MTDN: Learning Multiple Temporal Dynamics Representation for Emotional Valence Classification with EEG.

Area of Science:

• Neuroscience
• Computer Science
• Artificial Intelligence

Background:

• Emotion recognition from electroencephalogram (EEG) is vital for understanding human affective states.
• Accurate computational models require capturing spatial, spectral, and temporal features of emotional responses.
• Existing methods struggle with efficiently learning complex temporal dynamics in EEG data.

Purpose of the Study:

• To develop an efficient deep learning framework, MTDN, for enhanced emotion recognition from EEG.
• To effectively capture spectral, spatial, and temporal features for improved affective computing.
• To address the challenge of learning temporal dynamics in EEG signals.

Main Methods:

• A novel deep learning framework, MTDN, was designed.
• MTDN incorporates a filterbank module for spectral feature extraction and a spatial convolution block for spatial feature learning.
• Parallel long short-term memory (LSTM) embedding and self-attention modules were employed to jointly learn multiple temporal dynamics by segment embedding and inter-correlation.

Main Results:

• The MTDN framework was evaluated on the publicly available DEAP dataset.
• MTDN demonstrated improved performance compared to existing state-of-the-art methods.
• Significant improvements were observed specifically on the valence dimension of the DEAP dataset.

Conclusions:

• The proposed MTDN framework offers an effective approach for emotion recognition from EEG.
• MTDN successfully captures crucial spectral, spatial, and temporal dynamics for affective computing.
• The framework shows promise for advancing the field of emotion recognition and human-computer interaction.