Online Adaptation via Dual-Stage Alignment and Self-Supervision for Fast-Calibration Brain-Computer Interfaces

2025-09-23

AI Summary

Severe inter-subject variability in electroencephalographic (EEG) signals critically limits the online practicality of EEG-based brain–computer interfaces (BCIs). To address this, we propose an online adaptive calibration algorithm for unseen subjects, enabling real-time model optimization within a single online trial. Our method comprises two key components: (i) a dual-stage Euclidean-space alignment incorporating dynamic batch normalization statistic updates, and (ii) label-free self-supervised learning leveraging soft pseudo-labeling and Shannon entropy regularization. The framework is agnostic to BCI paradigms—including steady-state visual evoked potentials (SSVEP) and motor imagery—and compatible with mainstream decoder architectures. Extensive evaluation across five public datasets and seven decoders demonstrates that, using only one online trial, our approach improves average classification accuracy by 4.9% for SSVEP and 3.6% for motor imagery tasks. It significantly enhances cross-subject generalizability and deployment efficiency without requiring subject-specific labeled data.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Cognitive Modeling & Cognitive Systems: Adaptive Behavior
• Computer Vision: Learning & Optimization for CV

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• User Modeling, Personalization and Recommendation: User modeling for targeted and personalized online advertising
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments

Abstract

Individual differences in brain activity hinder the online application of electroencephalogram (EEG)-based brain computer interface (BCI) systems. To overcome this limitation, this study proposes an online adaptation algorithm for unseen subjects via dual-stage alignment and self-supervision. The alignment process begins by applying Euclidean alignment in the EEG data space and then updates batch normalization statistics in the representation space. Moreover, a self-supervised loss is designed to update the decoder. The loss is computed by soft pseudo-labels derived from the decoder as a proxy for the unknown ground truth, and is calibrated by Shannon entropy to facilitate self-supervised training. Experiments across five public datasets and seven decoders show the proposed algorithm can be integrated seamlessly regardless of BCI paradigm and decoder architecture. In each iteration, the decoder is updated with a single online trial, which yields average accuracy gains of 4.9% on steady-state visual evoked potentials (SSVEP) and 3.6% on motor imagery. These results support fast-calibration operation and show that the proposed algorithm has great potential for BCI applications.

Problem

• Overcoming individual brain activity differences in EEG-based BCI systems
• Enabling fast calibration for unseen subjects during online BCI operation
• Adapting decoders across different BCI paradigms and architectures

Innovation

• Dual-stage alignment in EEG and representation spaces
• Self-supervised loss using entropy-calibrated pseudo-labels
• Single-trial online adaptation for fast calibration