Federal learning poisoning detection method and device based on feature confrontation

Abstract

The invention discloses a federal learning poisoning detection method and device based on feature confrontation, and the method comprises the steps: dividing all clients of each round of parameter training into benign clients and defense clients, and configuring a defense patch data set for the defense clients; in each round of training, enabling the benign client to optimize the benign model by using the local data set, enabling the defense client to optimize the defense model by using the defense patch data set and the local data, and enabling the server to aggregate all the benign models and the defense models to obtain a federated learning model; after multiple rounds of training are finished, using the federated learning model of the last round for detecting a poisoning sample, and during detection, according to a prediction result of a target label of a test sample in the federated learning model, and judging whether the test sample is poisoned or not by judging whether the prediction result of the defense target label in the federated learning model meets the label mapping relation after the optimal defense patch data is added into the test sample, namely realizing federated learning poisoning detection.

Description

technical field [0001] The invention belongs to the field of federated learning, and in particular relates to a federated learning poisoning detection method and device based on feature confrontation. Background technique [0002] With the rapid development of data-driven intelligent applications, the machine learning paradigm is also facing new difficulties and challenges. On the one hand, the machine learning paradigm hopes to provide a robust and efficient functional service for all users. On the other hand, it is difficult to fully share data as the nutrition of learning algorithms. [0003] In order to solve this problem, federated learning emerged as a potential solution. Its main innovation is that it provides a distributed machine learning framework with privacy protection characteristics, and can coordinate data in a distributed manner Thousands of participants iteratively train on a particular machine learning model. Since the training data is still stored local...

AI Technical Summary

Problems solved by technology

Model training failure or backdoor implantation will directly or indirectly cause huge economic losses to AI service platforms and ordinary users

[0006] In short, due to the privacy mechanism of federated learning, local data cannot be observed, and malicious data nodes are difficult to detect

Images