Bayesian classifier security generation system and method for multi-party collaboration

Abstract

The invention belongs to the technical field of information security, and discloses a secure multi-party Bayesian classifier generation system and a method, and the method comprises system initialization, system security parameter generation by a key distribution center, a distributed data encryption key and an aggregated data decryption key. Encrypting the local training data, and sending the encrypted data to a model generator; the model generator performs aggregation calculation on the received ciphertext training data to generate ciphertext global training data, and decrypts the ciphertextglobal training data by using an aggregation data decryption key to obtain Bayesian classifier training parameters; and the Bayesian classification model generator uses the obtained Bayesian trainingparameters to calculate a corresponding conditional probability and a corresponding preamble probability, and generates a Bayesian classifier. The method can be used for generating and training the Bayesian classifier in a distributed scene, and can realize safe aggregation and privacy protection of sensitive data of multiple data centers while ensuring that a model generator acquires the high-precision Bayesian classifier.

Description

technical field [0001] The invention belongs to the technical field of information security, and in particular relates to a system and method for securely generating a Bayesian classifier oriented to multi-party collaboration. Background technique [0002] At present, the closest existing technology: With the rapid growth of Internet data volume and the continuous development of information technology, machine learning has received widespread attention. Naive Bayesian classifier, as a typical machine learning algorithm, can provide accurate and efficient data classification services through learning modeling, and has been widely used in many fields such as finance, medical care, and transportation. In the traditional Bayesian data classification service, the model generator aggregates the data in the data center to train the classifier, and then provides the data classification service. In the above process, the sensitive data (such as sample data, statistical analysis data...

AI Technical Summary

Problems solved by technology

The disadvantage of this invention is that when order-preserving encryption is used for multi-dimensional data, since the order-preserving encrypted ciphertext maintains the size relationship of the corresponding plaintext, the order of the plaintext and the correlation between data of different dimensions will be leaked, and there is security. sexual inadequacy

The disadvantage of this invention is that: the construction of decision tree classifier requires complex mathematical operations, the proposed method fails to ensure the security of the original data in the process, and only protects the privacy of sensitive data in the process of data classification

[0005] To sum up, the problems existing in the existing technology are: the existing data classification privacy protection scheme has insufficient security and cannot support data protection in the classifier training process, etc.

At the same time, most of the solutions do not construct secure multi-data source classifier generation methods for data distributed storage scenarios

[0006] Difficulty in solving the above technical problems: Existing privacy protection schemes do not support complex calculations, and there are mutual constraints between security and usability

At the same time, the existing homomorphic encryption technology is difficult to be used in distributed computing scenarios, and cannot provide effective privacy protection for multiple data sources

Images