Neural Spatiotemporal Point Processes: Trends and Challenges
Authors: Sumantrak Mukherjee, Mouad Elhamdi, George Mohler, David Antony Selby, Yao Xie, Sebastian Josef Vollmer, Gerrit Großmann
TMLR 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The existing literature on neural STPPs mainly underscores their applications in settings where forecasts of when and where events occur support operational decisions. We organize this section by domain: natural disasters, crime, traffic, and epidemiology, because each area requires different modeling choices concerning space, time, and contextual factors. ... Performance is reported using R2 for rate prediction, accuracy for zone classification, and error metrics such as MAE, MSE, and RMSE; the neural components are compared against ETAS using these measures. ... Section 5 Evaluation Metrics |
| Researcher Affiliation | Academia | Sumantrak Mukherjee EMAIL Data Science and its Applications, DFKI; Mouad Elhamdi EMAIL College of Computing, Mohammed VI Polytechnic University and Data Science and its Applications, DFKI; George Mohler EMAIL Department of Computer Science, Boston College; David A. Selby EMAIL Data Science and its Applications, DFKI; Yao Xie EMAIL Georgia Institute of Technology; Sebastian Vollmer EMAIL Data Science and its Applications, DFKI; Gerrit Großmann EMAIL Data Science and its Applications, DFKI |
| Pseudocode | No | The paper describes various modeling approaches and methods, but it does not contain any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | No | A unified, open-source library with standardized data handling, reproducibility controls, modular architectural components, and configurable experimental protocols could facilitate more transparent, rigorous, and scalable research in neural STPPs. |
| Open Datasets | Yes | Synthetic datasets play an important role in neural STPP research by providing controlled environments for understanding model behavior under known conditions. Common examples include Pinwheel (Chen et al., 2021) for spatial heterogeneity and multimodal patterns, spatiotemporal Hawkes and self-correcting processes for self-exciting and self-correcting behaviors (Zhou et al., 2022), and Hawkes GMM (Yuan et al., 2023) for cluster-based triggering with long-range dependencies. ... Real-world datasets in neural STPP research capture diverse spatiotemporal phenomena that challenge current methods. Earthquake catalogs from Japan (U.S. Geological Survey, 2020) and California (NCEDC, 2014; Ross et al., 2019) exhibit long-range dependencies with continuous processes and varying background rates. Epidemic datasets using COVID-19 data from New Jersey (Chen et al., 2021), ACLED conflict records, and EMPRES-i disease (Okawa et al., 2022) outbreak reports reveal self-exciting behavior with seasonal effects and entangled spatiotemporal dynamics. Urban mobility datasets, including Citibike NYC (Chen et al., 2021) and taxi records, capture spatial heterogeneity through high-volume flows with clustering patterns. Public safety datasets involving Atlanta 911 calls (Zhu et al., 2022), Chicago crime records (Okawa et al., 2019), Vancouver crime data, and NYC collision reports (Zhang et al., 2023) showcase both self-exciting and self-correcting patterns with heterogeneous event types across different urban environments. |
| Dataset Splits | No | The literature points to several effective data-splitting strategies for testing generalization: temporal splits (e.g., roll-forward training on past events, testing on future periods), spatial splits (holding out entire regions), combined spatiotemporal splits (unseen regions and future periods), and cross-region splits (evaluating transferability across distinct geographic contexts). |
| Hardware Specification | No | This is a review paper that summarizes existing research. It does not describe any specific experiments conducted by the authors that would require hardware specifications. |
| Software Dependencies | No | This is a review paper that summarizes existing research. It does not describe any specific experiments conducted by the authors that would require software dependencies. |
| Experiment Setup | No | This is a review paper that summarizes existing research. It does not describe any specific experiments conducted by the authors that would require experimental setup details like hyperparameters or training configurations. |