questions about SOM neural network

[sawaak]

Hi,
i have certain confusions about Self-Organizing Maps in neural networks.

As we know, we have to assign random weights as initial values in a SOM network. what is the rational behind this?? what will happen if we assign all the initial weights equally?? or assign them a value of zero??

waiting 4 quick reply..


thanks
sawaak

 

[Beetroots]

With most SOM algorithms you don't have to assign random weights. If all the initial stored reference vectors were the same the network will still train correctly.

 

[sawaak]

Hi,
but if we assign equal initial weights to every neuron, then each neuron will behave in the same manner to inputs, so we can not get a 'winner neuron' in this case, then how the process proceeds??


thanks
sawaak