Dynamic Graph Structure Estimation for Learning Multivariate Point Process using Spiking Neural Networks

2025-04-01

AI Summary

To address poor generalizability in multivariate point process modeling caused by restrictive static graph-structure assumptions, this paper proposes a spiking neural network (SNN)-based dynamic graph learning framework. It pioneers the integration of spike-timing-dependent plasticity (STDP) into event sequence modeling, enabling data-driven, online estimation of spatiotemporal functional graphs. The method eliminates reliance on predefined graph topologies by unifying dynamic graph neural networks with event-time encoding. Evaluated on real-world datasets—including NYC Taxi, 911 emergency calls, and Reddit activity—it achieves significant accuracy improvements over state-of-the-art methods while maintaining low computational overhead. Key contributions are: (i) establishing an STDP-driven paradigm for dynamic graph learning; (ii) enabling robust, assumption-light multivariate event modeling; and (iii) providing an interpretable, adaptive spatiotemporal dependency modeling tool applicable to neuroscience, epidemiology, and related domains.

Technology Category

• Cognitive Modeling & Cognitive Systems: Neural Spike Coding
• Machine Learning: Graph-based Machine Learning
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems

Application Category

• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Models for Web evolution

Abstract

Modeling and predicting temporal point processes (TPPs) is critical in domains such as neuroscience, epidemiology, finance, and social sciences. We introduce the Spiking Dynamic Graph Network (SDGN), a novel framework that leverages the temporal processing capabilities of spiking neural networks (SNNs) and spike-timing-dependent plasticity (STDP) to dynamically estimate underlying spatio-temporal functional graphs. Unlike existing methods that rely on predefined or static graph structures, SDGN adapts to any dataset by learning dynamic spatio-temporal dependencies directly from the event data, enhancing generalizability and robustness. While SDGN offers significant improvements over prior methods, we acknowledge its limitations in handling dense graphs and certain non-Gaussian dependencies, providing opportunities for future refinement. Our evaluations, conducted on both synthetic and real-world datasets including NYC Taxi, 911, Reddit, and Stack Overflow, demonstrate that SDGN achieves superior predictive accuracy while maintaining computational efficiency. Furthermore, we include ablation studies to highlight the contributions of its core components.

Problem

• Estimating dynamic spatio-temporal functional graphs from event data
• Improving predictive accuracy in temporal point process modeling
• Addressing limitations of static graph structures in learning dependencies

Innovation

• Uses spiking neural networks for dynamic graphs
• Learns spatio-temporal dependencies from events
• Combines SNNs with spike-timing-dependent plasticity