training "large scale" neural network

Hello,

I am learning to train a neural network to get good regression results (curve fitting). I have matrix size of 3600*1000 as input, and my output is 1*1000. So each sample in the input has 3600 parameters. According to some credible reference tackling similar problems, I am using 1 hidden layer with 50 neurons. I am using "feedforwardnet" function. I would like to ask what's the reasonable step/choice of choosing a training function, and parameters setup? What's the good place to look at for detail information on selecting training functions and setting up properties for large scale problems?

Thank you all in advance.

Answer 1

[Srivardhan Gadila]

You can refer to the following documentation pages: Choose a Multilayer Neural Network Training Function, Improve Shallow Neural Network Generalization and Avoid Overfitting & Modeling and Prediction with NARX and Time-Delay Networks.

You can also try using Deep Neural Networks over Shallow Neural Networks for regression.