Adversarial examples design by deep generative models
Thesis title in Czech: | Tvorba nepřátelských vzorů hlubokými generativními modely |
---|---|
Thesis title in English: | Adversarial examples design by deep generative models |
Key words: | Hluboké učení|klasifikace|generativní modely|nepřátelské vzory |
English key words: | Deep learning|classification|generative models|adversarial examples |
Academic year of topic announcement: | 2018/2019 |
Thesis type: | diploma thesis |
Thesis language: | angličtina |
Department: | Department of Theoretical Computer Science and Mathematical Logic (32-KTIML) |
Supervisor: | Mgr. Roman Neruda, CSc. |
Author: | hidden - assigned and confirmed by the Study Dept. |
Date of registration: | 04.03.2019 |
Date of assignment: | 04.03.2019 |
Confirmed by Study dept. on: | 22.03.2019 |
Date and time of defence: | 22.06.2021 09:00 |
Date of electronic submission: | 21.05.2021 |
Date of submission of printed version: | 21.05.2021 |
Date of proceeded defence: | 22.06.2021 |
Opponents: | Mgr. Martin Pilát, Ph.D. |
Guidelines |
Current deep learning systems are vulnerable to artificially constructed adversarial examples that can lead the model to misclassify its output. While the main approach to adversarial examples is based on gradient methods, the possibility to use generative models such as GAN has been proposed in recent research. The goal of this thesis is to explore the possibility of generative models to design successful adversarial examples for deep neural network classifiers. The student will propose an algorithm that uses generative models in the context of adversarial attacks on a trained neural network. The algorithm will be implemented and tested on currently used benchmark data sets to assess its usability both in successful attacks and in the possible defences against them. |
References |
[1] A. Chakraborty et al.: Adversarial Attacks and Defences: A Survey. ACM Comp. Surv. (to appear), 2019. (arXiv:1810.00069)
[2] Y. Song et al: Constructing Unrestricted Adversarial Examples with Generative Models. Advances in Neural Information Processing Systems 31, 8312-8323, 2018. [3] Ch. Xiao et al: Generating Adversarial Examples with Adversarial Networks. 2019. (arXiv:1801.02610v5) [4] I. Goodfellow et al: Deep Learning. MIT Press, 2016. |