Proceedings of the Annual Meeting of the Cognitive Science Society

Traditional collaborative Brain Computer Interface (cBCI) pipelines treat participants as independent entities, inadvertently suppressing inter-brain neural dependencies critical for hyperscanning-based tasks. We propose an Inter-brain Coupling-Constrained Independent Component Analysis (ICC-ICA) method that incorporates a coupling gradient into the objective function to preserve shared neural signatures. Validated on a motor imagery (MI) dataset, our approach demonstrates that integrating inter-brain constraints does not compromise signal quality, yielding a stable 1.76% accuracy improvement. Crucially, functional network analysis reveals enhanced recovery of inter-brain links primarily localized in task-relevant motor regions. Furthermore, ICC-ICA shows superior retention of both intra-frequency and cross-frequency coupling across the entire task duration. These findings demonstrate that explicitly regularizing the ICA update rule with inter-participant constraints effectively safeguards social brain markers, providing a robust foundation for group-level neural decoding and more valid collaborative BCI systems.