Variational Bayesian optimal experimental design
Bayesian optimal experimental design (BOED) is a principled framework for making efficient use of limited experimental resources. Unfortunately, its applicability is hampered by the difficulty of obtaining accurate estimates of the expected information gain (EIG) of an experiment. To address this, w...
| Autori principali: | , , , , , , |
|---|---|
| Natura: | Conference item |
| Pubblicazione: |
Conference on Neural Information Processing Systems
2019
|
| _version_ | 1826276298112106496 |
|---|---|
| author | Foster, A Jankowiak, M Bingham, E Horsfall, P Tee, YW Rainforth, T Goodman, N |
| author_facet | Foster, A Jankowiak, M Bingham, E Horsfall, P Tee, YW Rainforth, T Goodman, N |
| author_sort | Foster, A |
| collection | OXFORD |
| description | Bayesian optimal experimental design (BOED) is a principled framework for making efficient use of limited experimental resources. Unfortunately, its applicability is hampered by the difficulty of obtaining accurate estimates of the expected information gain (EIG) of an experiment. To address this, we introduce several classes of fast EIG estimators by building on ideas from amortized variational inference. We show theoretically and empirically that these estimators can provide significant gains in speed and accuracy over previous approaches. We further demonstrate the practicality of our approach on a number of end-to-end experiments. |
| first_indexed | 2024-03-06T23:11:51Z |
| format | Conference item |
| id | oxford-uuid:65c01da9-4ee3-46ff-951a-d212be4c21df |
| institution | University of Oxford |
| last_indexed | 2024-03-06T23:11:51Z |
| publishDate | 2019 |
| publisher | Conference on Neural Information Processing Systems |
| record_format | dspace |
| spelling | oxford-uuid:65c01da9-4ee3-46ff-951a-d212be4c21df2022-03-26T18:27:32ZVariational Bayesian optimal experimental designConference itemhttp://purl.org/coar/resource_type/c_5794uuid:65c01da9-4ee3-46ff-951a-d212be4c21dfSymplectic ElementsConference on Neural Information Processing Systems2019Foster, AJankowiak, MBingham, EHorsfall, PTee, YWRainforth, TGoodman, NBayesian optimal experimental design (BOED) is a principled framework for making efficient use of limited experimental resources. Unfortunately, its applicability is hampered by the difficulty of obtaining accurate estimates of the expected information gain (EIG) of an experiment. To address this, we introduce several classes of fast EIG estimators by building on ideas from amortized variational inference. We show theoretically and empirically that these estimators can provide significant gains in speed and accuracy over previous approaches. We further demonstrate the practicality of our approach on a number of end-to-end experiments. |
| spellingShingle | Foster, A Jankowiak, M Bingham, E Horsfall, P Tee, YW Rainforth, T Goodman, N Variational Bayesian optimal experimental design |
| title | Variational Bayesian optimal experimental design |
| title_full | Variational Bayesian optimal experimental design |
| title_fullStr | Variational Bayesian optimal experimental design |
| title_full_unstemmed | Variational Bayesian optimal experimental design |
| title_short | Variational Bayesian optimal experimental design |
| title_sort | variational bayesian optimal experimental design |
| work_keys_str_mv | AT fostera variationalbayesianoptimalexperimentaldesign AT jankowiakm variationalbayesianoptimalexperimentaldesign AT binghame variationalbayesianoptimalexperimentaldesign AT horsfallp variationalbayesianoptimalexperimentaldesign AT teeyw variationalbayesianoptimalexperimentaldesign AT rainfortht variationalbayesianoptimalexperimentaldesign AT goodmann variationalbayesianoptimalexperimentaldesign |