Neural Point Process

Neural point processes (NPPs) are a class of models using neural networks to predict the timing and location of events in continuous time and space, addressing limitations of traditional point process models. Current research focuses on improving model efficiency and expressiveness through architectures like variational autoencoders and incorporating spatiotemporal dependencies via graph neural networks, often within a Bayesian framework for better uncertainty quantification. These advancements enable more accurate predictions in diverse applications, such as traffic congestion forecasting, event sequence modeling, and spatiotemporal dynamical system analysis, improving decision-making in these domains.