Graph Convolutional Network With Connectivity Uncertainty for EEG-Based Emotion Recognition.

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Automatic emotion recognition using electroencephalography (EEG) is crucial for advancing human-computer interaction.
• Existing methods face challenges in learning long-range dependencies, handling topological ambiguities in EEG data, and mapping signal qualities to labels.

Purpose of the Study:

• To develop a robust Graph Convolutional Network (GCN) model for accurate emotion recognition from multichannel EEG signals.
• To address challenges in spatial dependency representation, temporal-spectral relativeness, and noisy label mitigation.

Main Methods:

• Introduced a distribution-based uncertainty method within a GCN architecture to capture spatial and temporal-spectral features.
• Employed graph mixup technique to enhance connectivity and reduce the impact of noisy labels.
• Integrated uncertainty learning with GCN weights, termed Connectivity Uncertainty GCN (CU-GCN).

Main Results:

• The proposed CU-GCN model demonstrated superior performance in emotion recognition tasks on SEED and SEEDIV datasets.
• Significant improvements were observed compared to existing methodologies.
• Ablation studies validated the effectiveness of individual components of the CU-GCN model.

Conclusions:

• The CU-GCN approach effectively represents spatial dependencies and temporal-spectral relativeness in EEG signals.
• The method successfully mitigates over-smoothing and noisy label issues, leading to enhanced emotion recognition accuracy.
• This work offers a promising advancement for emotion recognition in human-computer interaction.