基于标准化流的灵活、高效Hawkes过程

Hawkes过程是一类可以描述自激励或者互激励现象的时间点过程，在地震预测、神经脉冲序列、高频金融交易等领域都有着广泛的应用。传统的Hawkes过程为了推断方便存在参数化假设，导致其表达能力受限。为了增强其灵活性，频率学派、贝叶斯学派和神经网络学派均提出了相应的方法。这些方法普遍存在推断或者采样效率低下的问题，严重影响Hawkes过程在大规模数据集场景中的应用。针对该问题，本项目提出了基于标准化流的灵活、高效Hawkes过程，立足Hawkes过程和标准化流两大研究领域，细分成三个研究阶段，包含了四个研究部分，共同聚焦于Hawkes过程的灵活性与高效性，旨在解决传统Hawkes过程的上述问题。

Hawkes processes are a class of temporal point processes that have the ability to model the self- and mutual-exciting phenomena. It covers a wide range of applications, e.g., seismology, neural spike train and high-frequency finance. The vanilla Hawkes processes are assumed to be parametric, which eases the inference at the cost of model expressiveness. To achieve better expressiveness, various flexible Hawkes processes have been proposed from the frequentist, Bayesian or neural network perspectives. However, all those models are inefficient in the inference or sampling, which hampers their application to large-scale problems. To address these issues, we propose the normalizing flow Hawkes processes (NF-Hawkes) in this project, which integrates Hawkes processes with normalizing flow to achieve the better flexibility and efficiency.