How to apply constraints on the training process of neural networks? Like turn off the learning of some weights.
4 views (last 30 days)
Show older comments
"Constraints that hold entire weight matrices and/or bias vectors to be constant by turning off their learning, using commands such as:
net.inputWeights{i,j}.learn = false; net.layerWeights{i,j}.learn = false; net.biases{i}.learn = false"__
For example, we now have a single-hidden layer neural network. We then turn off the learning of the output weights of hidden layer, says LW. Which case will happen in MATLAB?
Case 1: the gradient descent operation only perform on the rest of weights and biases(which are not turned off). We don't calculate the gradient deviation with regard to LW, because we don't need it.
Case 2: the gradient descent operation perform on the whole parameters including the the output weights of hidden layers. We calculate the gradient deviation with regard to LW, although we don't need it. But the gradient deviation will not be applied to LW.
Is there a way to make the 1st case happen? THANKS A LOT.
1 Comment
Greg Heath
on 14 Mar 2014
If you cannot find anything in
type train %Or what ever training function you are using
Try contacting the Matlab staff directly.
Answers (0)
See Also
Categories
Find more on Sequence and Numeric Feature Data Workflows in Help Center and File Exchange
Products
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!