Summary of Selective Aggregation For Low-rank Adaptation in Federated Learning, by Pengxin Guo et al.
Selective Aggregation for Low-Rank Adaptation in Federated Learning
by Pengxin Guo, Shuang Zeng, Yanran Wang, Huijie Fan, Feifei Wang, Liangqiong Qu
First submitted to arxiv on: 2 Oct 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: None
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 We explore LoRA (Low-Rank Adaptation) in federated learning, focusing on the asymmetry analysis of learned matrices A and B. Our investigation reveals that A matrices learn general knowledge, while B matrices focus on capturing client-specific knowledge. Based on this finding, we introduce Federated Share-A Low-Rank Adaptation (FedSA-LoRA), which employs two low-rank trainable matrices A and B to model the weight update. Only A matrices are shared with the server for aggregation. We also analyze other LoRA variants, such as rsLoRA and VeRA, revealing a consistent pattern. Our method, FedSA-LoRA, is extended to these variants, resulting in FedSA-rsLoRA and FedSA-VeRA. This establishes a general paradigm for integrating LoRA with FL, offering guidance for future work on subsequent LoRA variants combined with FL. Experimental results on natural language understanding and generation tasks demonstrate the effectiveness of our proposed method. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary We’re studying how to improve a type of machine learning called federated learning. We found that two types of learned matrices, A and B, are doing different jobs. The A matrices learn general knowledge, while the B matrices learn specific things about individual clients. Based on this discovery, we created a new method that combines these matrices in a way that only shares some information with the server. We also looked at other similar methods and found they work similarly. Our new method can be used for different types of federated learning and works well for tasks like understanding and generating natural language. |
Keywords
» Artificial intelligence » Federated learning » Language understanding » Lora » Low rank adaptation » Machine learning
