RP-AG-SOM: A growing self-organizing map with assymetric neighborhood function and variable radius

Abstract

To deal with the problem of unit exhaustion and high computational cost of Kohonen’s Self-Organizing Map (SOM), Growing SOM was proposed in last century. Meanwhile, to avoid topological twist of the map, Aoki and Aoyagi proposed an asymmetric neighborhood function to instead of the conventional symmetric one. Furthermore, Masuda et al. proposed a heuristic method named Radius Parallel Self-Organizing Map to decide the adaptive bounds of SOM during learning process. In this paper, we propose to compose these 3 methods to construct a novel efficient SOM named RP-AG-SOM: a Growing Self-Organizing Map with asymmetric neighborhood function and Variable Radius. Additionally, RP-AG-SOM is applied to a voice instruction recognition system successfully.