Multi-party collaborative model updating method, device and system for realizing privacy protection

Abstract

The embodiment of the invention provides a multi-party collaborative model updating method and device for realizing privacy protection. In the collaborative model updating method, each participant i determines a corresponding local gradient vector according to a local sample set and a current model parameter, by using a randomization algorithm satisfying differential privacy, random binarization processing is carried out on each element in the local gradient vector to obtain a disturbance gradient vector. Each participant i sends the disturbance gradient vector determined by the participant i to the server. And the server aggregates the n perturbation gradient vectors, and performs binarization representation on each element according to the sign of each element in the current aggregation result to obtain a target gradient vector. And each participant i receives the target gradient vector from the server, and updates the current model parameter according to the target gradient vector for the next round of iteration. And after multiple rounds of iteration, each participant i takes the obtained current model parameter as a business prediction model cooperatively updated with other participants.

Description

technical field [0001] One or more embodiments of this specification relate to the field of computer technology, and in particular to a method, device and system for implementing a privacy-protected multi-party collaborative update model. Background technique [0002] The emergence of federated learning (also known as federated learning) has revolutionized traditional centralized machine learning, allowing participants to collaboratively build more accurate models without uploading local data. [0003] At present, federated learning is often realized by sharing model parameters or gradients among participants. However, since model parameters or gradients are usually high-dimensional private data, traditional federated learning is accompanied by high communication overhead and privacy issues to a certain extent. leakage etc. Contents of the invention [0004] One or more embodiments of this specification describe a method, device and system for implementing privacy-protect...

AI Technical Summary

Problems solved by technology

[0003] At present, federated learning is often realized by sharing model parameters or gradients among participants. However, since model parameters or gradients are usually high-dimensional private data, traditional federated learning is accompanied by high communication overhead and privacy issues to a certain extent. Leakage and other issues

Images