Funded Projects

Research Projects

• Privacy Analytics. Increasingly large amounts of data are being collected about users, and increasingly sophisticated analytics are being applied to this user data. This data is collected for purposes spanning improving services, marketing products, surveillance, and social control. Privacy analytics are machine learning and data mining algorithms applied by end-users to their data for the purpose of helping them manage both private information and their self-presentation. This research develops privacy analytics that help users answer three key, interconnected questions about their online persona (1) What data do I consider sensitive, and in what contexts should I share it?; (2) What does my data say about me?; and (3) Who knows what? These privacy analytics introduce a novel, inverse data mining problem where users analyze their own data to estimate the impact and conclusions the data will produce when shared and incorporated into larger data sets. This project designs new algorithms for quantitative and automated methods to detect privacy-related phenomena that have been observed qualitatively. These algorithms support the development of usable privacy enhancing technologies. These new analytics will give users tools to cope with and manage their data in an increasingly complicated data environment. While it is not possible for people to control what happens to their data in every situation, these tools will provide awareness to users about how and where their data is being used. These analytics will also help answer questions critical to the development of privacy law and policy.
  1. Publications
     • Aylin Caliskan-Islam (2015). How do we decide how much to reveal? (Hint: Our privacy behavior might be socially constructed.). to appear in ACM Computers and Society, Special Issue on Security, Privacy and Human Behavior. (Related blog post on Freedom to Tinker)
     • M. Juarez, S. Afroz, G. Acar, C. Diaz, and R. Greenstadt, A Critical Evaluation of Website Fingerprinting Attacks. ACM Conference on Computer and Communications Security (CCS), 2014. (AR 19.5%)
     • A. Caliskan-Islam, J. Walsh, and R. Greenstadt. Privacy Detective: Detecting Private Information and Collective Privacy Behavior in a Large Social Network. in Workshop on Privacy in the Electronic Society, November 2014. (AR 25.3%)
     • Sadia Afroz, Aylin Caliskan Islam, Jordan Santell, Aaron Chapin. Rachel Greenstadt. How Privacy Flaws Affect Consumer Perception. in 3rd workshop on Socio-Technical Aspects in Security and Trust (STAST '13), June 2013.
     • Matthew Burkholder and Rachel Greenstadt. Privacy in Online Review Sites. IEEE Workshop on Semantic Computing and Security (SPW 2012).
  2. Software
     • Our privacy detective code to mine Twitter and detect private information is available on github.
     • Github repository with source code for Obsecure final project (prototype project for sanitizing documents).
• Adversarial Stylometry. This project, funded by DARPA's Computer Science Study Group (CSSG) program, is developing the potential of automated linguistic analysis, or stylometry, for authorship recognition in modern, adversarial domains such as intelligence analysis and forensics. The intelligence community makes use of stylometry for entity disambiguation. However, levels of confidence in these techniques are critical for the actionability of the information. Our research in multilingual and cross genre stylometry will expand the cases where stylometry can be brought into play. Our work in open-world stylometry and adversarial stylometry will develop confidence measures that can assess how the information from them should be integrated with other entity disambiguation and attribution techniques.
• Secure Computing Research for Users' Benefit. This center is a collaborative project with UC Berkeley and Intel Labs. A guiding theme of the center is to conduct research that benefits end users of technology. In particular, a core goal of the center is to develop technology and knowledge that will make computing safer and more trustworthy for users. Our group works on developing robust security metrics and analytics for (1) Human-centric security in adversarial environments, (2) Machine Learning in adversarial environments, and (3) Large-scale Machine Learning for security.
• Active Authentication via Linguistic Modalities. This project is part of DARPA's Active Authentication Program and is a joint project with the Drexel Data Fusion Laboratory and Juola and Associates. We aim to adapt stylometric techniques for the authentication problem, gather a large data of routine computer use for development and testing of the modality, and integrate HCI modalities and stylometric techniques with a fusion algorithm to identify the context and circumstances under which this technology would be most useful.
• Socially Intelligent Community Filtering. This interdisciplinary NSF project applies both statistical machine learning and ethnographic studies to understand the mechanisms by which online communities regulate content online. This work uses machine learning to replicate crowd-sourced trust metrics on web forums such as Slashdot.
• Behavior-Based Access Control. In current enterprise environments, information is becoming more readily accessible even across domains due to recent advances in cross domain discovery and service invocation. However, a large portion of information still lacks annotations for identifying authoritative data sources, leaving information consumers without guidance on which information should be used. In addition, trustworthiness of documents and actors is not explicitly measured as part of the quality of information. In other words, missions operate unaware of how latest security events may have impacted the trustworthiness of the information being used and the actors involved. This leads to situations where producers give documents to consumers they should not trust and consumers use information from non-reputable documents or producers. This is especially true for cross domain information sharing due to identity management concerns. Our view of behavior-based access control (BBAC) unifies access control with trust management by not only considering whether actors are allowed to access documents, but assessing whether actors should consume or produce documents. In both cases, the question of trust can be applied to the actor and the document being manipulated and can be viewed as an access control decision in a general sense. This is a joint project with BBN and the Air Force Research Laboratory.
• Reconfigurable Antennas For Wireless Security. Channel characteristics for reconfigurable antennas are used as a means to certify identity in a wireless network. Using reconfigurable antennas creates a more robust classification which makes spoofing extremely difficult. Additional research is being done for creating security keys based on these channel characteristics. This research is done in collaboration with the Drexel University Wireless Systems Laboratory (DWSL).
• Robust Distributed Constraint Reasoning. Distributed constraint reasoning (DCR) has recently generated much interest due to its ability to solve many real world problems without centralizing all of the information. However, many DCR algorithms, however, are prone to failure if even a single agent fails. Further, while these algorithms have the capacity to reduce the unnecessary spread of information, they often share more information than a centralized algorithm would. This project seems to create algorithms for distributed constraint reasoning that are more robust to failure and leak less information.