I'm interested in how the collective dynamics of neural systems make
information processing and computation possible.  Neural
systems exhibit internally generated activity which is often much stronger than
any externally stimulated perturbations.  The structure of this variability has
yet to be fully understood and so historically it has been treated as "noise"
to be averaged out.  Yet it is difficult to reconcile this viewpoint with the fact
that we can reliably perceive complex stimuli at short (100s of milliseconds) time scales,
time scales which would seem to preclude the possibility of  temporal averaging.
I believe that such robust computation in what is apparently a very noisy system must
rely on conserved structure in the collective interactions between neurons 
rather than the noisy responses of individual neurons themselves. My research
focuses on understanding collective behavior at scales ranging from small 
(tens to hundreds of neurons) to large (the interactions between functional 
regions across the whole brain.

There are two main themes to my research.   The first involves the development
of statistical tools optimized specifically for understanding experimental neural 
data.  The complexity of neural systems dictates that most of what we learn
must be gleaned from experiments, in contrast to say physics where much 
has been learned through theory.  I collaborate with several labs performing
electrophysiological experiments which record action potentials (spikes)
and local field potentials (brain 'waves').  Projects currently ongoing 
involve both the visual and motor cortices of awake behaving primates
and also the somatosensory (barrel) cortex of awake and anesthetized 
rodents.  I develop various statistical models, often generalized linear models,
specifically geared towards testing experimental hypotheses.

The second theme of my research is more theoretical and involves thinking
about how the structure and complexity of collective neural activity can be used 
for actual computation.  Many studies have quantified the *information* present 
in neural data, but information is closely related to entropy which quantifies
randomness.  However randomness is devoid of any *relevant* information, what
is important , I feel, is the complexity of structure.  One line of inquiry being 
pursued are how the statistical complexity of neural activity changes during
different brain states and behaviors.  Another involves treating recurrent networks
of neurons as classifiers.  It has been shown that plasticity can allow such
recurrent networks to self organize into "reservoir computers" which can 
perform classification, even when there is a great deal of noise at the single
neuron level.