Neural Point Process for Forecasting Spatiotemporal Events

Zihao Zhou, Xingyi Yang, Xinyi He, Ryan Rossi, Handong Zhao, Rose Yu

Keywords: spatiotemporal point process, deep sequence models, time series

Abstract: Forecasting events occurring in space and time is a fundamental problem. Existing neural point process models are only temporal and are limited in spatial inference. We propose a family of deep sequence models that integrate spatiotemporal point processes with deep neural networks. Our novel Neural Spatiotemporal Point Process model is flexible, efficient, and can accurately predict irregularly sampled events. The key construction of our approach is based on space-time separation of temporal intensity function and time-conditioned spatial density function, which is approximated by kernel density estimation. We validate our model on the synthetic spatiotemporal Hawkes process and self-correcting process. On many benchmark spatiotemporal event forecasting datasets, our model demonstrates superior performances. To the best of our knowledge, this is the first neural point process model that can jointly predict the continuous space and time of events.

One-sentence Summary: A novel Neural Spatiotemporal Point Process model for irregularly sampled spatiotemporal event forecasting, which integrates deep neural networks with spatiotemporal point processes.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Reviewed Version (pdf): https://openreview.net/references/pdf?id=TFrxJzhI6f