A continuous time framework for discrete denoising models
We provide the first complete continuous time framework for denoising diffusion models of discrete data. This is achieved by formulating the forward noising process and corresponding reverse time generative process as Continuous Time Markov Chains (CTMCs). The model can be efficiently trained using...
| Main Authors: | , , , , , |
|---|---|
| Format: | Conference item |
| Language: | English |
| Published: |
Curran Associates
2023
|
| _version_ | 1826311192152375296 |
|---|---|
| author | Campbell, A Benton, J De Bortoli, V Rainforth, T Deligiannidis, G Doucet, A |
| author_facet | Campbell, A Benton, J De Bortoli, V Rainforth, T Deligiannidis, G Doucet, A |
| author_sort | Campbell, A |
| collection | OXFORD |
| description | We provide the first complete continuous time framework for denoising diffusion models of discrete data. This is achieved by formulating the forward noising process and corresponding reverse time generative process as Continuous Time Markov Chains (CTMCs). The model can be efficiently trained using a continuous time version of the ELBO. We simulate the high dimensional CTMC using techniques developed in chemical physics and exploit our continuous time framework to derive high performance samplers that we show can outperform discrete time methods for discrete data. The continuous time treatment also enables us to derive a novel theoretical result bounding the error between the generated sample distribution and the true data distribution. |
| first_indexed | 2024-03-07T08:04:43Z |
| format | Conference item |
| id | oxford-uuid:28af9179-92c0-47ea-8709-7e2baafd2a4e |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-03-07T08:04:43Z |
| publishDate | 2023 |
| publisher | Curran Associates |
| record_format | dspace |
| spelling | oxford-uuid:28af9179-92c0-47ea-8709-7e2baafd2a4e2023-10-30T09:35:52ZA continuous time framework for discrete denoising modelsConference itemhttp://purl.org/coar/resource_type/c_5794uuid:28af9179-92c0-47ea-8709-7e2baafd2a4eEnglishSymplectic Elements Curran Associates2023Campbell, ABenton, JDe Bortoli, VRainforth, TDeligiannidis, GDoucet, AWe provide the first complete continuous time framework for denoising diffusion models of discrete data. This is achieved by formulating the forward noising process and corresponding reverse time generative process as Continuous Time Markov Chains (CTMCs). The model can be efficiently trained using a continuous time version of the ELBO. We simulate the high dimensional CTMC using techniques developed in chemical physics and exploit our continuous time framework to derive high performance samplers that we show can outperform discrete time methods for discrete data. The continuous time treatment also enables us to derive a novel theoretical result bounding the error between the generated sample distribution and the true data distribution. |
| spellingShingle | Campbell, A Benton, J De Bortoli, V Rainforth, T Deligiannidis, G Doucet, A A continuous time framework for discrete denoising models |
| title | A continuous time framework for discrete denoising models |
| title_full | A continuous time framework for discrete denoising models |
| title_fullStr | A continuous time framework for discrete denoising models |
| title_full_unstemmed | A continuous time framework for discrete denoising models |
| title_short | A continuous time framework for discrete denoising models |
| title_sort | continuous time framework for discrete denoising models |
| work_keys_str_mv | AT campbella acontinuoustimeframeworkfordiscretedenoisingmodels AT bentonj acontinuoustimeframeworkfordiscretedenoisingmodels AT debortoliv acontinuoustimeframeworkfordiscretedenoisingmodels AT rainfortht acontinuoustimeframeworkfordiscretedenoisingmodels AT deligiannidisg acontinuoustimeframeworkfordiscretedenoisingmodels AT douceta acontinuoustimeframeworkfordiscretedenoisingmodels AT campbella continuoustimeframeworkfordiscretedenoisingmodels AT bentonj continuoustimeframeworkfordiscretedenoisingmodels AT debortoliv continuoustimeframeworkfordiscretedenoisingmodels AT rainfortht continuoustimeframeworkfordiscretedenoisingmodels AT deligiannidisg continuoustimeframeworkfordiscretedenoisingmodels AT douceta continuoustimeframeworkfordiscretedenoisingmodels |