2

Self-Attention Between Datapoints: Going Beyond Individual Input-Output Pairs in Deep Learning

Modelling Videos of Physically Interacting Objects