Uncertainty-Guided Self-Paced Learning Framework for Long-Term Brain-Machine Interfaces
A new framework called Uncertainty-guided Self-paced Cycling (UnSPC) is proposed to address neural drift in invasive Brain-Machine Interfaces (BMIs). Neural drift degrades decoding performance over time, requiring frequent recalibration. Existing methods using only domain adaptation (DA) or domain generalization (DG) fail to capture fine-grained distribution shifts across neural subdomains. UnSPC synergizes DA and DG for target domain refining, employing an Uncertainty-guided Self-paced Pseudo-labeling (UnSPL) mechanism to iteratively mine reliable pseudo-labels and handle subdomain neural drift. The framework aims to improve long-term BMI deployment for rehabilitation, human performance augmentation, and human-centered robotics.
Key facts
- UnSPC framework synergizes domain adaptation and domain generalization
- Neural drift degrades BMI decoding performance over time
- Existing methods fail to capture fine-grained distribution shifts
- UnSPL mechanism iteratively mines reliable pseudo-labels
- Framework targets long-term invasive BMI deployment
- BMIs hold potential in rehabilitation and human performance augmentation
- Frequent recalibration is currently needed due to neural drift
- UnSPC refines target domain under uncertainty guidance
Entities
—