Differentiable particle filtering via entropy-regularized optimal transport
Particle Filtering (PF) methods are an established class of procedures for performing inference in non-linear state-space models. Resampling is a key ingredient of PF necessary to obtain low variance likelihood and states estimates. However, traditional resampling methods result in PF-based loss fun...
| Main Authors: | , , , |
|---|---|
| Format: | Conference item |
| Language: | English |
| Published: |
Journal of Machine Learning Research
2021
|
| _version_ | 1826300517433737216 |
|---|---|
| author | Corenflos, A Thornton, J Deligiannidis, G Doucet, A |
| author_facet | Corenflos, A Thornton, J Deligiannidis, G Doucet, A |
| author_sort | Corenflos, A |
| collection | OXFORD |
| description | Particle Filtering (PF) methods are an established class of procedures for performing inference in non-linear state-space models. Resampling is a key ingredient of PF necessary to obtain low variance likelihood and states estimates. However, traditional resampling methods result in PF-based loss functions being non-differentiable with respect to model and PF parameters. In a variational inference context, resampling also yields high variance gradient estimates of the PF-based evidence lower bound. By leveraging optimal transport ideas, we introduce a principled differentiable particle filter and provide convergence results. We demonstrate this novel method on a variety of applications. |
| first_indexed | 2024-03-07T05:18:22Z |
| format | Conference item |
| id | oxford-uuid:de06530c-9226-4bbf-b088-33002410cc25 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T05:18:22Z |
| publishDate | 2021 |
| publisher | Journal of Machine Learning Research |
| record_format | dspace |
| spelling | oxford-uuid:de06530c-9226-4bbf-b088-33002410cc252022-03-27T09:29:09ZDifferentiable particle filtering via entropy-regularized optimal transportConference itemhttp://purl.org/coar/resource_type/c_5794uuid:de06530c-9226-4bbf-b088-33002410cc25EnglishSymplectic ElementsJournal of Machine Learning Research2021Corenflos, AThornton, JDeligiannidis, GDoucet, AParticle Filtering (PF) methods are an established class of procedures for performing inference in non-linear state-space models. Resampling is a key ingredient of PF necessary to obtain low variance likelihood and states estimates. However, traditional resampling methods result in PF-based loss functions being non-differentiable with respect to model and PF parameters. In a variational inference context, resampling also yields high variance gradient estimates of the PF-based evidence lower bound. By leveraging optimal transport ideas, we introduce a principled differentiable particle filter and provide convergence results. We demonstrate this novel method on a variety of applications. |
| spellingShingle | Corenflos, A Thornton, J Deligiannidis, G Doucet, A Differentiable particle filtering via entropy-regularized optimal transport |
| title | Differentiable particle filtering via entropy-regularized optimal transport |
| title_full | Differentiable particle filtering via entropy-regularized optimal transport |
| title_fullStr | Differentiable particle filtering via entropy-regularized optimal transport |
| title_full_unstemmed | Differentiable particle filtering via entropy-regularized optimal transport |
| title_short | Differentiable particle filtering via entropy-regularized optimal transport |
| title_sort | differentiable particle filtering via entropy regularized optimal transport |
| work_keys_str_mv | AT corenflosa differentiableparticlefilteringviaentropyregularizedoptimaltransport AT thorntonj differentiableparticlefilteringviaentropyregularizedoptimaltransport AT deligiannidisg differentiableparticlefilteringviaentropyregularizedoptimaltransport AT douceta differentiableparticlefilteringviaentropyregularizedoptimaltransport |