Adversarial attack defense method based on adversarial sample training

Abstract

The invention discloses an adversarial attack defense method based on adversarial sample training for an adversarial attack defense problem of a deep learning network. The adversarial attack defense method includes the steps: firstly, prior distribution of training sample category labels is introduced, and the labels are subjected to smooth correction, so that the network does not excessively believe that the training samples are completely correct during training; secondly, adversarial samples are added in the network training process, meanwhile, an original loss function is modified, the contribution of the adversarial samples to the loss function is reflected, so that the network has resistance to gradient attacks; and finally, random inactivation processing is performed on the networkneurons, so that over-fitting is reduced to a certain extent. According to the adversarial attack defense method, overfitting is inhibited, and the generalization ability of the network is enhanced, so that the robustness of the network for defending against resistance sample attacks is improved.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence and relates to a defense method for deep learning adversarial sample attacks, in particular to an adversarial attack defense method based on adversarial sample training. Background technique [0002] Deep learning technology has been widely used in image classification and recognition, natural language processing, speech processing and other fields. However, artificial intelligence systems face security risks against sample attacks. Adversarial example attack refers to adding perturbations that are difficult for humans to detect on the input samples, so that the samples that humans will not make mistakes in classification are misidentified and classified by machines. [0003] In actual application scenarios, adversarial sample attacks sometimes cause huge security problems. For example, in face recognition, attackers can use adversarial samples to bypass verification and gain per...

AI Technical Summary

Problems solved by technology

Although the recognition accuracy of the deep learning network after adding adversarial training can be improved, there is still a large gap from the original sample recognition accuracy, and from the perspective of the robustness of the intelligent unmanned system defense against adversarial attacks, this accuracy is not reliable enough

Images