Safe decentralized graph federal learning method

Abstract

The invention discloses a safe decentralized graph federal learning method. The method comprises the following steps of S1, sequentially numbering all clients, initializing graph neural network model parameters and sending a ring communication topological graph to all the clients; S2, enabling each client to determine a first-level neighbor client and a second-level neighbor client according to the ring communication topological graph, and performing key negotiation with each corresponding second-level neighbor client to generate a corresponding shared key; S3, training the local graph neural network model by each client, and updating parameters of the local graph neural network model; S4, enabling each client to receive the graph neural network model parameters sent by the first-level neighbor client to update a local graph neural network model; and S5, repeatedly executing the steps S3 to S4 till that the graph neural network model converges. According to the invention, the data privacy and security of each client can be protected, the communication bottleneck is reduced, and the communication time is shortened.

Description

technical field [0001] The invention relates to the technical field of graph federated learning, in particular to a secure decentralized graph federated learning method. Background technique [0002] In the past few years, the rise and application of neural networks have successfully promoted the research of pattern recognition and data mining. Traditional deep learning methods have achieved great success in extracting Euclidean space data features, but many of the data in the actual scene are generated from non-Euclidean spaces, and the performance of deep learning methods on such data is unsatisfactory. , as shown in the figure, the number of neighbor nodes of each node in the network is not fixed, resulting in some important operations (such as convolution) are easy to calculate on the image, but it is no longer suitable for direct use in the graph. Moreover, deep learning is based on the assumption that the training data satisfies independent and identical distribution,...

AI Technical Summary

Problems solved by technology

The above solution does not protect the local model parameters sent to the central server in the global model aggregation stage to prevent possible information leakage; secondly, the central server responsible for aggregating model information requires a trusted neutral third party Institutions, for modeling between institutions, it is difficult to guarantee such a trusted neutral third party; finally, this centralized architecture puts high demands on the IO capabilities of the central server, and all clients must Waiting for all clients to upload the model parameters to the central server successfully, and then the central server distributes the updated global model parameters to the clients before the client can proceed to the next cycle, which is undoubtedly very time-consuming

Images