Federal learning method and system for longitudinal xgboost decision tree

Abstract

The invention discloses a federated learning method and system for a longitudinal xgboost decision tree, relates to the technical field of federated learning and machine learning decision trees, and can ensure the privacy of feature data of samples held by all parties and enhance the security of the feature data of all parties. According to the main technical scheme, a combined training process and a combined reasoning process of a longitudinal xgboost decision tree are provided, splitting points are calculated in the combined training process, each node is judged in the combined reasoning process, and information disclosed in the combined training process is the maximum splitting value of each participant. The feature information of each participant is not directly leaked, and the safety of the joint reasoning process depends on a threshold homomorphic encryption scheme. The method is mainly applied to joint training and joint reasoning processes of decision trees.

Description

technical field [0001] The invention relates to the technical field of federated learning and machine learning decision tree, in particular to a federated learning method and system of vertical xgboost decision tree. Background technique [0002] Vertical xgboost decision tree federated learning means that when the feature information and label information of each sample are held by different owners, all parties jointly carry out the training of xgboost decision tree, and all parties do not want to send information to any other party during the training and inference process. Leaking feature information or label information about samples. [0003] For example, the complete sample information required for xgboost decision tree training is as follows, as shown in Table 1: [0004] ID x1 x2 x3 x4 x5 x6 y u1 ... ... ... ... ... ... ... u2 ... ... ... ... ... ... ... u3 ... ... ... ... ... ... ... [0005] Among them, ID repr...

AI Technical Summary

Problems solved by technology

[0022] As mentioned above, in the training process and the joint reasoning process of the SecureBoost scheme, there is a serious leakage of the feature information of the passive party, so in the process of longitudinal xgboost decision tree federated learning, it is difficult to protect the privacy and security of the feature information of samples from all parties sex

Images