LSTM Example for Multi input and Multi outputs

I have seen many examples for multi input single output regression but i am unable to find the solution for multi output case.I am trying to train the LSTM with three inputs and two outputs.I am using sequence-to-sequence regression type of LSTM.The predicted outputs are of same value or the predicted outputs are wrong.I tried changing the training parameters but nothing worked.Please suggest some solution to work on LSTM with muti output case.This is my training graph and loss never settles to zero.

[Percy Hu]

Hi，i also have the same question.

As i know, multi-output network must be completed by defining custom training progress.

However, during the mentioned progress, the function 'dlgradient' which must be used in it does not support the sequence layer like lstm or gru.

Hope someone else to solve the issue.

[Samuel Somuyiwa]

dlgradient can be used with lstm or gru except when the workflow requires computation of higher order derivatives. See the Limitations Section of dlgradient's doc page for details.

Answer 1

[Samuel Somuyiwa]

You can train a multi-output LSTM network using a custom training loop. Here is an example of how to train a network with multiple outputs: https://www.mathworks.com/help/deeplearning/ug/train-network-with-multiple-outputs.html

You may also find this doc page helpful: https://www.mathworks.com/help/deeplearning/ug/multiple-input-and-multiple-output-networks.html.