Hiding in the Crowd: Federated Data Augmentation for On-Device Learning

作     者：Jeong, Eunjeong Oh, Seungeun Park, Jihong Kim, Hyesung Bennis, Mehdi Kim, Seong-Lyun 

作者机构：Yonsei Univ Seoul 03722 South Korea Deakin Univ Sch IT Waurn Ponds Vic 3216 Australia Samsung Res Seoul 06765 South Korea Univ Oulu Ctr Wireless Commun Oulu 90570 Finland Yonsei Univ Sch Elect & Elect Engn Seoul 03722 South Korea 

出 版 物：《IEEE INTELLIGENT SYSTEMS》 (IEEE智能系统)

年 卷 期：2021年第36卷第5期

页      面：80-86页

核心收录：

学科分类：0808[工学-电气工程] 08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：Bio-Mimetic Robot Research Center - Defense Acquisition Program Administration Agency for Defense Development [UD190018ID] 

主　　题：Servers Privacy Data models Training Intelligent systems Distributed databases Generators Machine learning Distributed networks Distributed artificial intelligence Wireless communication 

摘      要：To cope with the lack of on-device machine learning samples, this article presents a distributed data augmentation algorithm, coined federated data augmentation (FAug). In FAug, devices share a tiny fraction of their local data, i.e., seed samples, and collectively train a synthetic sample generator that can augment the local datasets of devices. To further improve FAug, we introduce a multihop-based seed sample collection method and an oversampling technique that mixes up collected seed samples. Both approaches enjoy the benefit from the crowd of devices, by hiding data privacy from preceding hops and feeding diverse seed samples. In the image classification tasks, simulations demonstrate that the proposed FAug frameworks yield stronger privacy guarantees, lower communication latency, and higher on-device ML accuracy.