T-TIME: Test-Time Information Maximization Ensemble for Plug-and-Play BCIs

TL;DR

T-TIME tackles the challenge of calibration-free, online MI-BCI by enabling real-time test-time adaptation. It builds an ensemble of $M$ source-models trained on Euclidean-aligned data, predicts labels for each incoming trial using a spectral meta-learner, and updates each target model with conditional entropy minimization and adaptive marginal distribution regularization on a sliding batch of size $B$. The approach leverages incremental Euclidean alignment, online ensemble fusion, and principled loss terms to handle class-imbalance and non-stationarity, achieving superior online performance across three public MI-BCI datasets—surpassing about 20 TL methods. This work demonstrates the feasibility of plug-and-play, calibration-free BCIs and provides practical insights into online TL for EEG data, including continual adaptation and privacy-preserving deployment avenues.

Abstract

Objective: An electroencephalogram (EEG)-based brain-computer interface (BCI) enables direct communication between the human brain and a computer. Due to individual differences and non-stationarity of EEG signals, such BCIs usually require a subject-specific calibration session before each use, which is time-consuming and user-unfriendly. Transfer learning (TL) has been proposed to shorten or eliminate this calibration, but existing TL approaches mainly consider offline settings, where all unlabeled EEG trials from the new user are available. Methods: This paper proposes Test-Time Information Maximization Ensemble (T-TIME) to accommodate the most challenging online TL scenario, where unlabeled EEG data from the new user arrive in a stream, and immediate classification is performed. T-TIME initializes multiple classifiers from the aligned source data. When an unlabeled test EEG trial arrives, T-TIME first predicts its labels using ensemble learning, and then updates each classifier by conditional entropy minimization and adaptive marginal distribution regularization. Our code is publicized. Results: Extensive experiments on three public motor imagery based BCI datasets demonstrated that T-TIME outperformed about 20 classical and state-of-the-art TL approaches. Significance: To our knowledge, this is the first work on test time adaptation for calibration-free EEG-based BCIs, making plug-and-play BCIs possible.

Figures (6)

• Figure 1: Three different TL settings, when the target domain is completely unlabeled. (a) Unsupervised domain adaptation (UDA); (b) source-free unsupervised domain adaptation (SFUDA); and, (c) test-time adaptation (TTA).
• Figure 2: Flowchart of the proposed T-TIME.
• Figure 3: Conditional entropy minimization and adaptive marginal distribution regularization in target model update.
• Figure 4: Performance of different ensemble strategies on BNCI2014001 as the number of EEGNet base models varies. (a) binary classification (left/right hand); and, (b) 4-class classification.
• Figure 5: Performance of T-TIME w.r.t. different hyperparameter values. (a) Temperature scaling factor $T$; and, (b) pseudo-labeling confidence threshold $\tau$.