Personalized federated learning method with efficient communication and privacy protection

Abstract

The invention provides a personalized federated learning method with efficient communication and privacy protection, which comprises the following steps of S1, pulling a current global model Wt from a central server to all clients, initializing a local model of each client; S2, executing E rounds of local training to obtain a new local model; S3, sending model parameters to the central server; s4, aggregating the received model parameters in the central server to obtain an aggregation result Wt + 1; s5, updating the local models of all the clients according to Wt + 1; S6, judging whether preset iteration times are completed or not; if so, completing personalized federated learning; and if not, enabling t to be equal to t + 1, and returning to the step S2 to carry out the next round of personalized federation learning. The invention provides a personalized federated learning method with efficient communication and privacy protection. The problem that an existing personalized federated learning method does not balance a local model and a global model of a personalized client is solved.

Description

technical field [0001] The present invention relates to the technical field of federated learning, and more specifically, to a personalized federated learning method for efficient communication and privacy protection. Background technique [0002] Machine learning has made great achievements in fields such as computer vision, speech recognition and natural language processing. In order to use massive data to complete large-scale machine learning training tasks, distributed machine learning has been proposed and attracted widespread attention. Federated learning is a new type of distributed machine learning. In federated learning, since the central central server uses the federated average (FedAvg) algorithm to aggregate model updates from various clients, all parties involved in federated training obtain a unified global model after training. Most of the existing federated learning algorithms focus on improving the global model effect of federated learning. However, in a ...

AI Technical Summary

Problems solved by technology

However, the existing personalized federated learning methods only emphasize the importance of the personalized client model, and fail to achieve a balance between the personalized client local model and the global model, resulting in a significant loss of the global model effect

[0003] In the existing technology, such as the Chinese patent published on August 28, 2020, a decentralized federated machine learning method under privacy protection, the publication number is CN111600707A, which solves the problem that the existing federated learning is vulnerable to DoS attacks and parameter centralization. Disadvantages such as server single point of failure; combined with PVSS verifiable secret distribution protocol to protect participant model parameters from model inversion attacks and data membership inference attacks

At the same time, it is ensured that parameters are aggregated by different participants in each training task. When there is an untrusted aggregator or it is attacked, it can return to normal by itself, which increases the robustness of federated learning; The performance of learning effectively improves the security training environment of federated learning, but does not achieve the balance between the personalized client local model and the global model

Images