Add a new layer to pre-trained neural networks

Hey,

I have three feed-forward neural networks in matlab:

...
[net1, tr1] = train(net1,X_train,Y_train) % First neural network
... // testing and evaluating net1
[net2, tr2] = train(net2,X_train,Y_train) % Second neural network
... // testing and evaluating net2
[net3, tr3] = train(net3,X_train,Y_train) % Second neural network
... // testing and evaluating net3
    

The neural networks are trained seprately. All NNs have the same architecture (122-25-1). I want to build a bigger neural network that adds a fully connected layer with softmax activation to the pretrained NNs, and train the new neural network on the new parameters and keep the "old" layers untouched.

I want to create net4 that has a final output layer which connects the pretrained NNs. This layer has three neurons that are comming from the 3 NNs, and one output neuron (with softmax activation max). When I train net4, I only learn the parametrs of the created layer, and not change the learned parameters for net1, net2 and net3.

Thank you for your help,

Bests,