Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Biswadeep Chakraborty, Hemant Kumawat, Beomseok Kang, Saibal Mukhopadhyay

TL;DR

SDGN tackles learning multivariate temporal point processes by dynamically estimating time-varying spatio-temporal graphs using RSNNs and STDP. The approach introduces membrane potential-based adaptive bases, STDP-driven temporal learning, and a principled dynamic graph estimation mechanism, enabling robust event modeling without predefined graphs. Theoretical results establish learning capacity, temporal resolution, and convergence, while empirical results on synthetic and real-world datasets demonstrate superior predictive accuracy and efficiency compared to strong baselines. This work offers a scalable, energy-efficient framework for modeling complex temporal interactions in domains such as transportation, social platforms, and emergency-response systems.

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.

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

TL;DR

SDGN tackles learning multivariate temporal point processes by dynamically estimating time-varying spatio-temporal graphs using RSNNs and STDP. The approach introduces membrane potential-based adaptive bases, STDP-driven temporal learning, and a principled dynamic graph estimation mechanism, enabling robust event modeling without predefined graphs. Theoretical results establish learning capacity, temporal resolution, and convergence, while empirical results on synthetic and real-world datasets demonstrate superior predictive accuracy and efficiency compared to strong baselines. This work offers a scalable, energy-efficient framework for modeling complex temporal interactions in domains such as transportation, social platforms, and emergency-response systems.

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.

Paper Structure

This paper contains 45 sections, 103 equations, 3 figures, 1 table, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Methods
  4. Problem Formulation
  5. Learning Algorithm
  6. Intensity Function Estimation
  7. Implementation
  8. Dynamic Neural Basis Construction
  9. Graph Structure Learning
  10. Learning Algorithm
  11. Optimization
  12. Theoretical Analysis
  13. Experiments and Empirical Results
  14. Datasets
  15. Baselines
  16. ...and 30 more sections

Figures (3)

  • Figure 1: Performance of (a) the graph estimation algorithm using Structural Similarity Index (SSI) between the underlying and estimated temporal graphs vs the Number of Nodes (b) RMSE for the event prediction task for different sparsity levels for the synthetic dataset (c), (d) Performance of Event Prediction with other baselines on the synthetic dataset for sparsity = 0.01, 0.3 respectively
  • Figure 2: Performance of the event prediction vs Number of Nodes for different sparsity levels for the synthetic dataset. The results are compared across models with random graphs and only spatial functional graphs
  • Figure 3: Visualization of the Synthetic Dataset with varying levels of sparsity and the corresponding average firing rate of the neurons

Theorems & Definitions (3)

  • proof
  • proof
  • proof