Computer Science · Artificial Intelligence
Disconnecting Self-Organizing Map
Haruna Matsushita, Yoshifumi Nishio
See on the mapAbstract
The Self-Organizing Map (SOM) has problems with some in-active neurons which have affected a result of clustering. In this study, we propose a new SOM algorithm which is the Disconnecting Self-Organizing Map (DSOM). In the initial state, all neurons of DSOM are directly or indirectly con-nected each other. However, connections between neurons located at distant locations are cut with learning. We can confirm that the result of using DSOM includes no inactive neurons and DSOM can obtain a more effective map reflect-ing the distribution state of input data than the conventional SOM. 1.