Related Experiment Video
Updated: Jan 8, 2026
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
5.2K
ELAI-SGCN: An explainable lightweight adaptive information-perceiving spiking graph convolutional network for
Jingxin Liu1, Zikai Song1, Xihang Qiu1
1Key Laboratory of Brain Health Intelligent Evaluation and Intervention, Ministry of Education, Beijing, 100081, China; School of Medical Technology, Beijing Institute of Technology, Beijing, 100081, China.
Summary
This study introduces ELAI-SGCN, a novel framework for efficient and interpretable electroencephalography (EEG) analysis. The model achieves high accuracy in emotion recognition while significantly reducing computational complexity for real-time applications.
Area of Science:
- Neuroscience
- Artificial Intelligence
- Biomedical Engineering
Background:
- Emotion recognition is crucial for diagnosing mental health conditions.
- Electroencephalography (EEG) offers high temporal resolution for brain activity analysis.
- Existing EEG models struggle with dynamic neural connectivity, spatial topology, and computational efficiency.
Purpose of the Study:
- To develop a lightweight and explainable framework for EEG-based emotion recognition.
- To address limitations in modeling dynamic neural connectivity and spatial topology in EEG data.
- To improve computational efficiency and performance of EEG analysis models.
Main Methods:
- Proposed ELAI-SGCN, a framework utilizing a trainable spiking encoder for sparse, event-driven EEG signal representation.
- Implemented a graph convolution module with spike-based operations for adaptive inter-regional connectivity modeling.
- Focused on preserving temporal dynamics and enabling interpretable, resource-efficient analysis.
Main Results:
- ELAI-SGCN achieved 87.08% valence and 89.96% arousal accuracy on the DEAP dataset with minimal parameters (60.48 K) and low computational cost (0.36 M FLOPs).
- On the SEED dataset, ELAI-SGCN reached 94.63% accuracy for three-class emotion recognition.
- Demonstrated significant reductions in parameters (99.8%) and computational cost (over 250-fold) compared to Dynamic Graph Convolutional Neural Networks.
Conclusions:
- ELAI-SGCN offers an efficient and interpretable method for modeling EEG spatiotemporal dynamics using a spiking-based dynamic graph convolutional framework.
- The model surpasses existing approaches in accuracy and computational efficiency, making it suitable for real-time EEG emotion recognition.
- Its lightweight design facilitates deployment on clinical devices, advancing intelligent psychological assessment systems.
