Learning Domain- and Class-Disentangled Prototypes for Domain-Generalized EEG Emotion Recognition
The framework learns emotion recognition from EEG data across different subjects without needing target-specific training data. It achieves this by separating features into domain-invariant and class-specific components, aggregating related subjects into superdomains using MMD, and aligning class prototypes adaptively. This allows robust classification on completely new subjects while handling label noise.