Backdoor attack defense system for artificial intelligence model

Abstract

A backdoor attack defense system for an artificial intelligence model comprises a model pre-detection module and a real-time defense module, the model pre-detection module scans an operation behavior of reading or modifying a local file and data output by each layer in a to-be-detected image classification model so as to judge whether the model contains a backdoor; and the real-time defense module adds a filter to the picture containing the trigger, so that the back door in the image classification model is invalid. The method can be applied to models in the fields of image classification and image recognition, defense is enhanced for the models, and then the safety performance in the field of deep learning is improved.

Description

technical field [0001] The invention relates to a technology in the field of information security, in particular to a backdoor attack defense system for artificial intelligence models. This technology can be applied to the protection of artificial intelligence models in the image field in the industrial field, such as image classification, face recognition, automatic driving, etc. Background technique [0002] Adding or changing neurons to a normal artificial intelligence model can turn it into a model with a backdoor. Once the input image contains a trigger that triggers the backdoor, the deep learning model will not only output classification results, but also execute the malicious neurons of the model, causing damage to the local system. There is currently no effective defense against this type of backdoor attack. After testing, mainstream antivirus software such as Norton, Kaspersky, McAfee, etc. cannot identify whether the model contains a backdoor. Contents of the ...

AI Technical Summary

Problems solved by technology

After testing, mainstream antivirus software such as Norton, Kaspersky, McAfee, etc. cannot identify whether the model contains a backdoor

Images