Enhancing Deep Fingerprinting Attack On Tor Network Using Lottery Ticket Hypothesis

Tor is one of the popular choices as an anonymity tool to protect the identity of users in digital-based communication. It is used to avoid any eavesdropping that could be performed by local adversaries for monitoring purposes. Website Fingerprinting is a technique by which an eavesdropper determines which websites a user is visiting over an encrypted connection such as Tor. In an encrypted and anonymized network observing patterns of data flows such as packet size and direction helps in identifying a webpage that is being accessed by a user. A new Fingerprinting attack called Deep Fingerprinting makes use of deep learning for finding out the best features to identify a website. It makes use of a Convolutional neural network with a sophisticated architecture design that achieves high accuracy even against the modern defenses proposed for minimizing the effect of fingerprinting. In this paper, we propose to apply The Lottery Ticket Hypothesis on Deep Fingerprinting Attack to reduce the complexity of the model. Thus, the revised model will be efficient in terms of computation costs, storage, and latency. The improvised model is not expected to degrade the accuracy of the model.