Model-agnostic meta-learning for EEG-based inter-subject emotion recognition.

Area of Science:

• Affective Computing
• Neuroscience
• Machine Learning

Background:

• Inter-subject emotion recognition from neural signals is challenging due to individual differences in brain activity.
• Existing algorithms struggle to achieve high accuracy in recognizing emotions across diverse subjects.

Purpose of the Study:

• To develop an efficient and generalizable method for inter-subject emotion recognition using electroencephalogram (EEG) data.
• To create a model-agnostic meta-learning algorithm for adaptable emotion decoders at a population level.

Main Methods:

• Proposed a model-agnostic meta-learning algorithm with pre-training and one-shot adaptation steps.
• The meta-decoder learns from diverse subjects and adapts to new subjects efficiently.
• Algorithm is compatible with various mainstream machine learning decoders.

Main Results:

• Evaluated on SEED, DEAP, and DREAMER EEG datasets.
• The adapted meta-emotion decoder achieved state-of-the-art inter-subject emotion recognition accuracy.
• Outperformed classical supervised learning baselines across different decoder architectures.

Conclusions:

• The proposed meta-learning algorithm significantly improves inter-subject generalizability for emotion recognition.
• Results show promise for enhancing future affective brain-computer interfaces.
• Offers a robust solution for heterogeneous neural signal characteristics in emotion detection.