Approximate Bayesian computation with path signatures
Simulation models often lack tractable likelihood functions, making likelihood-free inference methods indispensable. Approximate Bayesian computation generates likelihood-free posterior samples by comparing simulated and observed data through some distance measure, but existing approaches are often...
| Những tác giả chính: | , , |
|---|---|
| Định dạng: | Conference item |
| Ngôn ngữ: | English |
| Được phát hành: |
Proceedings of Machine Learning Research
2024
|
| _version_ | 1826315067113603072 |
|---|---|
| author | Dyer, J Cannon, P Schmon, SM |
| author_facet | Dyer, J Cannon, P Schmon, SM |
| author_sort | Dyer, J |
| collection | OXFORD |
| description | Simulation models often lack tractable likelihood functions, making likelihood-free inference methods indispensable. Approximate Bayesian computation generates likelihood-free posterior samples by comparing simulated and observed data through some distance measure, but existing approaches are often poorly suited to time series simulators, for example due to an independent and identically distributed data assumption. In this paper, we propose to use path signatures in approximate Bayesian computation to handle the sequential nature of time series. We provide theoretical guarantees on the resultant posteriors and demonstrate competitive Bayesian parameter inference for simulators generating univariate, multivariate, and irregularly spaced sequences of non-iid data. |
| first_indexed | 2024-12-09T03:17:26Z |
| format | Conference item |
| id | oxford-uuid:c8475c0c-e38f-4200-b7a3-e58c265e4d43 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-12-09T03:17:26Z |
| publishDate | 2024 |
| publisher | Proceedings of Machine Learning Research |
| record_format | dspace |
| spelling | oxford-uuid:c8475c0c-e38f-4200-b7a3-e58c265e4d432024-10-24T17:29:07ZApproximate Bayesian computation with path signaturesConference itemhttp://purl.org/coar/resource_type/c_5794uuid:c8475c0c-e38f-4200-b7a3-e58c265e4d43EnglishSymplectic ElementsProceedings of Machine Learning Research2024Dyer, JCannon, PSchmon, SMSimulation models often lack tractable likelihood functions, making likelihood-free inference methods indispensable. Approximate Bayesian computation generates likelihood-free posterior samples by comparing simulated and observed data through some distance measure, but existing approaches are often poorly suited to time series simulators, for example due to an independent and identically distributed data assumption. In this paper, we propose to use path signatures in approximate Bayesian computation to handle the sequential nature of time series. We provide theoretical guarantees on the resultant posteriors and demonstrate competitive Bayesian parameter inference for simulators generating univariate, multivariate, and irregularly spaced sequences of non-iid data. |
| spellingShingle | Dyer, J Cannon, P Schmon, SM Approximate Bayesian computation with path signatures |
| title | Approximate Bayesian computation with path signatures |
| title_full | Approximate Bayesian computation with path signatures |
| title_fullStr | Approximate Bayesian computation with path signatures |
| title_full_unstemmed | Approximate Bayesian computation with path signatures |
| title_short | Approximate Bayesian computation with path signatures |
| title_sort | approximate bayesian computation with path signatures |
| work_keys_str_mv | AT dyerj approximatebayesiancomputationwithpathsignatures AT cannonp approximatebayesiancomputationwithpathsignatures AT schmonsm approximatebayesiancomputationwithpathsignatures |