Interpretable Spatio-Temporal Point Processes

Summary

I work on point process models for event data where the goal is not only forecasting, but also understanding dynamics, stress-testing models, and supporting downstream scientific reasoning.

Research Question

What kinds of point process models remain interpretable while still being expressive enough for realistic spatio-temporal event data?

Methodology

Neural and physics-informed point process models
Benchmark design for spatio-temporal pattern complexity
Counterfactual analysis for event-generating processes

Key Contributions

Surveys and organizes the design space of neural spatio-temporal point processes
Builds evaluation frameworks for comparing event-model complexity
Studies counterfactual behavior and physically structured event models

Outputs

Publication: (Mukherjee et al., 2025)
Publication: (Aalaila et al., 2025)
Publication: (Mukherjee et al., 2025)
Publication: (Grossmann et al., 2024)

References

2025

event modeling

Neural Spatio Temporal Point Processes: Trends and Challenges

Sumantrak Mukherjee, Mouad Elhamdi, George Mohler, and 4 more authors

Transactions on Machine Learning Research, 2025

Bib Paper

@article{mukherjee2025nstpp,
  title = {Neural Spatio Temporal Point Processes: Trends and Challenges},
  author = {Mukherjee, Sumantrak and Elhamdi, Mouad and Mohler, George and Selby, David A. and Xie, Yao and Vollmer, Sebastian and Grossmann, Gerrit},
  journal = {Transactions on Machine Learning Research},
  year = {2025},
  themes = {event_modeling},
  note = {2025},
}

event modeling

HawkesNest: A Multi-Axis Benchmark for Spatio-Temporal Pattern Complexity

Yahya Aalaila, Sumantrak Mukherjee, Gerrit Grossmann, and 1 more author

ECML, 2025

Bib

@article{aalaila2025hawkesnest,
  title = {HawkesNest: A Multi-Axis Benchmark for Spatio-Temporal Pattern Complexity},
  author = {Aalaila, Yahya and Mukherjee, Sumantrak and Grossmann, Gerrit and Vollmer, Sebastian},
  journal = {ECML},
  year = {2025},
  themes = {event_modeling},
  note = {Under review},
}

event modeling

SQUID: A Bayesian Approach for Physics-Informed Event Modeling

Sumantrak Mukherjee, Sebastian Vollmer, and Gerrit Grossmann

In EurIPS DiffSys Workshop, 2025

Bib Paper

@inproceedings{mukherjee2025squid,
  title = {SQUID: A Bayesian Approach for Physics-Informed Event Modeling},
  author = {Mukherjee, Sumantrak and Vollmer, Sebastian and Grossmann, Gerrit},
  booktitle = {EurIPS DiffSys Workshop},
  year = {2025},
  themes = {event_modeling},
  note = {Workshop paper, 2025},
}

2024

event modeling

Peculiarities of Counterfactual Point Process Generation

Gerrit Grossmann, Sumantrak Mukherjee, and Sebastian Vollmer

In STCausal Workshop at ACM SIGSPATIAL, 2024

Bib Paper

@inproceedings{grossmann2024peculiarities,
  title = {Peculiarities of Counterfactual Point Process Generation},
  author = {Grossmann, Gerrit and Mukherjee, Sumantrak and Vollmer, Sebastian},
  booktitle = {STCausal Workshop at ACM SIGSPATIAL},
  year = {2024},
  themes = {event_modeling},
  note = {Workshop paper, 2024}
}