| |
| |
|
|
| cff-version: 1.2.0 |
| title: >- |
| Improving Single Noise Level Diffusion Samplers with |
| Restricted Gaussian Oracles |
| message: 'If you use this code for your projects, please cite:' |
| type: software |
| authors: |
| - given-names: Leello |
| family-names: Dadi |
| email: leello.dadi@epfl.ch |
| affiliation: 'EPFL STI IEM LIONS, Lausanne, Switzerland' |
| orcid: 'https://orcid.org/0000-0003-2580-4913' |
| - given-names: Andrej |
| family-names: Janchevski |
| email: andrej.janchevski@epfl.ch |
| affiliation: 'EPFL STI IEM LIONS, Lausanne, Switzerland' |
| orcid: 'https://orcid.org/0000-0001-9568-0966' |
| - given-names: Volkan |
| family-names: Cevher |
| email: volkan.cevher@epfl.ch |
| affiliation: 'EPFL STI IEM LIONS, Lausanne, Switzerland' |
| orcid: 'https://orcid.org/0000-0002-5004-201X' |
| identifiers: |
| - type: url |
| value: 'https://openreview.net/forum?id=xkiI5tou6J' |
| repository-code: >- |
| https://github.com/Bani57/multi-prox-diffusion-iclr-delta-2025 |
| abstract: >- |
| Diffusion models and diffusion Monte-Carlo schemes that |
| sample from unnormalized log-densities, both rely on |
| denoisers ( or score estimates) at different noise scales. |
| This complicates the sampling process as denoising |
| schedules require careful tuning and nested inner-MCMC |
| loops. In this work, we propose a single noise level |
| sampling procedure that only requires a single low-noise |
| denoiser. Our framework results from improvements we bring |
| to the multimeasurement Walk-Jump sampler of Saremi et al. |
| 2021 by mixing in ideas from the proximal sampler of Shen |
| et al. 2020. Our analysis shows that annealing (or |
| multiple noise scales) is unnecessary if one is willing to |
| pay an increased memory cost. We demonstrate this by |
| proposing an entirely log-concave sampling framework. |
| license: CC-BY-1.0 |
| preferred-citation: |
| type: conference-paper |
| title: 'Improving Single Noise Level Denoising Samplers with Restricted Gaussian Oracles' |
| collection-title: 'ICLR 2025 Workshop on Deep Generative Model in Machine Learning: Theory, Principle and Efficacy' |
| year: 2025 |
| url: 'https://openreview.net/forum?id=xkiI5tou6J' |
| authors: |
| - given-names: Leello |
| family-names: Dadi |
| email: leello.dadi@epfl.ch |
| affiliation: 'EPFL STI IEM LIONS, Lausanne, Switzerland' |
| orcid: 'https://orcid.org/0000-0003-2580-4913' |
| - given-names: Andrej |
| family-names: Janchevski |
| email: andrej.janchevski@epfl.ch |
| affiliation: 'EPFL STI IEM LIONS, Lausanne, Switzerland' |
| orcid: 'https://orcid.org/0000-0001-9568-0966' |
| - given-names: Volkan |
| family-names: Cevher |
| email: volkan.cevher@epfl.ch |
| affiliation: 'EPFL STI IEM LIONS, Lausanne, Switzerland' |
| orcid: 'https://orcid.org/0000-0002-5004-201X' |
