Improving robustness of neural network with adversarial training

Resource type
Thesis type
(Thesis) M.Sc.
Date created
Author: Zhou, Yajie
Deep neural networks have been applied in computer vision recognition and achieved great performance on many image classification tasks. However, deep neural networks are not as robust we expected all the time, and may be vulnerable to adversarial examples, which are images with some imperceptible changes to originals. In this work, we enhance the robustness of a given deep neural network using an accumulating adversarial training algorithm. We also propose an adaptive boosting method, a group sampling boosting method and a stochastic mini batch boosting method to boost the performance of the accumulating adversarial training algorithm. Moreover, we show that our proposed methods can enable a given deep neural network to protect against several adversarial attacking algorithms at the same time.
Copyright statement
Copyright is held by the author.
This thesis may be printed or downloaded for non-commercial research and scholarly purposes.
Scholarly level
Supervisor or Senior Supervisor
Thesis advisor: Pei, Jian
Member of collection