The product-moment correlation coefficient is often viewed as a natural measure of dependence. However, this equivalence applies only in the context of elliptical distributions, most commonly the multivariate gaussian, where linear correlation indeed sufficiently describes the underlying dependence structure. Should the true probability distributions deviate from those with elliptical contours, linear correlation may convey misleading information on the actual underlying dependencies. It is often the case that probability distributions other than the gaussian distribution are necessary to properly capture the stochastic nature of single neurons, which as a consequence greatly complicates the construction of a flexible model of covariance. We show how arbitrary probability densities can be coupled to allow greater flexibility in the construction of multivariate neural population models.

This paper describes a neural network for approximation problems on the sphere. The von Mises basis function is introduced, whose activation depends on polar rather than Cartesian input coordinates. The architecture of the von Mises Basis Function (VMBF) neural network is presented along with the corresponding gradient-descent learning rules. The VMBF neural network is used to solve a particular spherical problem of approximating acoustic parameters used to model perceptual auditory space. This model ultimately serves as a signal processing engine to synthesize a virtual auditory environment under headphone listening conditions. Advantages of the VMBF over standard planar Radial Basis Functions (RBFs) are discussed.