Federated Learning with Personalization Layers

• DOI: 10.48550/arxiv.1912.00818
• Publication Date: 2019-01-01
• Authors: Manoj Ghuhan Ariv...; Vinay K. Aggarwal; Aaditya Kumar Singh; Sunav Choudhary
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Computer Science - Distributed, Parallel, and Cluster Computing; Statistics - Machine Learning; 0202 electrical engineering, electronic engineering, information engineering; Machine Learning (stat.ML); 02 engineering and technology; Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)

The emerging paradigm of federated learning strives to enable collaborative training machine models on the network edge without centrally aggregating raw data and hence, improving privacy. This sharply deviates from traditional necessitates design algorithms robust various sources heterogeneity. Specifically, statistical heterogeneity across user devices can severely degrade performance standard averaging for applications like personalization with deep learning. paper pro-posesFedPer, a base + layer approach feedforward neural networks, which combat ill-effects We demonstrate effectiveness ofFedPerfor non-identical partitions ofCIFARdatasetsand personalized image aesthetics dataset Flickr.