Re: initialization of weights and bias of neural network
- From: "Greg Heath" <heath@xxxxxxxxxxxxxxxx>
- Date: 11 Apr 2006 01:37:18 -0700
I want to establish several neural network models. Each model will be
trained for 50 times. The neural network are initialized with
function "init" before every training started. I found that the
initialization of weights and bias of the neural network seems
related with time. It is restored after the MATLAB is restored.
Now I want to restore the initialization when a new model is to be
trained for the first time without the need to close and restore the
MATLAB first, is it possible?
Yes. Just restore the RNG to its original state. Go to the CSSM
archives in Google Groups and search on
greg-heath integer-states -RNG
Hope this helps.
- Prev by Date: Swapping
- Next by Date: Hi can you solve it
- Previous by thread: initialization of weights and bias of neural network
- Next by thread: Matrix must be square error when using Quad