V.S. Rostovtsev, P.R. Potapov, E.S. Hramkov
The article offers a solution to the problem of data clustering that change dynamically over time. For this, in the proposed neural network model implemented the idea of "falling asleep" and "awakening" of neurons, depending on the situation.
Comparison of the proposed neural network model and the Kohonen network was held on the same training samples by the number of iterations required to achieve the same level of error.
The proposed neural network model more attractive than the Kohonen network, as it can "tweak" the structure of a neural network for input samples and selects as many clusters as provided in the input samples.
The developed model is suitable for clustering problems with dynamically changing in time the input vectors, where some clusters are lost and new clusters are appear.
For the experiments developed a program on a language in C#.