Towards Federated Learning with on-device Training and Communication in 8-bit Floating Point
(2024) FedKDD: International Joint Workshop on Federated Learning for Data Mining and Graph Analytics- Abstract
- Recent work has shown that 8-bit floating point (FP8) can be used for efficiently training neural networks with reduced computational overhead compared to training in FP32/FP16. In this work, we investigate the use of FP8 training in a federated learning context. This brings not only the usual benefits of FP8 which are desirable for on-device training at the edge, but also reduces client-server
communication costs due to significant weight compression. We present a novel method for combining FP8 client training while maintaining a global FP32 server model and provide convergence analysis. Experiments with various machine learning models and datasets show that our method consistently yields communication reductions of at least 2.9x... (More) - Recent work has shown that 8-bit floating point (FP8) can be used for efficiently training neural networks with reduced computational overhead compared to training in FP32/FP16. In this work, we investigate the use of FP8 training in a federated learning context. This brings not only the usual benefits of FP8 which are desirable for on-device training at the edge, but also reduces client-server
communication costs due to significant weight compression. We present a novel method for combining FP8 client training while maintaining a global FP32 server model and provide convergence analysis. Experiments with various machine learning models and datasets show that our method consistently yields communication reductions of at least 2.9x across a variety of tasks and models compared to an FP32 baseline.
(Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/record/a92e2aa8-8bcb-45ae-b2ee-ec7275da1399
- author
- Wang, Bokun ; Berg, Axel LU ; Acar, Durmus Alp Emre and Zhou, Chuteng
- organization
- publishing date
- 2024-08-26
- type
- Contribution to conference
- publication status
- published
- subject
- keywords
- federated learning, quantization, fp8
- conference name
- FedKDD: International Joint Workshop on Federated Learning for Data Mining and Graph Analytics
- conference location
- Barcelona, Spain
- conference dates
- 2024-08-26 - 2024-08-26
- language
- English
- LU publication?
- yes
- id
- a92e2aa8-8bcb-45ae-b2ee-ec7275da1399
- alternative location
- https://openreview.net/pdf?id=YfGoRu5gR6
- date added to LUP
- 2024-09-03 07:36:40
- date last changed
- 2024-10-03 10:34:32
@misc{a92e2aa8-8bcb-45ae-b2ee-ec7275da1399, abstract = {{Recent work has shown that 8-bit floating point (FP8) can be used for efficiently training neural networks with reduced computational overhead compared to training in FP32/FP16. In this work, we investigate the use of FP8 training in a federated learning context. This brings not only the usual benefits of FP8 which are desirable for on-device training at the edge, but also reduces client-server<br/>communication costs due to significant weight compression. We present a novel method for combining FP8 client training while maintaining a global FP32 server model and provide convergence analysis. Experiments with various machine learning models and datasets show that our method consistently yields communication reductions of at least 2.9x across a variety of tasks and models compared to an FP32 baseline.<br/>}}, author = {{Wang, Bokun and Berg, Axel and Acar, Durmus Alp Emre and Zhou, Chuteng}}, keywords = {{federated learning; quantization; fp8}}, language = {{eng}}, month = {{08}}, title = {{Towards Federated Learning with on-device Training and Communication in 8-bit Floating Point}}, url = {{https://openreview.net/pdf?id=YfGoRu5gR6}}, year = {{2024}}, }