Professor Geman’s research is in the areas of machine and natural vision, statistical theory and modeling, and inference for images, text, and neurophysiological data. His current projects, all of which are collaborations, include: (i) Compositional Vision - building a recursive (part/whole) hierarchical system that learns explicit and reusable representations of objects and their relationships. (ii) The Folding Problem - in biology, structure is function. A central challenge of the genomic era is to predict the three-dimensional structures of molecules (e.g. proteins and enzymatic RNAs) from their known molecular constituents. We are developing stochastic methods for efficiently computing folding pathways. (iii) Effective Timescale of Neurons - what carries information in neural circuits, spiking rates or the precise timings of individual spikes? The question, fundamental to any theory of representation and computation, remains unanswered and contentious. Geman and his collaborators are developing new statistical methods that use raw electrode data to deliver high statistical power for the presence of fine-temporal structure.
James Manning Professor, Division of Applied Mathematics
Consulting Deputy Director, Data Science Initiative