Sandbox mechanism for federated learning

Abstract

The invention discloses a sandbox mechanism for federated learning. The sandbox mechanism comprises the following steps: establishing a federated learning sandbox training environment, establishing afederated learning sandbox market environment, measuring federated learning related parameters, and reporting and decrypting encrypted measured parameters. The invention provides a set of sandbox mechanism for the federated learning system, so that a federated learning intention user can test the federated learning system in the sandbox before officially adopting the federated learning system, measure required parameters in advance, help participants to quantify own attributes in advance, formulate a federated learning strategy and establish confidence of joint modeling; an effective way is provided for participants to know own attributes; the quantified participant attributes are helpful for the federation learning system to distribute the income generated by the federation learning modelmore effectively; the sandbox mechanism provides a trial and error way, supports participant tests and applies an optimal strategy; and the safety, the stability and the operability of federated learning in practical application are improved.

Description

technical field [0001] The present invention relates to federated learning in computer science, in particular to the sandbox mechanism in the federated learning process. Background technique [0002] In recent years, artificial intelligence applications that rely on data to drive have been in trouble. Although artificial intelligence algorithms are developing rapidly, training artificial intelligence models usually requires a large amount of data, which is often distributed in the hands of different organizations, resulting in data islands due to strict supervision to protect data privacy and security. Therefore, the Federated Machine Learning system came into being. Federated learning establishes a multi-party shared machine learning model through an encrypted parameter exchange mechanism on the premise of satisfying data security and user privacy protection, so that the machine learning model can be jointly established without any party's data being leaked. Island proble...

AI Technical Summary

Problems solved by technology

However, sandbox mechanisms in these other computer fields are difficult to be directly applied to federated learning scenarios

Images