Machine Learning and Point Processes for Insights into Earthquakes and Volcanoes

最近在处理地震数据方面的进展导致全球现有地震目录的规模急剧增加。然而，这些数据以及地震学领域相对没有引起机器学习界的注意。在我的博士学位中，我的目标是通过调查哪些机器学习方法可以应用于地震预测任务来弥合这一差距。更具体地说，由于地震事件通常被建模为点过程，我将着眼于开发学习点过程的新方法。这将从研究神经点过程模型是否比最新的参数模型--流行病类型余震序列模型--更好地捕捉地震的真实模式开始。在发展点过程方法的基础上，我试图创建可供地震学界在实践中使用的软件，以试图减少地震灾害对人类生活的影响。

Recent advances in the processing of seismic data have resulted in a drastic increase in the size of earthquake catalogs available globally. This data as well as the field of Seismology, however, has gone relatively unnoticed by the machine learning community. In my PhD I aim to bridge this gap through an investigation into what machine learning methods can be applied to the task of earthquake forecasting. More specifically, since earthquake events are generally modelled as a point process, I will look at developing new methodology for the learning of point processes. This will begin with an investigation into whether neural point process models can better capture the real patterns of earthquakes than the most up to date parametric model, the Epidemic-Type Aftershock Sequence model. Further to developing the methodology of point processes, I seek to create software that the seismology community can use in practice in an attempt to reduce the impact on human lives caused seismic hazards.