Multi Input Multi Output (MIMO) for time series forecasting


I’m dealing with a problem where I have several different sources of multivariate time series, and these sources have an interaction, in a way that the dynamics of the overall system is affected by the dynamics of each source.

I would like to try a MIMO model (using LSTMs OR TFTs) for dealing with this scenario. My idea is to model each source as an input and consider the separate outputs of each source as the outputs of my network.

I think that in this way I would have a stronger supervisor signal for my model and it would be able to learn the interactions between the several sources.

Is there any good example of code using Pytorch that implements a MIMO model that is similar to this one?