Learning Algorithm of Kohonen Network With Selection Phase
Mohamed Ettaouil, Essafi Abdelatifi, Fatima Belhabib · 2012
The learning algorithm of kohonen network realized away from any notion of class. So, the labelling phase is necessary to find the class associated to data. Generally, the size of the topological map is randomly chosen. Such choice effect the performance of the Kohonen algorithm. Consequently, the labelling phase becomes difficult to realize. To overcome this problem, we add to learning Kohonen algorithm a phase called selection stage. This phase is based on a function called selection function; to construct, we use a sub-set of the data set. In addition, we divided the topological map on two parts: The first contains the used neurons; while the second one is formed by the unused ones. It should be noted that the size of the first and the second parts are modified by the use of the selection function. To compare our method with the classical one, some experiments results are performed.