Summary of Sfprompt: Communication-efficient Split Federated Fine-tuning For Large Pre-trained Models Over Resource-limited Devices, by Linxiao Cao et al.
SFPrompt: Communication-Efficient Split Federated Fine-Tuning for Large Pre-Trained Models over Resource-Limited Devices
by Linxiao Cao, Yifei Zhu, Wei Gong
First submitted to arxiv on: 24 Jul 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper introduces SFPrompt, an innovative privacy-preserving fine-tuning method for large pre-trained models in a federated setting where data cannot be directly uploaded due to privacy concerns and local devices are resource-constrained. The approach combines split learning with federated learning to handle these challenges. Specifically, the model is partitioned into client and server components, reducing computational demands on local resources. Soft prompts are introduced to enhance fine-tuning performance, and novel algorithms for dataset pruning and local-loss updates are devised to minimize communication costs. Experimental results demonstrate that SFPrompt achieves competitive performance with federated full fine-tuning while using only 0.46% of local computing resources and reducing communication costs by 53%. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about a new way to help big models learn from smaller amounts of data without sharing the data itself. Right now, big models are super powerful but need lots of training data to work well. However, this can be a problem when we want to use these models for different tasks because it’s not always possible or safe to share all the data. The researchers came up with a solution that divides the model into smaller parts and lets each part learn from its own small amount of data. This helps reduce the need for big computers and also keeps the data private. They tested this method and found that it works just as well as other methods but uses much fewer resources. |
Keywords
* Artificial intelligence * Federated learning * Fine tuning * Pruning
