A Deep Learning Approach for Website Fingerprinting Attack

Website fingerprinting attack is an effective way to infer which web page a user is browsing through anonymous network. Previous studies focused more on the selection of traffic features and improvements in traditional classification methods. In this paper, we present a new website fingerprinting technique based on deep learning. The method uses a complex deep neural network to automatically learn features and classify website fingerprints. The model uses two-layer GRU network to extract the time feature and the 50-layer residual network to extract the spatial features of the website fingerprint. The main idea resides in the fact that deep neural network can achieve complex function approximation and more efficient feature extraction. Our technique performs better than the current state-of-the-art website fingerprint attack. We have achieved more than 99% accuracy over the current largest dataset which consists of 900 classes. The results of experiment show that deep learning is an efficient and robust technique for website fingerprint attack.