How to keep constant weights in a Neural Network?

8 views (last 30 days)
Hi, I am trying to train only certain parameters of neural network in MATLAB 2021a, but I have not found anything related in the help menu.
I have found that it is possible to keep constant all the parameters of a certain layer by changing the layer attribute "trainable" to false, but in this case I would like to train some parameters of the layer. (Ex. There are 6 parameters in the layer and I want to fix two and train the remanining ones).
Thanks for your attention.

Answers (1)

SALAH ALRABEEI on 5 Jun 2021
Assume u have 5 inputs 10 ne. (for only on hidden layer). First, generate yourself a random matrix of size 5x10, then save it. Then construct the net, but before the training, replace weight with the one you manually created as follows
if true
net.Layers(k).Weights = W;
where net is the name of ur ne, W is new weights, and k is the layer.
  1 Comment
Enrique Luna Villagomez
Enrique Luna Villagomez on 5 Jun 2021
Thanks for your answer, but if the train function is after that conditional, the weights would change at the end right ? I mean basically what I will be doing with that conditional is to initialize the weights.

Sign in to comment.




Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!

Translated by