EEG Emotional State Labeling and Model Training
TextEmotion Recognition
Labeled EEG-based emotion recognition data by designing and benchmarking SOGNN architectures on the SEED-IV dataset. Integrated focal loss, NT-Xent contrastive regularization, stratified sampling, and multimodal deep learning models for label accuracy. Built a complete EEG pipeline involving preprocessing, tensor generation, and balanced batch formation for improved label fairness.
• Generated labels for emotion classes in EEG data with high accuracy and macro-F1 scores.
• Used stratified batching and augmentations to improve balance quality for training data.
• Achieved a balanced accuracy-fairness trade-off through ablation study methodology.
• Supported deployment in BCI, mental health monitoring, and adaptive learning platforms.
Labeled EEG-based emotion recognition data by designing and benchmarking SOGNN architectures on the SEED-IV dataset. Integrated focal loss, NT-Xent contrastive regularization, stratified sampling, and multimodal deep learning models for label accuracy. Built a complete EEG pipeline involving preprocessing, tensor generation, and balanced batch formation for improved label fairness. • Generated labels for emotion classes in EEG data with high accuracy and macro-F1 scores. • Used stratified batching and augmentations to improve balance quality for training data. • Achieved a balanced accuracy-fairness trade-off through ablation study methodology. • Supported deployment in BCI, mental health monitoring, and adaptive learning platforms.
2026 - Present