Edge computing-oriented federated learning method

Abstract

The invention provides an edge computing-oriented federated learning method, which comprises the following steps that: all idle devices actively acquire a current global model from a server, asynchronously execute model training locally according to a preset optimization target, and upload the trained local model to the server; and the server receives the model formula uploaded by any one device and updates the global model by adopting a weighted average method. According to the method, asynchronous training and federation learning can be combined, in asynchronous federation optimization, allidle devices are used for asynchronous model training, a server updates a global model through weighted average, idle time of all edge devices is fully utilized, and model training is more efficient.

Description

technical field [0001] The present invention relates to the field of model training, in particular to an edge computing-oriented federated learning method and system. Background technique [0002] With the increase of various edge devices, such as smart grids, smartphones, IoT devices, etc., more and more data is used for machine learning training, so the data used for model training is transmitted to the server The traditional model training method of centralized training will bring many problems, such as huge communication overhead, limited computing resources, and privacy security risks. Compared with the traditional training of machine learning models directly in the server, which brings huge communication overhead and great security risks, federated learning can solve these problems well. In federated learning, the training of the model is transferred to each edge device or edge node, which solves the communication overhead problem caused by a large amount of data tran...

AI Technical Summary

Problems solved by technology

However, the computing resources and communication capabilities of most edge devices are limited, such as low battery energy, network congestion, limited number of GPUs and computing power, and network traffic charges, which cause the time required for them to train models and upload models. Longer, so the entire federated learning process is less efficient

Also, synchronous federated optimization fails to take full advantage of device idle time for model training

For example, during a certain round of training, those idle devices that were not selected were not used, or the device may be idle after uploading the updated local model and may no longer be selected.

In addition, due to the heterogeneity of edge devices, the data for training local models is not independent and identically distributed, so federated learning also needs to solve the problem of model differences caused by non-independent and identical distribution

Images