Anonymity Enhancement Against Website Fingerprinting Attacks

Please enable javascript in your browser.

Shiravi, Saeed | 2019

628 Viewed

Type of Document: M.Sc. Thesis
Language: Farsi
Document No: 51999 (19)
University: Sharif University of Technology
Department: Computer Engineering
Advisor(s): Jalili, Rasool
Abstract:
Website Fingerprinting is a trac analysis attack that allows an eavesdropper to determine the web activity of a client, even if the client is using privacy technologies such as Tor. Recent researches have shown that an attacker is able to detect which websites a user is visiting over than 98% accuracy, While previous countermeasures fail against this kind of attacks. In this research, we introduce two defenses. In the rst defense, we exploit deep neural network vulnerabilities by using Adversarial Example.In this method, we add small perturbation to trac which misleads classier to detect websites that the user has visited. In the second defense, we introduce a defense mechanism based on Occlusion experiment. In this method, we detect what a deep neural network learn as a pattern and try to modify trac patterns so deep neural network misclassify users tracs. The rst method is able to reduce one of the novel attacks from 96% to 63% only by adding 21% bandwidth overhead. Also, Second proposed defense reduces the accuracy of the state of the art attack from 98% to 19% while introducing zero latency overhead and just than 47% bandwidth overhead
Keywords:
Anonymity System ; Tor Network ; Machine Learning ; Deep Learning ; Adversarial Example ; Traffic Analysis ; Website Fingerprinting ; Deep Neural Networks