Split learning for health: Distributed deep learning without sharing raw\n patient data

Abstract

Can health entities collaboratively train deep learning models without\nsharing sensitive raw data? This paper proposes several configurations of a\ndistributed deep learning method called SplitNN to facilitate such\ncollaborations. SplitNN does not share raw data or model details with\ncollaborating institutions. The proposed configurations of splitNN cater to\npractical settings of i) entities holding different modalities of patient data,\nii) centralized and local health entities collaborating on multiple tasks and\niii) learning without sharing labels. We compare performance and resource\nefficiency trade-offs of splitNN and other distributed deep learning methods\nlike federated learning, large batch synchronous stochastic gradient descent\nand show highly encouraging results for splitNN.\n