Self-Supervised Consistency Enhanced Disentangled Learning for Neural Decoding Generalization in Brain-Machine Interface

The paper proposes Self-Supervised Consistency enhanced Disentangled Learning (SSCDL) for improving cross-day generalization in invasive brain-machine interfaces affected by neural drift. Its Consistency enhanced Neural Decoder (CND) uses a teacher-student consistency constraint with simulated neural-signal perturbations to learn drift-robust representations. A Complementary-Disentangled Generalization (CDG) mechanism then uses three CNDs to separate motor-related signals associated with velocity, direction, and speed. The abstract reports state-of-the-art decoding performance and improved cross-day stability, but provides no dataset names, participant counts, or numerical results.