Provable Sample-Efficient Transfer Learning Conditional Diffusion Models via Representation Learning
Ziheng Cheng Tianyu Xie Shiyue Zhang Cheng Zhang · 2025
Read Provable Sample-Efficient Transfer Learning Conditional Diffusion Models via Representation Learning by Ziheng Cheng Tianyu Xie Shiyue Zhang Cheng Zhang (2025) 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-2502-04491.txt.
Description
Ziheng Cheng, Tianyu Xie, Shiyue Zhang, and Cheng Zhang give transfer learning for diffusion models a theoretical footing in a 2025 paper. Diffusion models generate impressive content but need huge data to train from scratch, which many tasks lack. Borrowing from related tasks helps in practice, yet nobody had proved why for conditional diffusion models.
The authors assume all tasks share a low dimensional representation of their conditions, echoing real training recipes. They show that a good representation learned from source tasks cuts the data the target task needs by a lot. Real world application checks and numerical experiments support the math. The paper is a first step toward principled data saving for generative models.
Catalog record
| CREATOR | Ziheng Cheng Tianyu Xie Shiyue Zhang Cheng Zhang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 178 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-04491