Summary of Towards Federated Learning with On-device Training and Communication in 8-bit Floating Point, by Bokun Wang and Axel Berg and Durmus Alp Emre Acar and Chuteng Zhou
Towards Federated Learning with On-device Training and Communication in 8-bit Floating Point
by Bokun Wang, Axel Berg, Durmus Alp Emre Acar, Chuteng Zhou
First submitted to arxiv on: 2 Jul 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: 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 A novel approach to federated learning, leveraging 8-bit floating point (FP8) for training neural networks on edge devices, is presented in this study. The use of FP8 reduces both computational overhead and client-server communication costs due to weight compression. A new method for combining FP8 client training with a global FP32 server model is introduced, accompanied by convergence analysis. Experimental results demonstrate consistent communication reductions of at least 2.9x across various tasks and models compared to an FP32 baseline. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Researchers have discovered that using 8-bit floating point (FP8) can train neural networks quickly while saving computer power. This study shows how to use FP8 for training on devices at the edge, reducing both computer work and communication between devices. The researchers also created a new way to combine FP8 device training with a global server model that uses full 32-bit precision. Tests showed that this approach saved communication by at least 2.9 times compared to using full 32-bit precision. |
Keywords
» Artificial intelligence » Federated learning » Precision
