Past Projects
Learning from Social Metadata
Many social Web sites allow users to create content and annotate it with descriptive metadata, such as tags, and to organize content within personal hierarchies. Structured social metadata offers invaluable evidence for learning how a community organizes knowledge. But such metadata also tends to be sparse, shallow, ambiguous, noisy and inconsistent. We are developing machine learning methods to aggregate social metadata to improve information discovery and learn common taxonomies (folksonomies).
Geospatial Social Networks | Harvesting Folksonomies
Mathematical Modeling of Multi-Agent Systems
Mathematical analysis is a crucial tool for understanding, predicting and controlling emergent behavior of autonomous agent systems, whether human or robot. We have developed a stochastic processes-based framework for mathematical analysis of multi-agent systems, and have successfully compared analytic predictions with empirical results in several robotics applications. These include foraging, collaboration and dynamic task allocation.
In addition, we have developed a methodology for designing multi-robot systems that uses machine learning to automatically synthesize robotsÕ behavior rules. In another project with collaborators at USC and Brandeis University, we are modeling collective behavior of museum visitors and validating models experimentally by recording actual museum visitors.
Robot Swarms | TASK | Human Crowds
Pattern Formation in Spatially Extended Systems
My graduate research on pattern formation in spatially extended systems sparked my fascination with emergent behavior in complex adaptive systems, which continues to inspire my work.