Re-evaluation Shows Flexible Classical Models Match Neural Temporal Point Processes

Summary

A systematic study comparing classical temporal point processes (TPPs) with neural TPPs across five benchmarks and two large medical datasets finds that flexible Hawkes processes with learnable baselines and kernels close the previously reported performance gap. A Transformer with a spline-based intensity head matches intensity-free models, with gains mainly in modeling event timing rather than event type. The authors release TorchPP, an open-source framework with consistent implementations.

This is a reviewed summary. Read the complete article at the original source.

Why it matters by industry

• Technology

  This research provides reproducible evidence that flexible classical models can match neural TPPs, with implications for model selection in software and data science applications, and includes an open-source framework (TorchPP) for the industry.