Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Calibration-free online test-time adaptation for electroencephalography motor imagery decoding

Martin Wimpff, Mario Döbler, Bin Yang

TL;DR

This work tackles distribution shifts in EEG motor imagery decoding by introducing a source-free online test-time adaptation framework. It combines covariance alignment (EA/RA), online adaptive batch normalization, and entropy minimization, guided by an input buffer, to calibrate models during inference without accessing source data. Across cross-session, cross-subject, and continual settings on BCIC IV 2a/2b, the approach achieves state-of-the-art gains and demonstrates substantial privacy preservation, with improvements up to roughly 12% over non-adaptive baselines. The results suggest a practical shift toward online adaptation for robust, real-world BCI deployment.

Abstract

Providing a promising pathway to link the human brain with external devices, Brain-Computer Interfaces (BCIs) have seen notable advancements in decoding capabilities, primarily driven by increasingly sophisticated techniques, especially deep learning. However, achieving high accuracy in real-world scenarios remains a challenge due to the distribution shift between sessions and subjects. In this paper we will explore the concept of online test-time adaptation (OTTA) to continuously adapt the model in an unsupervised fashion during inference time. Our approach guarantees the preservation of privacy by eliminating the requirement to access the source data during the adaptation process. Additionally, OTTA achieves calibration-free operation by not requiring any session- or subject-specific data. We will investigate the task of electroencephalography (EEG) motor imagery decoding using a lightweight architecture together with different OTTA techniques like alignment, adaptive batch normalization, and entropy minimization. We examine two datasets and three distinct data settings for a comprehensive analysis. Our adaptation methods produce state-of-the-art results, potentially instigating a shift in transfer learning for BCI decoding towards online adaptation.

Calibration-free online test-time adaptation for electroencephalography motor imagery decoding

TL;DR

This work tackles distribution shifts in EEG motor imagery decoding by introducing a source-free online test-time adaptation framework. It combines covariance alignment (EA/RA), online adaptive batch normalization, and entropy minimization, guided by an input buffer, to calibrate models during inference without accessing source data. Across cross-session, cross-subject, and continual settings on BCIC IV 2a/2b, the approach achieves state-of-the-art gains and demonstrates substantial privacy preservation, with improvements up to roughly 12% over non-adaptive baselines. The results suggest a practical shift toward online adaptation for robust, real-world BCI deployment.

Abstract

Providing a promising pathway to link the human brain with external devices, Brain-Computer Interfaces (BCIs) have seen notable advancements in decoding capabilities, primarily driven by increasingly sophisticated techniques, especially deep learning. However, achieving high accuracy in real-world scenarios remains a challenge due to the distribution shift between sessions and subjects. In this paper we will explore the concept of online test-time adaptation (OTTA) to continuously adapt the model in an unsupervised fashion during inference time. Our approach guarantees the preservation of privacy by eliminating the requirement to access the source data during the adaptation process. Additionally, OTTA achieves calibration-free operation by not requiring any session- or subject-specific data. We will investigate the task of electroencephalography (EEG) motor imagery decoding using a lightweight architecture together with different OTTA techniques like alignment, adaptive batch normalization, and entropy minimization. We examine two datasets and three distinct data settings for a comprehensive analysis. Our adaptation methods produce state-of-the-art results, potentially instigating a shift in transfer learning for BCI decoding towards online adaptation.
Paper Structure (18 sections, 4 equations, 4 figures, 2 tables)

This paper contains 18 sections, 4 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Task and Datasets
  4. Settings
  5. Cross-session setting
  6. Cross-subject setting
  7. Cross-subject continual setting
  8. Model and Source training
  9. Input buffer
  10. Alignment
  11. Batch Normalization
  12. Entropy Minimization
  13. Results
  14. Cross-session
  15. Cross-subject
  16. ...and 3 more sections

Figures (4)

  • Figure 1: Methodology for the (a) cross-session and (b) cross-subject setting.
  • Figure 2: Test accuracy and buffer size.
  • Figure 3: Label smoothing and entropy minimization.
  • Figure 4: Results per subject for the 2a dataset.