2/10/04:                 Odelia Schwartz
                         Salk Institute<BR><BR>

Signal Statistics and Hierarchical Sensory Representations
<BR><BR>         
                                                                                                         
A critical question in the neural representation of the sensory
world, is how sensory systems can build multi-layer, hierarchical,
representations whose semantics are as uniform as possible at every
stage. I'll talk about some recent ideas suggesting that a particular
form of nonlinear transform on wavelet outputs can license effective
hierarchical representations. I'll discuss in relation to three
frameworks: Gaussian scale mixture models, ICA/sparse coding, and
divisive normalization.