Improving Neural network models for dynamic graphs

* Time Prediction: one of the core machine learning tasks on dynamic graphs is link prediction: Given a future timestamp and a pair of nodes, the network outputs whether a link exists between the two nodes at the given timestamp. While useful in some scenarios, many practical applications demand that the network predicts when the next interaction between two nodes is taking place. A tailor-made architecture will most likely use temporal point processes (generalisation of a Poisson point process, see also DyRep paper linked below) or neural ODEs.