Is the validation data in a Neural Network to fit a function using Bayesian Regularization irrelevant?

I'm using Bayesian Regularization to train a Neural Network and I notice the performance graph doesn't show the validation data. The algorithm doesn't use the validation data, in my understanding. Is it possible/viable to set to zero the percentage when training with Bayesian Regularization ?

Answer 1

[Greg Heath]

That is a MATLAB design flaw. The older versions of TRAINBR do not allow a val subset. I have complained several times about that, especially since recent publications have shown that a combination of Bayesian Regularization and Validation Stopping is superior to just one.

What version of the NN Toolbox do you have?

You should be able to combine VALSTOPPING and MSEREG with TRAINLM. I'm not sure about doing it with TRAINSCG.

If you are successful with the combination, please post.

Hope this helps.

Thank you for formally accepting my answer

Greg

Answer 2

[Lorenzo Ridolfi]

Thank you for the answer. I'm using 2016b. Do you have any link to a publication about combination of Bayesian Regularization and Validation Stopping?

Best Regards, Lorenzo