Towards Out-of-Distribution Sequential Event Prediction: A Causal Treatment

Then we devise a new learning objective based on backdoor adjustment and further harness variational inference to make it tractable for sequence learning problems.