# Define a custom linear layer

Hello,

Given a neural network model called S() and dataset X^{5000 \times 18} where 5000 is the number of samples and 18 its dimensionality.

At the first layer (linear) of S() l would like to learn X= X * A* theta such that A^{18 \times 18}

With A randomly initialized and theta the parameters of the model
And A=B * B^{T} with B^{18*60} randomly initiliazed.

I would like to learn A which is the same for all the samples (A as the parameter of model) .

Any trick to do that ?

Thank you