Network model training system, method, device and medium based on parameter sharing

Abstract

The invention discloses a network model training system, method, device and medium based on parameter sharing. The system includes a first parameter module and a second parameter module, the first parameter module is used to obtain the first model parameters formed when the first local end trains the network model, and synchronize the first model parameters to the block chain; the second parameter module is used to obtain the second model parameter requested by the second local end from the block chain and synchronize to the second local end; the second model parameter is used for the second local end Train the network model. The present invention realizes the sharing of model parameters through the block chain, which is beneficial to the large-scale application and mutual learning of network models; since there is no need to exchange network model training sets and test sets between local terminals, network model parameters can be efficiently propagated on the basis of avoiding the leakage of sensitive data. The invention is widely used in the technical field of artificial intelligence.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence, in particular to a network model training system, method, device and medium based on parameter sharing. Background technique [0002] Network models such as knowledge graphs and neural networks are important artificial intelligence tools with a wide range of uses. For example, knowledge graphs can be used for web search, link prediction, recommendation, and natural language processing. These artificial intelligence tools need to go through steps such as training and testing to have corresponding performance, and the model parameters obtained after training represent the shape of the network model, and the model parameters can be fixed by storage and used for the next use. [0003] According to the principle of the network model, the performance of the network model is related to its model parameters, and the model parameters are determined by the training set and test set used wh...

AI Technical Summary

Problems solved by technology

In some special fields, the training set and test set involved have certain privacy and confidentiality. For example, the training set and test set used in the medical field are generally hospital patient data, and the training set and test set used in the financial field are generally Insurance data of insurance companies, these data are difficult to communicate and use publicly, which is not conducive to the large-scale application of network models

Images