Modification of a geometrical algorithm of characterization k-valued threshold functions

The article provides an overview of the known approaches to the characterization (learning) k-valued threshold functions. Proposed a new algorithm characterization of k-valued threshold functions, which based on a known geometrical algorithm. For this new algorithm proved it's convergence. Also we give the results of experimental comparisons between new algorithm and a known geometrical algorithm of characterization and Obradovic learning algorithm.