Emergent online learning with a Gaussian zero-crossing discriminant function

This paper presents a modified Gaussian zero-crossing (GZC) discriminant function with a restricted receptive field width for realizing emergent online learning. An important advantage of the GZC function over existing linear discriminant functions is its locally tuned response characteristics. By using the GZC discriminant function, both incorrect interpolation and incorrect extrapolation of trained networks can be significantly prevented by adjusting two threshold limits of networks. We demonstrate that the trained networks based on the GZC discriminant function have the proper capability for rejecting unknown inputs