Abstract: The problem of neuro-fuzzy Kohonen network self-learning with fuzzy inference in tasks of clustering in conditions of overlapped classes is considered. The basis of the approach are probabilistic and possibilistic methods of fuzzy clustering. The main distinction of the introduced neuro-fuzzy network is the ability to adjust the values of fuzzifier and synaptic weights in on-line mode, as well as the presence except convenience Kohonen layer an additional layer to calculate the current values of the membership levels. The network characterized of computational simplicity, and is able to adapt to data uncertainty and detect new clusters appearance in real time. The experimental results confirm effectiveness of the approach developed.

Keywords: clustering, neuro-fuzzy network, self-learning algorithm, self-organizing Kohonen map.

ACM Classification Keywords: I.2.6 Artificial Intelligence - Learning - Connectionism and neural nets

Link:

ADAPTIVE NEURO-FUZZY KOHONEN NETWORK WITH VARIABLE FUZZIFIER

Yevgeniy Bodyanskiy, Bogdan Kolchygin, Iryna Pliss

http://www.foibg.com/ijita/vol18/ijita18-3-p02.pdf