Analyzing and Improving Diffusion Models for Time-Series Data Imputation: A Proximal Recursion Perspective
Zhichao Chen, Hao Wang, Fangyikang Wang, Licheng Pan, Zhengnan Li, Yunfei Teng, Haoxuan Li, Zhouchen Lin · 2026
Read Analyzing and Improving Diffusion Models for Time-Series Data Imputation: A Proximal Recursion Perspective by Zhichao Chen, Hao Wang, Fangyikang Wang, Licheng Pan, Zhengnan Li, Yunfei Teng, Haoxuan Li, Zhouchen Lin (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-2602-01182.txt.
Description
Zhichao Chen and co-authors probe diffusion nets for time-series fill-ins and build SPIRIT to fix their wobble. Shifting time trends skew fill paths and favor wild outputs, while diffusion training prizes varied samples over true point scores. A proximal-operator read finds a hidden Wasserstein pull that blocks trend fixes and boosts variety at truth's cost. SPIRIT loosens mass-keeping with an entropy Bregman gap, forming a semi-proximal gap proven tough to trend shifts, then drops the wasteful part for a full fill plan.
Catalog record
| CREATOR | Zhichao Chen, Hao Wang, Fangyikang Wang, Licheng Pan, Zhengnan Li, Yunfei Teng, Haoxuan Li, Zhouchen Lin |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 125426 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602.01182