Self-Supervised Consistency Enhanced Disentangled Learning for Neural Decoding Generalization in Brain-Machine Interfaces
Jiyu Wei, Di Hong, Zhanjie Zhang, Dazhong Rong, Qinming He, Yueming Wang
AI summary
Problem
Neural drift degrades BMI decoding accuracy over time, and existing generalization methods overlook that drift varies across specific motor parameters like direction and speed, often treating velocity as a unified target.
Approach
The framework employs a teacher-student consistency constraint with simulated signal perturbations to learn robust features from single-day data, combined with a Complementary-Disentangled Generalization mechanism that separately decodes velocity, direction, and speed before ensemble-fusing them.
Key results
- State-of-the-art cross-day decoding performance on non-human primate datasets
- Superior robustness across short, medium, and long temporal intervals without recalibration
- Enhanced fine-grained accuracy for direction and speed predictions
- Effective single-session training framework mitigating heterogeneous neural drift
Why it matters
Enables long-term, stable, and fine-grained BMI control, advancing practical neurorehabilitation and human-centered robotics applications.
Abstract
Brain–Machine Interfaces (BMIs) provide a di- rect communication pathway between the brain and external devices, enabling humans to control assistive and robotic tech- nologies, with potential applications in rehabilitation, human motor augmentation, and human-centered robotics. However, due to neural drift, the performance of BMIs decreases over time, posing challenges for long-term viability, particularly for invasive BMIs (iBMIs). Existing solutions suffer from two main drawbacks: (i) difficulty in learning robust neural representa- tions, and (ii) neglecting that neural drift varies across motor parameters (e.g., velocity, direction, and speed). To overcome these limitations, we propose Self-Supervised Consistency en- hanced Disentangled Learning (SSCDL), a neural decoding generalization framework built on two key innovations. We first design a backbone model named Consistency enhanced Neural Decoder (CND), using a novel teacher-student consistency constraint with simulated neural signal perturbations to learn robust representations invariant to neural drift. Then, we em- ploy three dedicated CNDs under Complementary-Disentangled Generalization (CDG) mechanism, which disentangles motor signals into velocity, direction and speed with the inspiration of neural preference theory. This disentangled learning enables SSCDL to capture invariant neural representations from diverse neural preference perspectives, significantly enhancing cross- day generalization. Extensive experimental results show that SSCDL delivers state-of-the-art decoding performance, exhibit- ing high robustness and cross-day stability. These capabilities underscore its strong potential for long-term interaction in human-centric robotic and fine-grained assistive applications.