Using MaxPool1D with on inputs with 4 dimensions (4D tensor)?

Question

I am trying to create a NN architecture for multiple instance learning, so the instances are actually bags of of time series segments. I specify input as having 4 dimensions and