How to build the architecture of the Kohonen hybrid network?

Question

Neural networks

Hello, I want to connect a Kohonen network and a multilayer perceptron. I found several articles on this subject, but they did not give a complete picture of the work of this bundle. At the output of the Kohonen network, I get a winner neuron, what data for training should I submit to the perceptron, a normalized array of the winner neuron and losers as input, and the initial data vector as output?
And in the end, after training, I must again run the input data vectors through two networks or only into the perceptron. Or in general, the perceptron is needed to correct the weights and then use them in the Kohonen network.
I'm just starting to understand all this, so I apologize if I wrote any contradictory judgments.
I solve the problem of clustering.
Thanks