Neural Processes for Large-Scale Spatiotemporal DataProf. Rose Yu (Computer Science and Engineering, UC San Diego)

Earthquakes prediction, COVID-19 forecasts, and weather projection require accurate modeling of large-scale spatiotemporal data. Traditional approaches either rely on strong modeling assumptions, or are too slow for real-time decision-making. Neural processes significantly enhance the expressivity and scalability of existing methods with deep neural networks. In this talk, I will discuss two examples of neural processes. (1) Deep Spatiotemporal Point Process (DeepSTPP), a deep dynamics model that integrates spatiotemporal point processes. Our method is flexible, efficient, and can accurately forecast irregularly sampled events over space and time. (2) Interactive Neural Process (INP), a Bayesian active learning framework to proactively learn a deep learning surrogate model and accelerate stochastic simulation. Our framework can faithfully imitate the behavior of a complex infectious disease simulator with a small number of examples, enabling rapid simulation and scenario exploration. I will demonstrate the use case of these models on earthquake prediction and COVID-19 scenario creation.