Matlab deep learning network for regression

Hi,

I built a neural network using Feedforwardnet, which performed well for my requirements. I have multiple inputs and outputs, which are not image data. However, I discovered that the results from this shallow network cannot be exported to the ONNX format. Only deep networks such as SeriesNetwork, DAGNetwork, dlnetwork, or layerGraph can be used for this purpose. Can you recommend one of these networks to replace Feedforwardnet and provide some simple examples of their use for applications other than image classification?

P.S.: My ultimate goal is to export the network to Fortran. I assume that once the network is exported in ONNX format, it can be imported into Fortran. If this is not possible, please explain it to me.

[prasanth s]

try 'fitcnet'

Answer 1

[Sandeep]

Hi Mary,

It is my understanding that you are expecting recommendations on what Deep network to use as a replacement for Feedforwardnet such that it can be exported to the ONNX format.

You can prefer using either SeriesNetwork or DAGNetwork as they have essential functions for exporting the network in ONNX format.

It is suggested to re-create your shallow neural network using trainNetwork and give that network to exportONNXNetwork. trainNetwork can be used to train deep neural networks for non-image/non-sequence data.

@prasanth s suggested the use of fitcnet inplace of Feedforwardnet. Though feedforwardnet and fitcnet share a similar architecture, they are used based on the nature of the problem i.e, feedforwardnet is for regression problems whereas fitcnet is for classification problems.

There is also functionality of creating custom layers to implement custom functionalities based on your requirements. You can refer to the examples in the following documentation link:

Define Custom Deep Learning Layers - MATLAB & Simulink (mathworks.com)