Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

On-device Learning of EEGNet-based Network For Wearable Motor Imagery Brain-Computer Interface

Sizhen Bian, Pixi Kang, Julian Moosmann, Mengxi Liu, Pietro Bonazzi, Roman Rosipal, Michele Magno

TL;DR

The proposed approach, applied to the well-established EEGNet architecture, enables real-time and accurate adaptation to EEG signals from unregistered users and achieves enhanced real-time performance without compromising inference accuracy.

Abstract

Electroencephalogram (EEG)-based Brain-Computer Interfaces (BCIs) have garnered significant interest across various domains, including rehabilitation and robotics. Despite advancements in neural network-based EEG decoding, maintaining performance across diverse user populations remains challenging due to feature distribution drift. This paper presents an effective approach to address this challenge by implementing a lightweight and efficient on-device learning engine for wearable motor imagery recognition. The proposed approach, applied to the well-established EEGNet architecture, enables real-time and accurate adaptation to EEG signals from unregistered users. Leveraging the newly released low-power parallel RISC-V-based processor, GAP9 from Greeenwaves, and the Physionet EEG Motor Imagery dataset, we demonstrate a remarkable accuracy gain of up to 7.31\% with respect to the baseline with a memory footprint of 15.6 KByte. Furthermore, by optimizing the input stream, we achieve enhanced real-time performance without compromising inference accuracy. Our tailored approach exhibits inference time of 14.9 ms and 0.76 mJ per single inference and 20 us and 0.83 uJ per single update during online training. These findings highlight the feasibility of our method for edge EEG devices as well as other battery-powered wearable AI systems suffering from subject-dependant feature distribution drift.

On-device Learning of EEGNet-based Network For Wearable Motor Imagery Brain-Computer Interface

TL;DR

The proposed approach, applied to the well-established EEGNet architecture, enables real-time and accurate adaptation to EEG signals from unregistered users and achieves enhanced real-time performance without compromising inference accuracy.

Abstract

Electroencephalogram (EEG)-based Brain-Computer Interfaces (BCIs) have garnered significant interest across various domains, including rehabilitation and robotics. Despite advancements in neural network-based EEG decoding, maintaining performance across diverse user populations remains challenging due to feature distribution drift. This paper presents an effective approach to address this challenge by implementing a lightweight and efficient on-device learning engine for wearable motor imagery recognition. The proposed approach, applied to the well-established EEGNet architecture, enables real-time and accurate adaptation to EEG signals from unregistered users. Leveraging the newly released low-power parallel RISC-V-based processor, GAP9 from Greeenwaves, and the Physionet EEG Motor Imagery dataset, we demonstrate a remarkable accuracy gain of up to 7.31\% with respect to the baseline with a memory footprint of 15.6 KByte. Furthermore, by optimizing the input stream, we achieve enhanced real-time performance without compromising inference accuracy. Our tailored approach exhibits inference time of 14.9 ms and 0.76 mJ per single inference and 20 us and 0.83 uJ per single update during online training. These findings highlight the feasibility of our method for edge EEG devices as well as other battery-powered wearable AI systems suffering from subject-dependant feature distribution drift.
Paper Structure (12 sections, 5 figures, 3 tables, 1 algorithm)

This paper contains 12 sections, 5 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methods
  4. Dataset
  5. Target device
  6. Neural Network Model
  7. Online training engine
  8. Evaluation
  9. Feature distribution drift evaluation
  10. On-device training performance evaluation
  11. On-device computing efficiency evaluation
  12. Conclusion

Figures (5)

  • Figure 1: Left: EEG electrode distribution and EEG-based motor imagery human-computer interface, which often faces the challenge of feature distribution drift across users and thus degrade the classification performance of a pre-trained neural network model; Middle: wearable EEG device with an data processing unit at the extreme edge (image from g.tec GmbH); Right: the implemented dense layer (classifier) parameter updates through gradient descent-based backpropogation during online training, and targeted edge MCU: the GAP9 with RISC-V-based hardware accelerator
  • Figure 2: Varied EEGNet applied in this work, where Ch means channel number (64, 19, 8), WL means window length (3s, 2s, 1s)
  • Figure 3: Training Engine
  • Figure 4: Time and energy consumed per inference
  • Figure 5: Time and energy consumed per update