Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Rethinking Self-Training Based Cross-Subject Domain Adaptation for SSVEP Classification

Weiguang Wang, Yong Liu, Yingjie Gao, Guangyuan Xu

TL;DR

The paper tackles cross-subject variability in SSVEP-based BCIs by introducing a frequency-aware data alignment (FBEA) combined with a Cross-Subject Self-Training (CSST) framework that uses adversarial pre-training and a Dual-Ensemble Self-Training strategy, augmented by Time–Frequency Augmented Contrastive Learning (TFA-CL). The approach aligns distributions across subjects, refines pseudo-labels through temporal and multi-view ensembles, and enhances discriminability via contrastive learning across time and frequency augmentations. Empirical results on Benchmark and BETA demonstrate state-of-the-art accuracy and information transfer rate across multiple signal lengths, with ablation analyses confirming the effectiveness of each component. The work advances practical cross-subject SSVEP BCIs by delivering robust, unsupervised domain adaptation capabilities with strong generalization.

Abstract

Steady-state visually evoked potentials (SSVEP)-based brain-computer interfaces (BCIs) are widely used due to their high signal-to-noise ratio and user-friendliness. Accurate decoding of SSVEP signals is crucial for interpreting user intentions in BCI applications. However, signal variability across subjects and the costly user-specific annotation limit recognition performance. Therefore, we propose a novel cross-subject domain adaptation method built upon the self-training paradigm. Specifically, a Filter-Bank Euclidean Alignment (FBEA) strategy is designed to exploit frequency information from SSVEP filter banks. Then, we propose a Cross-Subject Self-Training (CSST) framework consisting of two stages: Pre-Training with Adversarial Learning (PTAL), which aligns the source and target distributions, and Dual-Ensemble Self-Training (DEST), which refines pseudo-label quality. Moreover, we introduce a Time-Frequency Augmented Contrastive Learning (TFA-CL) module to enhance feature discriminability across multiple augmented views. Extensive experiments on the Benchmark and BETA datasets demonstrate that our approach achieves state-of-the-art performance across varying signal lengths, highlighting its superiority.

Rethinking Self-Training Based Cross-Subject Domain Adaptation for SSVEP Classification

TL;DR

The paper tackles cross-subject variability in SSVEP-based BCIs by introducing a frequency-aware data alignment (FBEA) combined with a Cross-Subject Self-Training (CSST) framework that uses adversarial pre-training and a Dual-Ensemble Self-Training strategy, augmented by Time–Frequency Augmented Contrastive Learning (TFA-CL). The approach aligns distributions across subjects, refines pseudo-labels through temporal and multi-view ensembles, and enhances discriminability via contrastive learning across time and frequency augmentations. Empirical results on Benchmark and BETA demonstrate state-of-the-art accuracy and information transfer rate across multiple signal lengths, with ablation analyses confirming the effectiveness of each component. The work advances practical cross-subject SSVEP BCIs by delivering robust, unsupervised domain adaptation capabilities with strong generalization.

Abstract

Steady-state visually evoked potentials (SSVEP)-based brain-computer interfaces (BCIs) are widely used due to their high signal-to-noise ratio and user-friendliness. Accurate decoding of SSVEP signals is crucial for interpreting user intentions in BCI applications. However, signal variability across subjects and the costly user-specific annotation limit recognition performance. Therefore, we propose a novel cross-subject domain adaptation method built upon the self-training paradigm. Specifically, a Filter-Bank Euclidean Alignment (FBEA) strategy is designed to exploit frequency information from SSVEP filter banks. Then, we propose a Cross-Subject Self-Training (CSST) framework consisting of two stages: Pre-Training with Adversarial Learning (PTAL), which aligns the source and target distributions, and Dual-Ensemble Self-Training (DEST), which refines pseudo-label quality. Moreover, we introduce a Time-Frequency Augmented Contrastive Learning (TFA-CL) module to enhance feature discriminability across multiple augmented views. Extensive experiments on the Benchmark and BETA datasets demonstrate that our approach achieves state-of-the-art performance across varying signal lengths, highlighting its superiority.
Paper Structure (12 sections, 11 equations, 3 figures, 3 tables)

This paper contains 12 sections, 11 equations, 3 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Filter-Bank Euclidean Alignment
  4. Cross-Subject Self-Training
  5. Time–Frequency Augmented Contrastive Learning
  6. Experiments
  7. Dataset and Experimental Setup
  8. Results and Comparisons
  9. Ablation Study
  10. Conclusion
  11. Acknowledgement
  12. Compliance with Ethical Standards

Figures (3)

  • Figure 1: Overview of the proposed cross-subject domain adaptation method. (a) The strategy of Filter-Bank Euclidean Alignment. (b) The framework of Cross-Subject Self-Training consisting of a Pre-Training with Adversarial Learning stage (top) and a Dual-Ensemble Self-Training stage (bottom), where Time-Frequency Contrastive Learning is applied in the second stage.
  • Figure 2: Comparisons of classification accuracy (a) and ITR (b) among six different methods on the Benchmark dataset. Standard errors are indicated by the bars.
  • Figure 3: Comparisons of classification accuracy (a) and ITR (b) among six different methods on the BETA dataset. Standard errors are indicated by the bars.