieves a new state-of-the-art performance on a real-world hardware design application. We also show that BOPrO converges faster even if the priors for the optimum are not entirely accurate and that it robustly recovers from misleading priors.
(Less)
- author
- Souza, Artur ; Nardi, Luigi LU ; Oliveira, Leonardo B. ; Olukotun, Kunle ; Lindauer, Marius and Hutter, Frank
- organization
- publishing date
- 2021
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- host publication
- Machine Learning and Knowledge Discovery in Databases. Research Track - European Conference, ECML PKDD 2021, Proceedings
- series title
- Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
- editor
- Oliver, Nuria ; Pérez-Cruz, Fernando ; Kramer, Stefan ; Read, Jesse and Lozano, Jose A.
- volume
- 12977 LNAI
- pages
- 32 pages
- publisher
- Springer Science and Business Media B.V.
- conference name
- European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML PKDD 2021
- conference location
- Virtual, Online
- conference dates
- 2021-09-13 - 2021-09-17
- external identifiers
-
- scopus:85115712403
- ISSN
- 1611-3349
- 0302-9743
- ISBN
- 9783030865221
- DOI
- 10.1007/978-3-030-86523-8_17
- language
- English
- LU publication?
- yes
- id
- fc65e94b-0ef4-4c84-a5ba-f9d93cb6d89a
- date added to LUP
- 2021-10-08 14:29:37
- date last changed
- 2025-02-09 17:40:20
@inproceedings{fc65e94b-0ef4-4c84-a5ba-f9d93cb6d89a, abstract = {{<p>While Bayesian Optimization (BO) is a very popular method for optimizing expensive black-box functions, it fails to leverage the experience of domain experts. This causes BO to waste function evaluations on bad design choices (e.g., machine learning hyperparameters) that the expert already knows to work poorly. To address this issue, we introduce Bayesian Optimization with a Prior for the Optimum (BOPrO). BOPrO allows users to inject their knowledge into the optimization process in the form of priors about which parts of the input space will yield the best performance, rather than BO’s standard priors over functions, which are much less intuitive for users. BOPrO then combines these priors with BO’s standard probabilistic model to form a pseudo-posterior used to select which points to evaluate next. We show that BOPrO is around 6.67 × faster than state-of-the-art methods on a common suite of benchmarks, and achieves a new state-of-the-art performance on a real-world hardware design application. We also show that BOPrO converges faster even if the priors for the optimum are not entirely accurate and that it robustly recovers from misleading priors.</p>}}, author = {{Souza, Artur and Nardi, Luigi and Oliveira, Leonardo B. and Olukotun, Kunle and Lindauer, Marius and Hutter, Frank}}, booktitle = {{Machine Learning and Knowledge Discovery in Databases. Research Track - European Conference, ECML PKDD 2021, Proceedings}}, editor = {{Oliver, Nuria and Pérez-Cruz, Fernando and Kramer, Stefan and Read, Jesse and Lozano, Jose A.}}, isbn = {{9783030865221}}, issn = {{1611-3349}}, language = {{eng}}, pages = {{265--296}}, publisher = {{Springer Science and Business Media B.V.}}, series = {{Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)}}, title = {{Bayesian Optimization with a Prior for the Optimum}}, url = {{http://dx.doi.org/10.1007/978-3-030-86523-8_17}}, doi = {{10.1007/978-3-030-86523-8_17}}, volume = {{12977 LNAI}}, year = {{2021}}, }