Model training method based on decentralized federated learning

Abstract

This application proposes a model training method based on decentralized federated learning, including: constructing a federated learning network, the federated learning network includes multiple nodes and a broadcast bus; As a slave node, the master node transmits the first model data to each slave node; each slave node performs training based on the first model data and the local data set to obtain the second model data, and adds noise data to the second model data to obtain the second model data The three-data model transmits the third model data to the master node; the master node receives and cleans the third model data, and performs model data aggregation on all the second model data obtained after cleaning. This method builds a decentralized federated learning network, realizes stability and guarantees that there is a master node in the federated learning network that can communicate with all nodes, and provides a noise protection mechanism to protect the privacy of model data.

Description

technical field [0001] The present application relates to the technical field of computer data processing, in particular to a model training method based on decentralized federated learning. Background technique [0002] Federated learning is essentially a distributed machine learning technology, or machine learning framework. The purpose is to collaboratively complete model training on the basis of ensuring data privacy security and legal compliance, achieve common modeling, and improve the detection effect of the model. That is to say, when training a certain model, it is necessary to use a large amount of private data containing users. In order to avoid user information leakage, the federated learning method is used to receive the original model from the central server, and then use the federated learning method. The user's local data set is trained to obtain model parameters. Finally, each user only needs to upload the model parameters to the central server, and the cent...

AI Technical Summary

Problems solved by technology

This has two disadvantages. One is that if there is only one problem with the noise transmission between nodes, the final result will deviate from the ideal result (not to mention the occurrence of malicious attacks and interference); the other is that the nodes transmit independently Noise needs one ring after another, and cannot be executed concurrently. If there are too many nodes, the efficiency will be low

In particular, the noise in this scheme is random one-way noise, while the actual model is multi-directional, that is, the model has three directions of x-axis, y-axis, and z-axis like a three-dimensional space (actually more, much more than three), so the added noise has limitations

Images