Skipping the Zeros in Diffusion Models for Sparse Data Generation
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2026
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866913163810701312 |
|---|---|
| author | Ostheimer, Phil Sidney Nagda, Mayank Balinskyy, Andriy Rodrigues, Gabriel Vicente Radig, Jean Herrmann, Carl Mandt, Stephan Kloft, Marius Fellenz, Sophie |
| author_facet | Ostheimer, Phil Sidney Nagda, Mayank Balinskyy, Andriy Rodrigues, Gabriel Vicente Radig, Jean Herrmann, Carl Mandt, Stephan Kloft, Marius Fellenz, Sophie |
| contents | Diffusion models (DMs) excel on dense continuous data, but are not designed for sparse continuous data. They do not model exact zeros that represent the deliberate absence of a signal. As a result, they erase sparsity patterns and perform unnecessary computation on mostly zero entries. With Sparsity-Exploiting Diffusion (SED), we model only non-zero values, preserving sparsity. SED delivers computational savings while maintaining or improving generation quality by skipping zeros during training and inference. Across physics and biology benchmarks, SED matches or surpasses conventional DMs and domain-specific baselines, while vision experiments provide intuitive insights into the limitations of dense DMs and the benefits of SED. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_01817 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Skipping the Zeros in Diffusion Models for Sparse Data Generation Ostheimer, Phil Sidney Nagda, Mayank Balinskyy, Andriy Rodrigues, Gabriel Vicente Radig, Jean Herrmann, Carl Mandt, Stephan Kloft, Marius Fellenz, Sophie Machine Learning Diffusion models (DMs) excel on dense continuous data, but are not designed for sparse continuous data. They do not model exact zeros that represent the deliberate absence of a signal. As a result, they erase sparsity patterns and perform unnecessary computation on mostly zero entries. With Sparsity-Exploiting Diffusion (SED), we model only non-zero values, preserving sparsity. SED delivers computational savings while maintaining or improving generation quality by skipping zeros during training and inference. Across physics and biology benchmarks, SED matches or surpasses conventional DMs and domain-specific baselines, while vision experiments provide intuitive insights into the limitations of dense DMs and the benefits of SED. |
| title | Skipping the Zeros in Diffusion Models for Sparse Data Generation |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2605.01817 |