Multivariate Spatio-Temporal Neural Hawkes Processes
New AI model analyzes terrorism data and other complex events by learning spatial and temporal dynamics simultaneously.
A team of researchers led by Christopher Chukwuemeka, Hojun You, and Mikyoung Jun has introduced a novel AI model called the Multivariate Spatio-Temporal Neural Hawkes Process, detailed in a paper submitted to IEEE Transactions on Knowledge and Data Engineering. This work addresses a critical gap in existing temporal Hawkes process models by demonstrating that likelihood-based performance metrics alone don't fully capture fitted intensity behavior. The proposed approach fundamentally extends continuous-time neural Hawkes processes by integrating spatial information directly into the latent state evolution, enabling the model to learn both temporal and spatial decay dynamics simultaneously without relying on predefined triggering kernels.
The technical innovation lies in the model's ability to capture complex spatio-temporal interactions across multiple event types, which existing temporal neural Hawkes processes fail to achieve. Simulation studies confirm the method successfully recovers sensible temporal and spatial intensity structures in multivariate spatio-temporal point patterns. In a practical application, the model demonstrated its capability by analyzing terrorism data from Pakistan, revealing intricate spatio-temporal dynamics across event types. This represents a significant advancement for fields requiring analysis of complex event sequences with spatial dimensions, including security analysis, epidemiology, and financial modeling, where understanding both when and where events occur and influence each other is crucial.
- Extends neural Hawkes processes by integrating spatial information into latent state evolution through learned decay dynamics
- Successfully recovers intensity structure where temporal-only approaches fail, proven in simulation studies
- Demonstrated practical application analyzing terrorism data from Pakistan to capture cross-event spatio-temporal interactions
Why It Matters
Enables better prediction and analysis of complex real-world events like crime, disease spread, and financial transactions by modeling both space and time.