EEGReXferNet: A Lightweight Gen-AI Framework for EEG Subspace Reconstruction via Cross-Subject Transfer Learning and Channel-Aware Embedding

TL;DR

EEGReXferNet tackles the challenge of noisy EEG signals limiting real-time decoding by introducing a lightweight Gen-AI framework that reconstructs EEG subspaces through cross-subject transfer learning and channel-aware embedding. It combines neighborhood-based input selection, band-specific SubWindowConv1D encoding/decoding, and a sliding temporal-statistics pipeline with Sliced Wasserstein latent regularization and reference-based scaling to ensure continuity across windows. The approach yields high spatial-temporal-spectral fidelity (mean PSD correlation ≥ 0.95; mean spectrogram RV-Coefficient ≥ 0.85) while reducing model weights by about 45% and maintaining sub-millisecond inference times, enabling robust, real-time EEG preprocessing for BCIs. Across ablation studies and downstream MI decoding, dynamic latent spaces with SWD consistently outperform baselines, demonstrating strong cross-subject generalization and practical impact for neurophysiological applications.

Abstract

Electroencephalography (EEG) is a widely used non-invasive technique for monitoring brain activity, but low signal-to-noise ratios (SNR) due to various artifacts often compromise its utility. Conventional artifact removal methods require manual intervention or risk suppressing critical neural features during filtering/reconstruction. Recent advances in generative models, including Variational Autoencoders (VAEs) and Generative Adversarial Networks (GANs), have shown promise for EEG reconstruction; however, these approaches often lack integrated temporal-spectral-spatial sensitivity and are computationally intensive, limiting their suitability for real-time applications like brain-computer interfaces (BCIs). To overcome these challenges, we introduce EEGReXferNet, a lightweight Gen-AI framework for EEG subspace reconstruction via cross-subject transfer learning - developed using Keras TensorFlow (v2.15.1). EEGReXferNet employs a modular architecture that leverages volume conduction across neighboring channels, band-specific convolution encoding, and dynamic latent feature extraction through sliding windows. By integrating reference-based scaling, the framework ensures continuity across successive windows and generalizes effectively across subjects. This design improves spatial-temporal-spectral resolution (mean PSD correlation >= 0.95; mean spectrogram RV-Coefficient >= 0.85), reduces total weights by ~45% to mitigate overfitting, and maintains computational efficiency for robust, real-time EEG preprocessing in neurophysiological and BCI applications.

Figures (16)

• Figure 1: Overview of EEGReXferNet architecture illustrating key processing blocks and workflow.
• Figure 2: Subject ‘a’: sMAPE across EEG channels/windows for relative $\delta$, $\theta$, $\alpha$, and $\beta$ band power.
• Figure 3: Model comparison across subjects using EEG metrics. Heatmaps show Wilcoxon ranks for (top) clean and (bottom) noisy data. In (top), * marks the best (min), in (bottom), # marks the best (max). Cells show mean scores. Color scale: green (min) $\rightarrow$ cyan $\rightarrow$ yellow $\rightarrow$ gray (max).
• Figure 4: Comparison of accuracy metrics (Downstream classification using EEGNet-8-2) across subjects - Baseline vs. Reconstructed misclassified EEG windows via Model-C and D.
• Figure 5: Channel-wise mean and standard deviation of EEG signals for subject 'a' before and after band-pass filtering (BPF).