Hawkes processes match neural TPPs on event timing, study finds

A new research paper re-evaluates the performance of Hawkes processes and neural temporal point processes (TPPs) on event sequence data. The study found that the performance gap between classical Hawkes processes and more flexible neural TPPs largely disappears when Hawkes processes are equipped with learnable baselines and kernels. The research also highlights that the advantage of neural TPPs primarily lies in modeling the timing of events rather than their type, with exceptions noted on complex medical datasets like MIMIC-IV. The authors have released their framework, TorchPP, as an open-source package to facilitate consistent implementation and comparison of TPP models. AI

IMPACT Clarifies the strengths and weaknesses of different temporal point process models, guiding future research and practical application in event sequence analysis.

RANK_REASON Academic paper evaluating and comparing statistical models for temporal point processes. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• Hawkes Processes
• Hendrik Alexander Mehrtens
• MIMIC-IV
• Neural Temporal Point Processes
• TorchPP
• transformer
• UK Biobank

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →