Retrieving activation data results for intermediate layers of a shallow neural network

Hi everyone,

I am looking for a function or another straightforward way to retrieve intermediate layer results (e.g. from the bottleneck layer) when training a shallow neural network which looks like this:

net = feedforwardnet(hiddenLayers,trainFcn)
net = configure(net,X,X);
net = train(net,X,X)
X_afterNN = net(X)

X are a series of modal parameters from a structure (i.e. input data that are not images) and I'd like to retrieve and further examine the intermediate outcomes of my NN.

In other words, I'm looking for an equivalent to the 'activations' function which works for deployed deep learning networks and returns intermediate layer activations for an image data and a specified layer, that is:

act = activations(workflowObject,image,layername)

Thank you for your help!

Answer 1

[Varun Sai Alaparthi]

Hello Ana,

Unfortunately, there is no direct function or straightforward approach as of MATLAB R2022b to directly get output of an intermediate layer of a shallow neural network.

However, as a workaround you can try to use the ‘dlnetwork’ class using which you can directly use the 'predict' function to get the output of any intermediate layer of the network.

[Y1,Y2,..] = predict(___,'Outputs',layerNames);

Use this code snippet to get the output of the specified layer

If you have any further queries, please feel free to reply to my answer

[gengoroshop gengoroshop]

This did not work for me. Can you please explain how to obtain outputs from intermeiate layers with an example of gloogLenet?