Precision feedback federated learning method for privacy protection

Abstract

The invention discloses a precision feedback federated learning method for privacy protection. The method comprises the following steps that: 1, each client performs data enhancement on a local data set by using a GAN, and uploads the obtained generated data to a server to form a shared data set; 2, the server initializes the model parameters and broadcasts the model parameters to each client; 3, each client performs local model training by using the downloaded global parameters, and uploads the parameters to the server after training; 4, the server tests each local model to obtain model precision and then generate a new aggregation weight; 5, the server performs model aggregation on the local model by using the aggregation weight; 6, the server performs global training on the aggregated parameters by using the shared data set, and broadcasts the parameters to each client after obtaining a global model; and 7, steps 3-6 are repeated until the model performance meets the requirement. According to the method, the influence on the global model performance caused by non-independent identical distribution of the client data and the training weight of the client can be improved on the premise of protecting the client data.

Description

technical field [0001] The invention belongs to the field of federated learning, and in particular relates to a privacy protection-oriented precision feedback federated learning method, which is dedicated to improving the impact of non-independent and identical distribution of local data of each client on the performance of a global model. Background technique [0002] Data is the foundation of machine learning. As the main direction of artificial intelligence, machine learning needs data to train artificial intelligence models. In most industries, due to issues such as industry competition, privacy security, and complex administrative procedures, data often exists in the form of isolated islands, and the performance of artificial intelligence models trained only using data in isolated data islands often cannot meet task requirements. . Aiming at the dilemma of data islands and data privacy, the framework of federated learning algorithm came into being. [0003] Under the ...

AI Technical Summary

Problems solved by technology

In the process of implementing this strategy, there are at least the following problems in the prior art: (1) Privacy issues should be emphasized during the application of federated learning methods, that is, the protection of private data of each client

However, this strategy proposes to extract part of the data from each client to form a shared data set, which to a certain extent caused the leakage of client data and violated the original intention of the federal method to maintain privacy.

(2) In the current federated averaging method, the weights of each client are fixed, and the weights of different clients are not adjusted correspondingly according to the performance and training rounds of different model training

Images