Today's fast-growing mobile device and online social media infrastructure is producing an unprecedented volume of data on communication, system and device-level interaction, and user behavior. Examples include cellphone data (including voice calls, SMS and Internet data, cellular signal strength) and activity on online social media (itself often sent from mobile devices). This presents both opportunities and challenges. On one hand, the increasing availability, high quality, and relatively easy collection of such data provides a way to better understand online human activity, infer activity patterns, and design new services. On the other hand, the combination of big data and powerful inferential techniques pose both computational challenges and privacy concerns. In both cases, there is a need for efficient algorithmic techniques that can support learning from mobile and social data sets, and that enable principled creation of synthetic data at similar scale for modeling, testing and anonymization purposes. This project aims to address problems in the above space by framing them in terms that are traditionally the domain of the physical sciences.
This project is based on the core insight that many problems in the above space can be framed in terms of pairwise interactions among spatially embedded entities, traditionally the domain of N-body problems in the physical sciences. As a consequence, revisiting and adapting N-body algorithms specifically to mobile and social data analysis and learning can increase our capacity to (i) work with such data at scale and (ii) do so in a privacy-preserving way. The methodological contributions will include novel N-body and parallel algorithms, specifically designed for processing hierarchical, and geospatially embedded, mobile and social data, where the size or access to datasets is prohibitive. In addition to the algorithmic design, this project will develop software modules (e.g., special purpose compilers, crowdsourcing tools and generators of synthetic datasets) that implement these methods. The design will be primarily applied to improve cellular network monitoring and control, mobile data privacy, and smart city applications.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
