Does 'traingd' function in nntraintool implement plain vanilla gradient descent? Or does it change the learning rate adaptively?

hiddenlayersize = [5 5];
net = fitnet(hiddenlayersize,'traingd');

Answer 1

[Gayathri]

Hello @Sai Srujana Vuppala,

The 'traingd' training function implements vanilla gradient descent. If you want to implement "Variable Learning Rate Gradient Descent" then you can use 'traindx' function.

For knowing about different training functions that can be used with "fitnet" function, please access the documentation with the following command.

doc fitnet

Hope this helps!