Use trainNetwork funcion to train a feedforwad net

Does the trainNetwork function can only be used to train CNNs and LSTMs as written in the documentation or can I use it to train a feedforwad net as well?

My idea was to create the ImageInputLayer the size of my feature vector and then use a couple of fullyConnectedLayers followed by RELULayers and dropoutLayers. The reason to want to use this intead of the regular feedforwardnet and train functions are the possibilities to use ReLU and dropout, more flexibility to choose the training parameters, and the usage of the ImageDatastore structure.

Will this method give me similar results to when feedforward and train are used or the way the trainNetwork function is implemented is not recommended for feedforward-only training?

Answer 1

[Saeed Mohammadi]

Hi,

I think this is a very interesting question. Can someone comment on this?

Thanks.

[Raphael Ruschel]

I tried using it as described on the question and it achieved good results, so I think it is not a problem!