Method for carrying out batch standardization in logic regression model of security federated learning

Abstract

The invention relates to the related field of federated learning, and particularly discloses a method for carrying out batch standardization in a logic regression model of security federated learning, which comprises the following steps: S1, federated learning participants prepare training data: the federated learning participants can have multiple parties P1, P2,..., Pn, each participant has training data data_1, data_2,..., data_n, one party has a label y, it is assumed that the participant with a label is P1, and in addition, a coordinator is used for gradient decryption; S2, data set fusion is performed: in the longitudinal federated learning, intersection is firstly carried out on the data, and a sample ID (Identity) commonly owned by each participant is screened to form a fused data set; S3, a logistic regression model in longitudinal federated learning is trained; S4, prediction is performed. According to the method, on one hand, model training failure caused by incapability of convergence due to overlarge gradient in logistic regression algorithm modeling of security federated learning can be avoided, and the training success rate of the model is improved;.

Description

technical field [0001] The invention relates to the field related to federated learning, in particular to a method for performing batch normalization in a logistic regression model of safe federated learning. Background technique [0002] Machine learning refers to the process of using certain algorithms to guide the computer to independently construct a reasonable model using known data, and using this model to make judgments on new situations. , financial risk management and other applications play a very important role. Traditionally, machine learning models are trained on a centralized corpus of data, which may be collected by a single or multiple data providers. Although parallel distributed algorithms have been developed to speed up the training process, the training data itself is still collected and stored centrally in a single data center. [0003] In May 2018, the European Union passed the General Data Protection Regulation (GDPR) bill, raising the requirements f...

AI Technical Summary

Problems solved by technology

[0005] In the logistic regression model of vertical federated learning, the approximate function of the sigmoid function used in the calculation of the homomorphically encrypted data cannot limit the upper and lower limits of the calculation results; in the model operation, there is no sum of the product of the model parameters and the input features Limiting, when this sum is too large, the gradient of the model will always increase and become unable to converge

Images