HawkesNest: A Multi-Axis Synthetic Benchmark for Spatiotemporal Pattern Complexity

2026-06-15 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, quick

Summary

HawkesNest is a new generator-aligned synthetic benchmark designed to evaluate spatiotemporal point process (STPP) models by controlling pattern complexity. Built on a multivariate Hawkes backbone, HawkesNest addresses the challenge of opaque real-world datasets where latent generative structures are unknown, hindering model failure attribution. The benchmark defines four distinct complexity axes: space--time entanglement, background heterogeneity, cross-type interaction, and domain topology. Each axis is associated with a deterministic index derived from the latent data-generating mechanism. By varying these axes while maintaining fixed global rate, stability, and simulation budget, HawkesNest enables diagnostic stress tests. Verification shows the indices are monotone and nearly orthogonal. For instance, Hawkes-family baselines degrade under joint heterogeneity--entanglement complexity, and AutoSTPP exhibits vulnerability to isolated increases in space--time entanglement. The code is available at https://github.com/YahyaAalaila/HawkesNest.

Key takeaway

For Machine Learning Engineers evaluating spatiotemporal point process (STPP) models, HawkesNest offers a critical tool to move beyond opaque real-world data. You should integrate this benchmark into your model development pipeline to precisely diagnose performance degradations under known structural difficulties, such as space--time entanglement or background heterogeneity. This allows you to identify specific vulnerabilities in both traditional Hawkes-family models and neural STPP architectures like AutoSTPP, guiding targeted improvements and more robust model selection.

Key insights

HawkesNest provides a multi-axis synthetic benchmark for diagnostic stress testing of spatiotemporal point process models.

Principles

• Generator-aligned benchmarks enable clear failure attribution.
• Controlled complexity axes reveal model sensitivities.
• Orthogonal indices allow targeted diagnostic testing.

Method

HawkesNest generates synthetic spatiotemporal point process data using a multivariate Hawkes backbone, varying four complexity axes (space--time entanglement, background heterogeneity, cross-type interaction, domain topology) via deterministic indices, while fixing global rate, stability, and simulation budget.

In practice

• Stress test STPP models under known structural difficulty.
• Diagnose neural-model sensitivity to specific complexities.
• Evaluate Hawkes-family baselines against joint heterogeneity.

Topics

• Spatiotemporal Point Processes
• Synthetic Benchmarking
• Hawkes Processes
• Model Evaluation
• Neural Models
• AutoSTPP

Code references

• YahyaAalaila/HawkesNest

Best for: AI Scientist, Machine Learning Engineer, Research Scientist

Related on AIssential

• Benchmarking Counterfactual Prediction in Epidemic Time Series with Time-Varying Interventions — A new benchmark provides realistic, observable counterfactuals for epidemic time series under dynamic interventions, addressing a critical gap.
• Simulation-Augmented Multi-Step Split Conformal Prediction for Aggregated Forecasts — Simulation-augmented multi-step split conformal prediction effectively quantifies uncertainty for aggregated time-series forecasts.
• CoReDiT: Spatial Coherence-Guided Token Pruning and Reconstruction for Efficient Diffusion Transformers — CoReDiT efficiently prunes Diffusion Transformers by leveraging spatial coherence for token selection and reconstruction, significantly reducing computational cost.
• SREGym: A Live Benchmark for AI SRE Agents with High-Fidelity Failure Scenarios — SREGym provides a high-fidelity, live benchmark for AI SRE agents, revealing significant performance gaps in handling complex, real-world failures.
• CHANI: Correlation-based Hawkes Aggregation of Neurons with bio-Inspiration — CHANI is a bio-inspired spiking neural network mathematically proven to learn classification tasks via local rules.
• SRT: Super-Resolution for Time Series via Disentangled Rectified Flow — SRT reconstructs high-resolution time series from low-resolution inputs using disentangled rectified flow and component decomposition.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.