Decision tree-oriented vertical federated learning method

A decision tree-oriented vertical federated learning method, comprising the following steps:

• (1) each participant sorting local samples according to each feature of the samples contained locally, and then dividing the sorted samples into different blocks in sequence, wherein each block is called a bucket and numbered in sequence, and each bucket contains a same number of samples;
• (2) for a group of samples corresponding to each feature, a bucket number of each sample under this feature having a certain probability to be replaced by other bucket numbers, and readjusting the bucket number corresponding to the sample for encryption, wherein the selected probability makes the encryption method meet the definition of differential privacy;
• (3) each participant sending the adjusted bucket numbers of all local samples under different features to a coordinator, who is a participant holding a label;
• (4) the coordinator calculating a gradient of each sample according to local label information, and summarizing the gradient of each bucket corresponding to each feature;
• for a group of samples corresponding to each feature, dividing data by bucket and calculating information gain, and finally, recording a partition position and a current feature corresponding to a maximum information gain, and returning the partition position and the current feature to the participant with the corresponding feature data;
• (5) repeating step (4) for each node to train a first tree;
• and
• (6) updating parameters, repeating steps (4) and (5), and training the whole model.