Abstract

The problems faced by the brain when dealing with odor signals are many and extremely complex.  Odors are multi-dimensional objects, which we usually experience as unitary percepts.  They are also noisy, variable and yet, we can classify and identify them well, for example over many orders of magnitude of concentration.  Olfactory systems have thus found solutions to complicated pattern learning and recognition problems.  We propose that part of this solution relies on a particular architecture that in turn imposes a dynamic format on odor codes: the olfactory system would thereby actively create a very large coding space in which to place odor representations and simultaneously optimize their distribution within it. This process uses both oscillatory and non-periodic dynamic processes that each appears to serve complementary roles: non-periodic processes enable decorrelation whereas oscillations enable sparsening and feature-binding.  I
will summarize evidence supporting these propositions.