Privacy protection machine learning training and reasoning method and system based on block chain

Abstract

The invention discloses a privacy protection machine learning training and reasoning method and system based on a block chain. The system comprises a machine learning model training module, a machine learning model reasoning service module, and a block chain service module which is connected with the machine learning model training module and the machine learning model reasoning service module and communicates with the machine learning model training module and the machine learning model reasoning service module. The method is applied to a block chain network, and the block chain network comprises a machine learning model owner node, a calculation contractor node, a trusted arbitration node and an inference service client node. Wherein the machine learning model owner node, the calculation contractor node and the trusted arbitration node form a model channel, and the machine learning model owner node, the reasoning service client node and the trusted arbitration node form a reasoning channel; according to the method, model training and reasoning are separated, reasoning services are separated, and good privacy protection is formed by information isolation between channels; and the safety and credibility of the model training and reasoning process are ensured.

Description

technical field [0001] The invention relates to a block chain-based privacy protection machine learning training and reasoning method and system, belonging to the technical field of data management. Background technique [0002] Data and models are the foundation of machine learning, and it is very important to ensure the security and privacy of data and models during the training and inference process of machine learning. Leakage of machine learning training data may lead to the theft of personally identifiable information and sensitive private information of user data, threatening user security. Leakage of machine learning models, such as model training algorithms, model topology, model weight parameters, activation functions, hyperparameters, and other information related to machine learning models, may lead attackers to launch model reverse attacks, that is, by Reverse analysis to obtain the training data of the restored model. The research of privacy-preserving machin...

AI Technical Summary

Problems solved by technology

However, privacy-preserving machine learning inevitably increases computing overhead, and outsourced computing provides a training path for model owners with limited device resources but privacy requirements

Images