Constrained Particle Seeking: Solving Diffusion Inverse Problems with Just Forward Passes
Hongkun Dou, Zike Chen, Zeyu Li, Hongjue Li, Lijun Yang, Yue Deng · 2026
Read Constrained Particle Seeking: Solving Diffusion Inverse Problems with Just Forward Passes by Hongkun Dou, Zike Chen, Zeyu Li, Hongjue Li, Lijun Yang, Yue Deng (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-01837.txt.
Description
Hongkun Dou, Zike Chen, Zeyu Li, and colleagues solve inverse problems with diffusion models and no gradient steps. Their 2026 work covers tasks like rebuilding an image from partial or noisy measurements. Most diffusion solvers need full knowledge of the measurement process to compute guiding gradients, which many real cases lack.
Their method, Constrained Particle Seeking, searches actively among candidate samples, called particles, for the best one. Constraints steer the search toward high-density regions of the learned prior, where plausible answers live. Only forward passes through the observation model are needed, so no gradient of that model is required. On image and scientific problems, results match gradient-based solvers in quality. Gradient-free rivals trail by a wide margin in the same tests, which makes the method handy where measurement models are opaque.
Catalog record
| CREATOR | Hongkun Dou, Zike Chen, Zeyu Li, Hongjue Li, Lijun Yang, Yue Deng |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 52190 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.01837