Question: Can a multivariate MIMO LSTM learn the relationships between the multiple feature outputs?

This question arose when I decided to modify a multivariate (Multiple Input - Single Output, MISO) LSTM to forecast more than 1 feature of multiple future timesteps. What I ended up with was a multivariate MIMO (multiple output) LSTM to forecast multiple features with each forward pass.

For some context, here are the shapes of my inputs and labels before and after the modification:

# MISO Multivariate LSTM
X.shape = (128, 18, 4) # (batch, seq_len, features)
y.shape = (128, 6) # (batch, seq_len, features)

# MIMO Multivariate LSTM
X.shape = (128, 18, 4) # (batch, seq_len, features)
y.shape = (128, 6, 2) # (batch, seq_len, features)

It is without a doubt that the learning of one feature output can affect the learning of the other as they all come from the same forward pass through the same model, but does the model learning the complex relationships between each feature output?

Is learning accelerated/improved/supplemented in a MIMO scenario as compare to a MISO one?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.