Group Resonance Network: Learnable Prototypes and Multi-Subject Resonance for EEG Emotion Recognition

Researcher Renwei Meng has introduced the Group Resonance Network (GRN), a novel machine learning architecture designed to overcome a persistent challenge in brain-computer interfaces: accurately recognizing human emotions from EEG (electroencephalography) data across different people. Current methods struggle with 'inter-subject variability,' where brain signals for the same emotion can look wildly different from person to person. GRN tackles this by not just looking for subject-invariant features but by explicitly modeling the 'group regularities'—the shared neural responses to emotional stimuli that occur across multiple individuals.

The GRN architecture is built around three key components. First, an individual encoder processes band-specific features from a single subject's EEG. Second, a set of learnable group prototypes acts as a dynamic reference for common emotional states. Third, a multi-subject resonance branch calculates neural synchrony (using PLV/coherence metrics) between the subject and a small reference group. A fusion module then intelligently combines these individual and group-level representations for the final emotion classification. In rigorous testing on the standard SEED and DEAP datasets, GRN consistently outperformed existing baseline models, with ablation studies confirming that both the prototype learning and multi-subject resonance mechanisms provide complementary benefits to the system's accuracy.