AWA: Adversarial Website Adaptation

One of the most important obligations privacy-enhancing technologies is to bring confidentiality and privacy users’ browsing activities on Internet. The website fingerprinting attack enables a local passive eavesdropper predict target user’s even she uses anonymous technologies, such as VPNs, IPsec, Tor. Recently, growth deep learning empowers adversaries conduct with higher accuracy. In this paper, we propose new defense against using adversarial approaches called Adversarial Website Adaptation (AWA). AWA creates transformer set in each run so that has unique transformer. Each generates traces evade adversary’s classifier. two versions, including Universal (UAWA) Non-Universal (NUAWA). Unlike NUAWA, there no need access entire trace order generate an UAWA. We accommodate secret random elements training phase transformers for various sets run. several times create multiple transformers. If adversary user select different transformers, accuracy classifier almost 19.52% 31.94% 22.28% 26.28% bandwidth overhead UAWA respectively. more powerful through trains them, 49.10% 25.93% 62.52% 64.33% NUAW,