Related Experiment Video
Updated: Jul 8, 2026
09:44
Recording and Analyzing Multimodal Large-Scale Neuronal Ensemble Dynamics on CMOS-Integrated High-Density Microelectrode Array
Published on: March 8, 2024
HALO-GNN: hallucination-resistant temporal graph neural networks for dynamic community detection.
Yanfei Ma1, Daozheng Qu2,3, Yibo Wang4
1Department of Computer Science, Fairleigh Dickinson University, Vancouver, V6B 2P6, Canada. yanfei.ma@ieee.org.
Scientific Reports
|July 6, 2026
Summary
Temporal graph learning can suffer from illusory dynamics. This study introduces a hallucination-resistant method to stabilize temporal representations for reliable dynamic community detection.
Area of Science:
- Graph Neural Networks
- Dynamic Network Analysis
- Machine Learning
Background:
- Temporal graph learning is challenged by illusory structural dynamics, where noise creates deceptive community movements.
- This instability in community assignments limits the effectiveness of temporal graph neural networks in dynamic environments.
Purpose of the Study:
- To develop a robust learning paradigm resistant to hallucinations in temporal graph data.
- To stabilize temporal representations and improve dynamic community detection accuracy.
Main Methods:
- Introduced a memory-guided structural regularization framework.
- Stabilized node embeddings by referencing historical graph structures.
- Reduced oscillations from high-frequency noise while preserving low-frequency community evolution.
Main Results:
- Demonstrated significant improvements in robustness across temporal graph benchmarks.
- Showcased enhanced temporal consistency and perturbation resilience.
- Validated the effectiveness of hallucination resistance in dynamic community detection.
Conclusions:
- Hallucination resistance is crucial for reliable dynamic community detection.
- The proposed method effectively stabilizes temporal representations.
- The framework offers improved performance in dynamic and streaming graph scenarios.
