Matching Items (2)

154767-Thumbnail Image.png

Secure and privacy-preserving microblogging services: attacks and defenses

Description

Microblogging services such as Twitter, Sina Weibo, and Tumblr have been emerging and deeply embedded into people's daily lives. Used by hundreds of millions of users to connect the people worldwide and share and access information in real-time, the microblogging

Microblogging services such as Twitter, Sina Weibo, and Tumblr have been emerging and deeply embedded into people's daily lives. Used by hundreds of millions of users to connect the people worldwide and share and access information in real-time, the microblogging service has also became the target of malicious attackers due to its massive user engagement and structural openness. Although existed, little is still known in the community about new types of vulnerabilities in current microblogging services which could be leveraged by the intelligence-evolving attackers, and more importantly, the corresponding defenses that could prevent both the users and the microblogging service providers from being attacked. This dissertation aims to uncover a number of challenging security and privacy issues in microblogging services and also propose corresponding defenses.

This dissertation makes fivefold contributions. The first part presents the social botnet, a group of collaborative social bots under the control of a single botmaster, demonstrate the effectiveness and advantages of exploiting a social botnet for spam distribution and digital-influence manipulation, and propose the corresponding countermeasures and evaluate their effectiveness. Inspired by Pagerank, the second part describes TrueTop, the first sybil-resilient system to find the top-K influential users in microblogging services with very accurate results and strong resilience to sybil attacks. TrueTop has been implemented to handle millions of nodes and 100 times more edges on commodity computers. The third and fourth part demonstrate that microblogging systems' structural openness and users' carelessness could disclose the later's sensitive information such as home city and age. LocInfer, a novel and lightweight system, is presented to uncover the majority of the users in any metropolitan area; the dissertation also proposes MAIF, a novel machine learning framework that leverages public content and interaction information in microblogging services to infer users' hidden ages. Finally, the dissertation proposes the first privacy-preserving social media publishing framework to let the microblogging service providers publish their data to any third-party without disclosing users' privacy and meanwhile meeting the data's commercial utilities. This dissertation sheds the light on the state-of-the-art security and privacy issues in the microblogging services.

Contributors

Agent

Created

Date Created
2016

157892-Thumbnail Image.png

Detecting Adversarial Examples by Measuring their Stress Response

Description

Machine learning (ML) and deep neural networks (DNNs) have achieved great success in a variety of application domains, however, despite significant effort to make these networks robust, they remain vulnerable to adversarial attacks in which input that is perceptually indistinguishable

Machine learning (ML) and deep neural networks (DNNs) have achieved great success in a variety of application domains, however, despite significant effort to make these networks robust, they remain vulnerable to adversarial attacks in which input that is perceptually indistinguishable from natural data can be erroneously classified with high prediction confidence. Works on defending against adversarial examples can be broadly classified as correcting or detecting, which aim, respectively at negating the effects of the attack and correctly classifying the input, or detecting and rejecting the input as adversarial. In this work, a new approach for detecting adversarial examples is proposed. The approach takes advantage of the robustness of natural images to noise. As noise is added to a natural image, the prediction probability of its true class drops, but the drop is not sudden or precipitous. The same seems to not hold for adversarial examples. In other word, the stress response profile for natural images seems different from that of adversarial examples, which could be detected by their stress response profile. An evaluation of this approach for detecting adversarial examples is performed on the MNIST, CIFAR-10 and ImageNet datasets. Experimental data shows that this approach is effective at detecting some adversarial examples on small scaled simple content images and with little sacrifice on benign accuracy.

Contributors

Agent

Created

Date Created
2019