pMixFed: Efficient Personalized Federated Learning through Adaptive Layer-Wise Mixup
Yasaman Saadati [email protected], Mohammad [email protected], M. Hadi Amini [email protected] · 2025
Read pMixFed: Efficient Personalized Federated Learning through Adaptive Layer-Wise Mixup by Yasaman Saadati [email protected], Mohammad [email protected], M. Hadi Amini [email protected] (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-2501-11002.txt.
Description
Federated learning trains one model across many phones or hospitals without moving their data, but one global model fits nobody whose data differ. Yasaman Saadati and two co-authors propose pMixFed in 2025, which blends each client's global and personal models layer by layer, mixing them in the style of the mixup training trick. Which layers stay personal adapts as training proceeds.
The blending eases the drift between local and global models and softens the forgetting that plagues the method. Gradual personalization lets clients settle into their share of the model. Across varied uneven data splits, the authors report faster training and hardier accuracy than rival personalized methods. Personalization, they argue, should be a dial, not a switch.
Catalog record
| CREATOR | Yasaman Saadati [email protected], Mohammad [email protected], M. Hadi Amini [email protected] |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 99 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-11002