A 5G edge device scheduling method for fast federated learning

Abstract

The invention discloses a 5G edge device scheduling method for fast federated learning. At the beginning of each training cycle of federated learning, the method considers the training delay, fairness and local model importance of each edge device, based on multiple Arm bandit machine theory for edge device scheduling. The present invention also builds a wireless federated learning system, and its federated learning training process includes: at the beginning of each training cycle, device scheduling is performed based on the multi-armed slot machine theory; during the training process of each training cycle, the scheduled devices perform local training , and upload the updated local model to the base station to generate a new global model; at the end of each training cycle, the base station broadcasts the updated global model to all devices for device selection and model training in the next cycle. The invention can obtain higher learning performance with lower training time delay under the condition that the device channel condition and local computing capability are unknown.

Description

technical field [0001] The invention relates to the fields of federated learning and edge computing, in particular to a 5G edge device scheduling method for fast federated learning. Background technique [0002] Due to the popularization of the Internet of Things, the number of edge devices has increased significantly, resulting in a large amount of data generated by the edge devices under the wireless network. These data processing and analysis require machine learning algorithms. Traditional machine learning algorithms require a central controller to collect a certain amount of data for model training. Edge devices may be reluctant to share local data due to privacy concerns. Therefore, this challenge can be solved by an innovative distributed machine learning algorithm, namely federated learning. In federated learning, a device trains a local model based on a local dataset. Then, the updated local model is uploaded to the central server for model aggregation. Since f...

AI Technical Summary

Problems solved by technology

However, due to limited wireless resources and a large number of edge devices, the scheduling requirements for edge devices are very high

Images