Federated learning system and federated metric anomaly detection and handling method

Abstract

The application provides a federal learning system and a federal index abnormality detection and processing method, and belongs to the technical field of federal learning. The method first collects abnormal signals of the federal index submitted by the federal learning participant, then quantizes and normalizes the collected abnormal signals, and calculates an abnormal score. If the abnormal score is greater than or equal to a preset threshold, the federal learning participant is subjected to federal index release freezing and settlement freezing. Then, the corresponding abnormal features are identified based on the collected abnormal signals, the minimum review data range is circled based on the abnormal features by reverse positioning, the abnormal review is performed based on the minimum review data range, and finally the ruling conclusion is generated according to the abnormal review result and the corresponding post-processing action is executed. Thus, the whole-process automation closed loop of abnormal monitoring, emergency freezing, accurate review and ruling disposal can be formed, the malicious attacks can be effectively resisted, and the fairness, credibility and governance efficiency of the federal learning system are significantly improved.

Images