论文笔记——Client Selection for Federated Learning with Heterogeneous Resources in Mobile Edge
This paper proposes a protocol based on edge computing to improve the federation learning algorithm.
The specific MEC platform in the wireless network composed of the server and the base station (BS) manages the behavior of the server and the client.
Use the MEC operator to select the client.
First, randomly select a certain percentage of clients and ask them to send their resource information (such as wireless channel status, computing power (for example, whether they can spare CPU or GPU to update the model) and the size of data resources related to them ) To the MEC operator.
The MEC operator then estimates the time required for the distribution and planned update and upload steps based on the information received, and determines which clients enter these steps in order to select the client.
FedCS agreement
Client selection
Use CIFAR-10 and Fashion-MNIST data sets to test protocol performance:
IID
non-IID
Will the client with a large amount of data have less chance to participate in training and waste a lot of data available for training?
