I want to know how to train LSTM neural network for regression with multiple inputs and multiple outputs?

I am fine with training a single input , single output LSTM neural network for regression using Matlab Deep Network Designer.But when working with multiple inputs and outputs, it is difficult to Standardizing the training Data

anybody can help me in this regard.

Thanks.

Answer 1

[Krishna]

Hello Iasini,

From what I gather, you're looking to develop a neural network capable of handling multi-input, multi-output sequences in MATLAB. My recommendation would be to utilize dlnetworks for this purpose. Essentially, multi-input, multi-output sequence processing networks have a similar structure, and the process of expanding the recurrent structure is identical. The difference lies in the computation of the loss for each observation within the sequence. Training the network is quite straightforward. You can prepare your training data in a format of axbxc, where 'a' represents the number of sequences in your training set, 'b' is the number of observations per sequence, and 'c' stands for the feature count of each sequence. Ensure that when you implement a recurrent layer, such as lstmLayer, you set the 'OutputMode' to 'sequence'. This ensures that the loss is computed at every timestep for a sequence. For additional details on lstmLayer, please visit the following link:

https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.lstmlayer.html

Moreover, you can find more information on how to train multi-input, multi-output networks using dlnetworks in this example:

https://www.mathworks.com/help/deeplearning/ug/sequence-to-sequence-regression-using-deep-learning.html

While it's feasible to construct a multi-output network using the Deep Network Designer, the tool does not support training such networks with trainNetwork. To train a multi-output network, one approach is to design the network with Deep Network Designer, export it to the workspace, convert the layer graph into a dlnetwork, and then apply custom training. Here's an example that guides you through creating a multi-output layer graph, transforming it into a dlnetwork, and then training the multi-output network with custom training methods:

https://www.mathworks.com/help/deeplearning/ug/train-network-with-multiple-outputs.html

I hope you find this information useful.