Federal learning-oriented sample poisoning attack resisting method

Abstract

A federated learning-oriented confrontation sample poisoning attack method comprises the following steps: defining the following scenes: assuming that m participants participate in training, mgt; the k-th participant is assumed to be an attacker, and the attack target is that the performance of the federal learning global model on the test set is as poor as possible after the local model parameters participate in aggregation; the method comprises the steps that firstly, an attacker adds certain adversarial disturbance which cannot be perceived by human eyes to local private training samples to generate toxic adversarial samples, and local training is carried out based on the samples; secondly, in order to dominate the training process of the global model, an attacker improves the training learning rate in the local training process so as to accelerate the generation of malicious model parameters; and finally, an attacker uploads local model parameters to a server side to participate in aggregation so as to influence a global model. Under the attack of the method, the performance of the federal global model is remarkably reduced, and the attack method shows good generalization performance.

Description

technical field [0001] The invention belongs to the field of federated learning security technology, and in particular relates to a federated learning-oriented anti-sample poisoning attack method. Background technique [0002] Although machine learning technology has been widely used in various fields, data silos and data privacy issues are still two major challenges that hinder its development, such as: in medical applications, to train a machine learning model with good performance, various medical institutions or Departments provide a wealth of information that can describe a patient's symptoms, and medical data is often highly private and sensitive. Similarly, a large amount of heterogeneous data will be generated in a city's emergency, logistics and security information departments, which exist in the form of data silos and cannot be integrated and utilized. In order to solve the above problems, federated learning technology came into being. Different from the machine ...

AI Technical Summary

Problems solved by technology

Currently, there are relatively few studies applying such attacks to the training phase of federated learning

Images