Data Mining Methods for Graph Discovery in Neuroinformatics

Neuroscientists are beginning to collect activation (spike-train) data
from hundreds of neurons with millisecond precision. Analysis of
these large data sets poses interesting data mining challenges. We
describe computational methods and associated significance tests to
discover sequential patterns in multi-neuronal spike trains. By
discovering these patterns, we are able to uncover the functional
connectivity (graphical structure) of the underlying neuronal networks
and observe their time-evolutions. We illustrate these on simulated
and real data sets and compare the data mining methods with
model-based estimation methods. We conclude with a brief discussion of
Hebb cell assemblies and neural codes and how data mining can help
discover them.
Unnikrishnan is a staff research scientist at the General Motors R&D
Center, Warren, Michigan. His research interests concern neural
computation in sensory systems, correlation-based algorithms for
learning and adaptation, dynamical neural networks, and temporal data
mining. Before joining GM, he was a postdoctoral member of the
technical staff at AT&T Bell Laboratories, Murray Hill, New Jersey. He
has also been an adjunct assistant professor at the University of
Michigan, Ann Arbor, a visiting associate at the California Institute
of Technology (Caltech), Pasadena, and a visiting scientist at the
Indian Institute of Science, Bangalore. He received the PhD degree in
Physics (Biophysics) from Syracuse University, Syracuse, New York, in
1987.