[Paper Review] Backpropagation-Free Test-Time Adaptation for Lightweight EEG-Based Brain-Computer Interfaces

Siyang Li, Jiayi Ouyang+5 more|arXiv (Cornell University)|Jan 12, 2026

EEG and Brain-Computer Interfaces0 citations

TL;DR

The paper introduces Backpropagation-Free Transformations (BFT), a test-time adaptation method for EEG-based BCIs that updates predictions without backpropagation, using knowledge-guided augmentations and an online ranking module to aggregate multiple transformed predictions.

ABSTRACT

Electroencephalogram (EEG)-based brain-computer interfaces (BCIs) face significant deployment challenges due to inter-subject variability, signal non-stationarity, and computational constraints. While test-time adaptation (TTA) mitigates distribution shifts under online data streams without per-use calibration sessions, existing TTA approaches heavily rely on explicitly defined loss objectives that require backpropagation for updating model parameters, which incurs computational overhead, privacy risks, and sensitivity to noisy data streams. This paper proposes Backpropagation-Free Transformations (BFT), a TTA approach for EEG decoding that eliminates such issues. BFT applies multiple sample-wise transformations of knowledge-guided augmentations or approximate Bayesian inference to each test trial, generating multiple prediction scores for a single test sample. A learning-to-rank module enhances the weighting of these predictions, enabling robust aggregation for uncertainty suppression during inference under theoretical justifications. Extensive experiments on five EEG datasets of motor imagery classification and driver drowsiness regression tasks demonstrate the effectiveness, versatility, robustness, and efficiency of BFT. This research enables lightweight plug-and-play BCIs on resource-constrained devices, broadening the real-world deployment of decoding algorithms for EEG-based BCI.

Motivation & Objective

• Address inter-subject variability and non-stationarity in EEG-based BCIs.
• Develop a backpropagation-free test-time adaptation method suitable for real-time, resource-constrained devices.
• Provide theoretical justification for aggregation of test-time transformations.
• Demonstrate versatility across classification and regression EEG decoding tasks on multiple datasets.

Proposed method

• Propose Backpropagation-Free Transformations (BFT) comprising two transformation families: BFT-A (knowledge-guided augmentations) and BFT-D (approximate Bayesian inference via Monte Carlo dropout).
• Generate multiple transformed versions of each test input and obtain multiple predictions per trial through a single forward pass, avoiding gradient updates.
• Introduce a learning-to-rank module to assign reliability-based weights to each transformed prediction, facilitated by a mapping module that produces rank-like scores.
• Aggregate predictions using reliability-weighted logits for classification or top-half averaging for regression, with temperature sharpening applied to logits prior to averaging.
• Provide theoretical variance-based justification showing uncertainty reduction from aggregating transformed predictions under BFT.

Experimental results

Research questions

• RQ1 Can backpropagation-free transformations reduce prediction uncertainty and improve transfer under domain shift in EEG decoding?
• RQ2 How can a ranking-based reliability scoring mechanism improve aggregation of multiple test-time predictions?
• RQ3 Do BFT methods generalize across both classification (MI) and regression (driver drowsiness) EEG tasks?
• RQ4 Can BFT be integrated with existing marginal-shift techniques (e.g., EA, BN-adapt) without conflict?
• RQ5 What is the practical impact of BFT on real-time, privacy-preserving BCI deployments on resource-constrained devices?

Key findings

• BFT provides a lightweight, privacy-preserving, noise-robust, and task-agnostic approach to test-time adaptation for EEG-based BCIs.
• Theoretical analysis shows aggregation of transformed predictions reduces uncertainty and improves transfer under domain shift.
• Empirical validation across five EEG datasets (MI classification and driver drowsiness regression) demonstrates the effectiveness and efficiency of BFT in online inference.
• BFT enables plug-and-play EEG decoding on devices with limited computational resources, reducing the need for per-use calibration.
• The ranking-based weighting mechanism helps prioritize more reliable transformations, enhancing robustness across tasks.