Stochastic Systems Group (SSG)

The Stochastic Systems Group (SSG) is led by Professor Alan S. Willsky, with additional leadership from Dr. John Fisher, Principal Research Scientist in the Computer Science and Artificial Intelligence Laboratory (CSAIL). The group includes graduate students, primarily based in LIDS but also from CSAIL, and several postdoctoral researchers and scientists. SSG has collaborative research efforts with colleagues and students in several other academic departments, laboratories, and centers at MIT, as well as with a number of other universities.
 
The focus of research within SSG is on developing statistically based methodologies for complex problems of information extraction and analysis from signals, images, and other sources of data. The work extends from basic mathematical theory to specific areas of application. Funding for this research comes from sources including the Air Force Office of Scientific Research, the Army Research Office, MIT Lincoln Laboratory and the Royal Dutch Shell Corporation.
 
One of the major research thrusts in SSG is in statistical inference and information fusion for complex graphical models. This research continues to yield new classes of models and signal and image processing algorithms that have provable performance properties and are scalable to very large problems. Recent applications are in computer vision, mapping of geophysical phenomena, and tracking of multiple vehicles from multiple sensors. Among the most recent advances are new classes of models that represent complex phenomena at multiple resolutions or granularities. Methods developed in SSG are being used by research and engineering organizations including Shell Oil, Lincoln Laboratory, and BAE Systems. This part of SSG’s research portfolio has received considerable international attention, including a string of best paper awards and extensive citations and influence on the work of others in fields ranging from chemical engineering to groundwater hydrology. 
 
Another area of focus is on curve evolution algorithms for the segmentation of imagery and extraction of the geometry from complex multimodal data. Recent accomplishments include machine learning methods that perform segmentation while learning the statistical differences between the regions being segmented, tracking dynamically evolving shapes, and capturing the inherent uncertainty in extracted geometry through curve-based Monte Carlo simulation methods. These methods have been applied to problems in medical image analysis and geophysical mapping. Research in this area has also received considerable recognition, including a recent best paper award.
 
An increasingly important component of research is in machine learning for the extraction of statistical models, usually in graphical form, of complex phenomena. One part of this research deals with the learning of “hidden” explanations of the complex behavior of observed data. A major new thrust is the development of so-called nonparametric methods for the extraction of behavioral models for dynamically evolving phenomena. Led by Emily Fox (a student of Professor Willsky) in collaboration with colleagues at UC Berkeley, this research has received considerable publicity and recognition.