Robust Inference for Delayed Point Processes

Reporting delays can distort the observed timing and likelihood structure of event sequences. This project develops a correction-centered Wasserstein distributionally robust optimization framework for point-process inference under delay-model misspecification.

The method combines event-level backward timestamp correction with sequence-level robustness, hedging against uncertainty that remains after correction. It is evaluated in controlled nonhomogeneous Poisson-process simulations and on California utility-reported wildfire events.

Correction-centered robust inference for delayed point processes

Timestamp correction moves delayed observations toward the latent process, while distributional robustness hedges against the remaining misspecification.

(Deng et al., 2026)

References

2026

Distributionally Robust Inference for Delayed Point Processes

Zewei Deng, Shuyi Chen, Shixiang Zhu, and 1 more author

Working paper. Workshop version forthcoming at the INFORMS Data Mining and Decision Analytics Workshop , 2026

Extended manuscript in preparation for submission to the Annals of Applied Statistics (AOAS).

Bib

@unpublished{deng2026robust,
  title = {Distributionally Robust Inference for Delayed Point Processes},
  author = {Deng, Zewei and Chen, Shuyi and Zhu, Shixiang and Xie, Liyan},
  note = {Extended manuscript in preparation for submission to the Annals of Applied Statistics (AOAS).},
  year = {2026},
}