A Gradient Descent Computation Method for Protecting Privacy Data

Abstract

The invention discloses a gradient descent calculation method for protecting privacy data. The method is used in the gradient function calculation of machine learning including one or more data providers, a decryption service provider and a computing resource provider. Use a polynomial function that is similar to the contour of the sigmoid function to fit the sigmoid function or use a piecewise function that is similar to the contour of the sigmoid function to fit and calculate the sigmoid function, including: homomorphic encryption key generation and distribution, training parameters Negotiation, data encryption and summarization and gradient descent process steps. This method has high precision, and the precision loss caused by data processing in the calculation process is in a controllable range; it has high security, and both the input and intermediate data in the calculation process can meet the requirements of semantic security; it has good flexibility, and can achieve two or more Multiple participants participate in the computation; good scalability, can be extended from the original gradient descent to Newton's method or batch gradient descent.

Description

technical field [0001] The invention relates to the field of privacy protection of machine learning, in particular to a gradient descent calculation method for protecting privacy data in machine learning. Background technique [0002] In modern society, machine learning technology is more and more widely used in various fields, such as medical, business, education and public safety. However, a large amount of private data is involved in the process of machine learning, especially in the scenario where these data belong to different data providers, there are a series of threats of privacy leakage. Therefore, machine learning algorithms with privacy protection have always been academic a research hotspot in the world. Among them, the gradient descent algorithm is an important optimization method in the field of machine learning, which is widely used in the training process of various machine learning algorithms, including algorithms such as logistic regression, matrix factori...

AI Technical Summary

Problems solved by technology

[0003] For the protection of private data in the gradient descent algorithm, most of the existing schemes adopt some less secure linear encryption or differential privacy methods to design schemes with certain privacy protection, but these schemes are difficult to ensure that all private data will not be Give way

There are also some schemes that use fully homomorphic encryption schemes to encrypt original data while ensuring data confidentiality and computing power, but the efficiency of fully homomorphic encryption algorithms at this stage is low, which limits the practical application of these schemes

Images