DefenseNet-A Resilent Network Against Adverserail Attacks On Recommendation Systems
Qurat Ul Ain, Nudrat Nida, Syed Aun Irtaza · 2021
On internet, one of the most progressing applications is recommendation system (RS), such as, youtube, google, Amazon, Netflix, etc. It is a subset of information filtering systems wherein, information about certain products or services or a person are categorized and are recommended for the concerned individual. The most popular recommender systems leverage the power of deep neural networks (DNNs) to make recommendations. But these recommendation systems implementing object detection and classification, are known to be susceptible to adversarial attacks. Small barely perceptible perturbations to the input can cause these algorithms to incorrectly classify inputs that they would have otherwise classified correctly. In our research work, we present an ensemble architecture of recommender system which combines the power of popular images recognition DNNs and presents higher defense ratio against adversarial attack. We have also shown that our DefenseNet’s accuracy is higher than the state-of-the-art recommendation under adversarial attacks on Amazon benchmark dataset.