An Empirical Study: Extensive Deep Temporal Point Process

Temporal point process as the stochastic process on continuous domain of time is commonly used to model the asyn-chronous event sequence featuring with occurence timestamps. Because the strong expressivity of deep neural networks, they areemerging as a promising choice for capturing the patterns in asynchronous sequences, in the context of temporal point process. In thispaper, we first review recent research emphasis and difficulties in modeling asynchronous event sequences with deep temporal pointprocess, which can be concluded into four fields: encoding of history sequence, formulation of conditional intensity function, relationaldiscovery of events and learning approaches for optimization. We introduce most of recently proposed models by dismantling theminto the four parts, and conduct experiments by remodularizing the first three parts with the same learning strategy for a fair empiricalevaluation. Besides, we extend the history encoders and conditional intensity function family, and propose a Granger causality discoveryframework for exploiting the relations among multi-types of events. Because the Granger causality can be represented by the Grangercausality graph, discrete graph structure learning in the framework of Variational Inference is employed to reveal latent structures of thegraph, and further experiments shows that the proposed framework with learned latent graph can both capture the relations and achievean improved fitting and predicting performance.