Asynchronous federated learning method and device for improving utilization efficiency of edge device, and medium

Abstract

The invention provides an asynchronous federated learning method and device for improving the utilization efficiency of edge equipment and a medium, which can reduce the data competition of a plurality of threads in a server on a global model and improve the concurrency performance of the server, and comprises the following steps: the edge equipment meeting a model data transmission condition actively requests the global model from the server; if the state of the event object is false, the server sends the global model to the edge device through the dispatcher component; the edge devices meeting the conditions train a global model through local data to obtain a local model; a collector component of the server enqueues the local model into a queue; the updater component pops up the local model and the shadow model from the queue to execute aggregation operation, assigns an aggregation result to the shadow model, assigns the value of the shadow model to the global model when a dequeue count value reaches a set value, and updates the number of global iterations; and continuously iterating until a set global total iteration number is reached.

Description

technical field [0001] The invention relates to the technical field of asynchronous federated learning in machine learning, in particular to an asynchronous federated learning method, device and medium for improving the utilization efficiency of edge devices. Background technique [0002] In recent years, various Internet companies have anonymously uploaded users' sensitive data to the company's central server, used the data to train mathematical models and used it to provide personalized services to their customers or help companies make business decisions. However, with the increase of data sources and data volume, with the help of tools such as information integration and big data, it is impossible to completely anonymize sensitive information. The 2006 AOL company search data leakage event triggered widespread concerns about personal privacy data leakage and heated discussions on data ownership. In recent years, legislators and regulators have put forward strict constra...

AI Technical Summary

Problems solved by technology

However, on the server side, due to data competition among multiple threads on the global model, the asynchronous federated optimization algorithm that uses this method to update the global model cannot run efficiently and concurrently on the central server, which reduces the training speed of the global model and reduces the utilization efficiency of edge devices. high

[0006] In addition, on the edge device side, there is also the problem of low utilization of edge devices. First, edge devices must meet some strict conditions before starting model training, for example: mobile phones should be in charging, standby, and can pass through the wireless network. Access to the Internet, etc., so that the training process will not disturb the normal use of mobile phones by mobile phone users; secondly, the network environment where edge devices are located is complex, and the connection between edge devices and servers often occurs

The above two situations will cause the model training process to be terminated, causing the device to go offline and enter an idle state, further reducing the utilization rate of the device, resulting in the inability to fully utilize the concurrent performance of the server.

Images