Fast and Flexible Temporal Point Processes with Triangular Maps

TL;DR

TriTPP introduces a non-recurrent, normalizing flow-based temporal point process model that enables parallel sampling and likelihood computation, significantly improving speed while maintaining flexibility, and is applicable to variational inference in complex systems.

Contribution

The paper presents TriTPP, a novel non-recurrent TPP model leveraging normalizing flows for parallel computation, enhancing speed without sacrificing model flexibility.

Findings

TriTPP achieves orders of magnitude faster sampling than RNN-based models.

The model maintains comparable flexibility to recurrent models.

Demonstrated effectiveness on synthetic and real-world datasets.

Abstract

Temporal point process (TPP) models combined with recurrent neural networks provide a powerful framework for modeling continuous-time event data. While such models are flexible, they are inherently sequential and therefore cannot benefit from the parallelism of modern hardware. By exploiting the recent developments in the field of normalizing flows, we design TriTPP -- a new class of non-recurrent TPP models, where both sampling and likelihood computation can be done in parallel. TriTPP matches the flexibility of RNN-based methods but permits orders of magnitude faster sampling. This enables us to use the new model for variational inference in continuous-time discrete-state systems. We demonstrate the advantages of the proposed framework on synthetic and real-world datasets.