Privacy protection method for collaborative deep learning model training

Abstract

The invention discloses a privacy protection method for collaborative deep learning model training, and the method comprises the steps: proposing a collaborative distributed deep learning model training method, wherein a participant makes model parameter gradient calculation locally through existing training data, transmitting the gradient data obtained through calculation to a parameter server,and updating model parameters ; proposing a privacy protection mechanism based on a double-trap-door public key cryptographic algorithm so that participants can realize safety training of a deep learning model on the premise of ensuring the privacy of respective training data; and designing a fine-grained deep learning model release method to ensure that only the data owners participating in training can obtain the model, so that the model training fairness is ensured. A simulation test result shows that the method can provide accurate model training service on the premise of ensuring data privacy of participants. Privacy protection can be provided for artificial intelligence and other new-generation computer technologies.

Description

technical field [0001] The invention belongs to the field of information security and relates to a privacy protection method, which can be used for collaborative security training of deep learning models in large-scale data. Background technique [0002] Machine learning is becoming a new engine for the development of the digital economy, especially driven by new theories and technologies such as mobile Internet, big data, supercomputing, sensor networks, and brain science, as well as the strong demand for economic and social development, machine learning will further empower All walks of life can promote the in-depth development of the digital economy. As a branch of machine learning, deep learning has also received more and more attention in industry and academia, and is widely used in medical diagnosis, speech recognition, image recognition and other fields. Deep learning is often based on massive data for model training. By analyzing the hidden correlation between data,...

AI Technical Summary

Problems solved by technology

The present invention proposes a privacy protection method for collaborative deep learning model training, which can solve the problem of multi-source data-oriented deep learning model security training, and solve the problems of traditional privacy protection schemes such as high computational overhead and model training accuracy cannot be guaranteed. In order to provide technical support for large-scale security applications of deep learning

Images