Image and Attribute Based Convolutional Neural Network Inference Attacks in Social Networks

Abstract

In modern society, social networks play an important role for online users. However, one unignorable problem behind the booming of the services is privacy issues. At the same time, neural networks have been swiftly developed in recent years, and are proven to be very effective in inference attacks. This article proposes a new framework for inference attacks in social networks, which smartly integrates and modifies the existing state-of-the-art convolutional neural network (CNN) models. As a result, the framework can fit wider applicable scenarios for inference attacks no matter whether a user has a legit profile image or not. Moreover, the framework is able to boost the existing high-accuracy CNN for sensitive information prediction. In addition to the framework, the article also shows the detailed configuration of fully connected neural networks (FCNNs) for inference attacks. This part is usually missing in the existing studies. Furthermore, traditional machine learning algorithms are implemented to compare the results from the constructed FCNN. Last but not least, this article also discusses that applying differential privacy (DP) can effectively undermine the accuracy of inference attacks in social networks.