Reputation: 66975
So there is grate sample (only one real sample we found). And it is quite limiting. It shows how to create an architecture of artificial neutral network where all neurons of one layer are connected (forward) to all neurons of following (next) layer. We want to manage this connections so to create a ANN where each neuron of second layer is only connected to 2 neurons of next layer. so we want:
Layer 1: 2 neurons (2 inputs)
Layer 2: 3 neurons (hidden layer) (each neuron is connected to 2 neurons of next Layer (3) )
Layer 3: 3 neurons (hidden layer)
Layer 4: 1 neurons (1 output)
Upvotes: 3
Views: 780
Reputation: 1837
May be you can initialize the we weight for the connection you dont want to zero and make the learning rate for those weights to zero.
Upvotes: 0