Multi-Source Co-adaptation for EEG-Based Emotion Recognition by Mining Correlation Information | JoVE Visualize

Area of Science:

Neuroscience and Artificial Intelligence
Biomedical Signal Processing
Machine Learning for Affective Computing

Background:

Subject-independent emotion recognition using electroencephalogram (EEG) is challenging due to high inter-subject variability in EEG patterns.
Existing cross-subject or cross-dataset classifiers often exhibit poor accuracy because they fail to generalize across different data distributions.
Domain adaptation techniques offer a promising approach for cross-distribution learning in affective computing.

Purpose of the Study:

To propose a robust multi-source co-adaptation framework (MACI) for automated emotion recognition using EEG features.
To address the limitations of subject-independent emotion classification by minimizing distribution differences between source and target domains.
To enhance the accuracy of EEG-based emotion recognition in cross-subject and cross-dataset scenarios.

Main Methods:

Developed a multi-source co-adaptation framework (MACI) incorporating mining diverse correlation information (MACI) among domains and features.
Employed L2,1-norm and correlation metric regularization to minimize statistical and semantic distribution discrepancies between domains.
Learned multiple subject-invariant classifiers jointly within a unified framework, leveraging knowledge from multiple sources via a developed correlation metric function.

Main Results:

The proposed MACI framework demonstrated superior performance in EEG-based emotion recognition compared to existing methods.
Experiments conducted on the DEAP and SEED datasets validated the effectiveness of MACI in improving classification accuracy.
The framework successfully learned subject-invariant classifiers by effectively adapting knowledge across different data domains.

Conclusions:

The MACI framework offers a robust and effective solution for subject-independent emotion recognition using EEG data.
This approach significantly improves the accuracy and generalizability of emotion classifiers across different subjects and datasets.
MACI represents a significant advancement in leveraging domain adaptation for enhanced affective computing with neurophysiological signals.