Generative Adversarial imitation learning

Hi everyone, excuse me for a such question that could be silly ! my question is if we can apply the generative adversarial imitation learning algorithm on a time-series data to make predictions (instead of supervised and unsupervised learning). so the objective will be to mimic the temporal behavior of an agent.

in this case the GAIL will learn from a time-series data of fuel consumption of a car I just say that as example but it may be any variable that is varying with time.
So here the expert demonstrations will be state-action pair from the dataset which is features-prediction (state=features and action=prediction) and the objective of GAIL is to learn the policy behind the taken action (i.e. to mimic the expert behavior, predictions of the model should be as close as possible to the expert demos)

There was a “generation by imitation” approach proposed at NeurIPS 2021: