Is it possible to include dropout in a fitting network?

I am trying to build a network for a regression problem and for this I would like to add a dropout layer into my network.

After some research I found out how to include dropout into a problem where the trainNetwork function is used. Since I am using a fitnet, the train function is used instead. So my question is if I can still add dropout to my network somehow and if I can, how to implement it?

I'd appreciate any help.

Answer 1

[Shashank Gupta]

Hi Carolin,

Input argument of fitnet function takes array as an input. This array contains the number of neuron to be set up in the model. This function is not yet customized to add different custom layer. There is a way you can add dropout layer, but things will become more tedious to handle. Customization of fitnet function can be done by accessing the output object, redefine the Layers property in the object and use it. Still I would suggest to just use trainNetwork, It gives you more flexibility to choose layer and even customize it.

I hope it helps.

[Carolin Bokelmann]

Hi Shashank,

first of all thank you for your answer. My problem with trainNetwork is that I am not sure which input layer to use. For my application I have concentrations, temperature and pressure as input and a reaction rate as output. So not an image neither a sequence as input.

Do you have any suggestions for me what input layer to use?

[Francesco Marcuzzi]

I am having the same problem. Apparently there is no way to implement a simple feedforward network with the deep learning design tool.