A Spherical Basis Function Neural Network for Modeling Auditory Space
Dr. Rick Jenison
Department of Psychology
University of Wisconsin-Madison
jenison@wavelet.psych.wisc.edu
2:30 pm Fri. Nov. 4 in 2310 Computer Sciences and Statistics Bldg.
This talk will describe a novel neural network for approximation problems on
the sphere. The von Mises basis function is introduced, whose activation
depends on spherical rather than Cartesian input coordinates. Spherical data
arise in many areas of science such as geophysics and meteorology, and
projection geometry has been the focus of cartographic interest for problems
involving a transformation of the spherical earth into a planar map. The
architecture of the von Mises Basis Function (VMBF) neural network will be
presented along with the corresponding gradient-descent learning rules for
optimizing the position and shape of the spherical basis functions. The VMBF
neural network has been used to solve a particular spherical problem of
approximating acoustic parameters used to model perceptual auditory space.
This model uses digital signal processing techniques to synthesize a virtual
auditory environment under headphone listening conditions. Advantages of using
VMBFs over standard Radial Basis Functions (RBFs) for spherical approximation
will be discussed.