MGMCL: Multi-Granularity Manifold Contrastive Learning With Neural ODEs for Cross-Subject EEG Emotion Recognition

2026-08-08

AI Summary

This study addresses the challenges of cross-subject EEG-based emotion recognition, which are primarily hindered by substantial inter-subject variability and inadequate modeling of the continuous nature of affective states. Existing approaches operating in Euclidean space often fail to preserve the intrinsic semantic structure of emotions. To overcome these limitations, this work proposes learning continuous emotion representations on the SPD Riemannian manifold. The method uniquely integrates multi-granularity manifold contrastive learning with neural ordinary differential equations (Neural ODEs) to capture emotion dynamics and employs the Gromov–Wasserstein distance for cross-subject manifold alignment. Furthermore, a weakly supervised framework is introduced to infer continuous valence–arousal–dominance dimensions from discrete labels while preserving their semantic ordinal relationships. Evaluated on SEED, SEED-IV, and DEAP datasets, the approach achieves accuracies of 91.23%, 73.82%, and 76.38%, respectively, outperforming state-of-the-art methods by 1.89%, 1.66%, and 1.28%.

Abstract

Cross-subject electroencephalogram (EEG)-based emotion recognition remains challenging due to substantial inter-individual variability and discrete formulation that overlooks affective continuity. Existing methods operate in Euclidean space and focus on marginal distribution alignment, failing to preserve the semantic structure of emotions across subjects. This article proposes MGMCL, reconceptualizing emotion recognition as learning continuous representations on symmetric positive definite (SPD) Riemannian manifolds. The frame?work introduces multi-granularity manifold contrastive learning at instance, emotion, and trajectory levels while preserving semantic ordering. Neural ordinary differential equations on manifolds model continuous emotion dynamics. Cross-subject generalization employs Gromov-Wasserstein manifold alignment. Weakly-supervised learning enables continuous valence-arousal-dominance prediction from discrete labels. Extensive experiments on three public datasets demonstrate state-of-the-art performance: 91.23% accuracy on SEED, 73.82% on SEED-IV, and 76.38% on DEAP, achieving consistent improvements of 1.89%, 1.66%, and 1.28% over previous best methods, respectively.

Problem

• cross-subject EEG
• emotion recognition
• inter-individual variability
• affective continuity
• semantic structure

Innovation

• Manifold Contrastive Learning
• Neural ODEs
• Riemannian Geometry
• Cross-Subject EEG
• Emotion Continuity