Vertical federated learning defense method based on auto-encoder

Abstract

The invention discloses a vertical federated learning defense method based on an auto-encoder. The method comprises the steps that (1) a terminal trains an edge model through local data, and aggregates the embedding features of adjacent nodes of each layer in the edge model in a training process; (2) the terminal constructs and trains an auto-encoder comprising an encoder and a decoder to obtain encoder parameters and decoder parameters, and encodes the embedded features by using the encoder to obtain encoding information; (3) the terminal uploads the decoder parameters to the server, and after the server constructs a decoding model according to the decoder parameters and carries out message verification with the terminal, the terminal uploads coded information to the server; and (4) the server decodes the received coded information by using the decoding model to obtain decoded information, aggregates all decoded information to obtain embedded information, trains the global model by using the embedded information, and feeds back gradient information to each terminal after training. The malicious participant can be effectively prevented from stealing the private data.

Description

technical field [0001] The invention relates to the technical fields of deep learning and privacy security, in particular to an autoencoder-based vertical federated learning defense method. Background technique [0002] With the research results of deep neural network (DNN), it has been widely used in machine translation, image recognition, unmanned driving, natural language processing, network map analysis, electromagnetic space countermeasures, biomedicine, finance and other fields. With its powerful feature extraction capabilities, deep learning gradually replaces humans in various fields for autonomous decision-making. However, once data and models are leaked, it will cause significant personal injury and property loss. [0003] Since Google proposed the concept of federated learning (FL) in 2016, all participants in the process of federated learning jointly train the joint model by exchanging gradient information or intermediate results, which effectively solves the pr...

AI Technical Summary

Problems solved by technology

However, the defense method against differential privacy attacks greatly reduces the accuracy of the original task while protecting the embedded feature information of nodes, which is not suitable for this vertical federated learning scenario for processing graph data.

Images