Task-Aware Parameter-Efficient Fine-Tuning of Large Pre-Trained Models at the Edge
Senkang Hu ⋆,† , Yanan Ma ⋆,† , Yihang Tao ⋆,† , Zhengru Fang ⋆,† , Zihan Fang ⋆,† , Yiqin Deng ⋆,† , Sam Kwong ‡ , Yuguang Fang ⋆,† ⋆ Hong Kong JC STEM Lab of Smart City, Hong Kong SAR. † Department of Computer Science, City University of Hong Kong, Hong Kong SAR. ‡ Department of Computing and Deci · 2025
Read Task-Aware Parameter-Efficient Fine-Tuning of Large Pre-Trained Models at the Edge by Senkang Hu ⋆,† , Yanan Ma ⋆,† , Yihang Tao ⋆,† , Zhengru Fang ⋆,† , Zihan Fang ⋆,† , Yiqin Deng ⋆,† , Sam Kwong ‡ , Yuguang Fang ⋆,† ⋆ Hong Kong JC STEM Lab of Smart City, Hong Kong SAR. † Department of Computer Science, City University of Hong Kong, Hong Kong SAR. ‡ Department of Computing and Deci (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-2504-03718.txt.
Description
Eight authors led by Senkang Hu tune large AI models for small edge gadgets without the full cost of retraining. Edge gear lacks the memory, power, and chip strength that full fine-tuning demands. Their frame, TaskEdge, scores which model weights matter most for the target task, then updates only that sliver. The score weighs both the weights and the signals that flow through them, for a fairer pick.
Chosen weights are spread even across the model, not bunched in one zone, by a task-fit spread rule. Less than a tenth of one percent of weights change, yet task skill holds near full tune marks. The sparse layout also mates with speed chips and a famed low-rank add-on. Phone and sensor kit can thus host tuned models cheaply.
Catalog record
| CREATOR | Senkang Hu ⋆,† , Yanan Ma ⋆,† , Yihang Tao ⋆,† , Zhengru Fang ⋆,† , Zihan Fang ⋆,† , Yiqin Deng ⋆,† , Sam Kwong ‡ , Yuguang Fang ⋆,† ⋆ Hong Kong JC STEM Lab of Smart City, Hong Kong SAR. † Department of Computer Science, City University of Hong Kong, Hong Kong SAR. ‡ Department of Computing and Deci |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 28 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-03718