Recent Advance in Temporal Point Process : from Machine Learning Perspective

Temporal point process (TPP) has served as a versatile framework for modeling event sequences in continuous time space. It spans a wide range of applications as event data is prevalent and becoming increasingly available such as online purchase, device failure. Tailored TPP learning algorithms are devised to different special processes, complemented by recent neural network based approaches. In general, traditional statistical TPP models are more interpretable and less data ravenous, which lay their success on appropriate selection of the intensity function via domain knowledge. In contrast, emerging network based models have higher capacity to digest massive event data with less reliance on model selection. However their physical meaning becomes less comprehensible. From machine learning perspective, this survey presents a literature review on these two threads of research. We walk through several working examples to provide a concrete disclosure of representative techniques.