Federated learning is a recently proposed paradigm that presents significant advantages in privacy-preserving machine learning services. It enables the deep learning applications on mobile devices, where a deep neural network (DNN) is trained in a decentralized manner among thousands of edge clients. However, directly apply the federated learning algorithm to the mobile edge computing environment will incur unacceptable computation costs in mobile edge devices. Moreover, among the training process, frequent model parameters exchanging between participants and the central server will increase the leakage possibility of the users' sensitive training data. Aiming at reducing the heavy computation cost of DNN training on edge devices while providing strong privacy guarantees, we propose a mobile edge computing enabled federated learning framework, called FedMEC, which integrating model partition technique and differential privacy simultaneously. In FedMEC, the most complex computations can be outsourced to the edge servers by splitting a DNN model into two parts. Furthermore, we apply the differentially private data perturbation method to prevent the privacy leakage from the local model parameters, in which the updates from an edge device to the edge server is perturbed by the Laplace noise. To validate the proposed FedMEC, we conduct a series of experiments on an image classification task under the settings of federated learning. The results demonstrate the effectiveness and practicality of our FedMEC scheme. [ABSTRACT FROM AUTHOR]
Copyright of Mobile Networks & Applications is the property of Springer Nature and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)
