Social awareness is pivotal for those who work with data analytics and is a key factor that affects the uses, benefits, and risks of big data. It is a common practice for both government agencies and private entities to collect and integrate large amounts of many different kinds of data, process it in real time, and deliver the product or service to consumers. There are increasing worries that both the acquisition and subsequent application of big data analytics could cause various privacy breaches, render security concerns, enable discrimination, and negatively affect diversity in our society. All these concerns affect public trust regarding big data analytics and the ability of institutions to safeguard against such negative social outcomes. As such, social awareness should be an integral part of research and training in the area of data analytics. In this theme, we are focused on developing cutting-edge socially aware data analytics to address social concerns and meet laws and regulations in national-priority applications, thus better enabling big data analytics to promote social good and prevent social harm.
Our major research goals are to develop novel techniques to provide privacy preservation, fairness, safety, and robustness to a variety of data analytics and learning algorithms including automated data curation, social media and network analysis, and deep learning, and ensure the adoption of the developed techniques meet regulations, laws and user expectations.
- Privacy-Preserving and Attack Resilient Deep Learning
- Faculty Lead: Xintao Wu
- Identify potential vulnerabilities of deep learning algorithms
- Develop a universal threat- and privacy-aware deep learning framework
- Conduct comprehensive evaluations of the proposed framework and models
- Socially Aware Crowdsourcing
- Faculty Lead: Chenyi Hu
- Improve crowdsourcing data quality with considerations of uncertainty
- Enhance available inference and learning models with novel algorithms for improved effectiveness and efficiency
- Verify and validate the robustness and trustworthiness of information from crowdsourcing data
- User-centric Data Sharing in Cyberspaces
- Faculty Lead: Ningning Wu
- Investigate on personal identifying information and their privacy issues
- Investigate appropriate multimodal deep learning techniques to identify discriminative and stigmatizing information
- Develop a user-centric privacy monitoring and protection framework
- Deep Learning for Preventing Cross Media-Discrimination
- Faculty Lead: Lu Zhang
- Explore deep learning-based techniques to detect cross-media discrimination
- Develop deep learning models and a causality-based deep learning framework for robust hate speech detection
- Cryptography-Assisted Secure and Privacy-Preserving Learning
- Faculty Lead: Qinghua Li
- Develop cryptography-aware privacy-preserving machine learning methods, and develop privacy protection for classification input data in machine learning applications.
- Causality-based Fairness in Social Networks
- Faculty Lead: Lu Zang
- Develop a deep-learning model for causality-based fair node classification in social networks.
Advancing the State of the Knowledge
Our developed technology can achieve meaningful and rigorous privacy protection when mining private data or collecting sensitive data from individuals; ensure non-discrimination, due process, and understandability in decision-making; achieve safe adoption, and robustness of machine learning and big data analytics techniques, especially in adversarial settings; and help incorporate social awareness in domain- or application-specific projects. Our research projects in this theme will advance the state of the knowledge through the goals and objectives listed above.