Convert a custom neural network to a dlnetwork or a layer of dlnetwork

Hi,

Is there a way to a convert a custom neural network to a dlnetwork or as a layer of a dlnetwork?

For example, I want to convert a simple custom neural network like below

net = network(1, 1, [0],[1],[0],[1]); % the number of input feature is 2.

to a dlnetwork or a layer of a dlnetwork.

I want this because I need a not-fully-connected-layer for dlnetworks.

I could make a not-fully-connected-layer easily using custom neural networks, but I could not use a custom loss function with custom neural networks.

I know that I can set all initial weights to 0, and disable updating all weights of a layer in dlnetwork by setting setLearnRateFactor of weghts to 0, but I could not set it for individual weights.

Thanks in advance.

Answer 1

[Angelo Yeo]

The ability to convert shallow neural network to dlnetwork is not available in Deep Learning Toolbox.

FYI, check out the linked doc to learn how to define custom deep learning layers: Define Custom Deep Learning Layers