An Ultra-Low Power Wearable BMI System with Continual Learning Capabilities
Abstract
This paper presents a wearable brain–machine interface that couples a compact convolutional neural network with continual learning to cope with inter-session variability. Across two in-house datasets, the on-device continual-learning workflow improves accuracy by 30.36% and 10.17% compared to static models. Implemented on the BioGAP platform with a GAP9 microcontroller, inference consumes 0.45 mJ per pass while adaptation steps run in 21.5 ms, enabling approximately 25 hours of operation from a 100 mAh battery.
Type
Key Highlights
- On-device continual-learning workflow boosts BMI accuracy by up to 30.36% versus static models.
- Runs on the BioGAP platform with GAP9, consuming 0.45 mJ per inference and 21.5 ms per adaptation step.
- Supports approximately 25 hours of operation from a 100 mAh battery, enabling adaptive wearable BMIs.
Thorir Mar Ingolfsson
Postdoctoral Researcher
Making AI run on microwatts: I build foundation models for EEG, ECG, and other biosignals, and compress them until they run in real time on wearable devices, not in the cloud.