My input features for LSTM are different dimension, Is it possible to combine and train the network

I have 6 input features for trining the LSTM network, but 1st coloumn has 2x28(contain 2 feature of 28 length) cells by 100 and all four are 1x100 each, Is it possible to combine all the feature and give it to network. since for LSTM we are using (sequenceInputLayer(N)) as input layer. Is there any alternate solution to this.

Answer 1

[Hiro Yoshino]

No, you cannot do that I think.

Why don't you conver the singular values into series data by using remat and put them under the 2x28 series data so you can use LSTM?

Another way to do this is creating a network with multi-inputs.

see: https://www.mathworks.com/help/deeplearning/ug/train-network-on-image-and-feature-data.html

[Manoj Devaraju]

Thank you so much for your response. I have tried first approach you mentioned worked well.

but creating a network with multi input, I tried to create but failed.

Can you please show me how to create multi-inputs network for my problem.