How to define a custom layer in deep learning toolbox which makes multiple copies of a matrix ?

I have to design a deep fully connected neural network for regression task. I am using examples from the deep learning toolbox for images. My input matrix size and precticted matrix Y has the same size of [1 1 32]. But my target matrix T has size [168 1 32]. Since the regressionLayer requires the predicted Y and target T matrices to be of same size, for the purpose of my loss function I need to replicate the predicted matrix Y 168 times.

So, I want to know how to add a Layer while designing neural network which just does repmat() before I pass it to the last regressionLayer.

Thank you.

[Jyotsna Talluri]

Can you share your model and dataset?

Answer 1

[Hon Wah Yeung]

I kind of have this problem a while ago but now fixed it. It can be done by the following, just one line.

suppose the [1 1 32] is your X1, and you want N row copies of this

Ones = ones([N 1],'like',X1);

Z=Ones.*X1;

and the 2019b version will automatically do the backward function for you.