FusionGen: Feature Fusion-Based Few-Shot EEG Data Generation

TL;DR

FusionGen tackles data scarcity and inter-subject variability in EEG-based BCIs by learning disentangled representations and fusing cross-subject features to generate diverse, label-consistent EEG trials in few-shot settings. The method aligns input distributions with Euclidean alignment, performs feature matching fusion within a U-Net generator, and trains as a denoising autoencoder. Extensive experiments on MI and SSVEP MOABB datasets show that FusionGen outperforms traditional augmentation and several few-shot generative baselines, with robust gains in cross-subject and within-subject scenarios. The approach offers a scalable data-generation engine that can support robust BCI models and future foundation-model-style transfer learning with limited labeled data.

Abstract

Brain-computer interfaces (BCIs) provide potential for applications ranging from medical rehabilitation to cognitive state assessment by establishing direct communication pathways between the brain and external devices via electroencephalography (EEG). However, EEG-based BCIs are severely constrained by data scarcity and significant inter-subject variability, which hinder the generalization and applicability of EEG decoding models in practical settings. To address these challenges, we propose FusionGen, a novel EEG data generation framework based on disentangled representation learning and feature fusion. By integrating features across trials through a feature matching fusion module and combining them with a lightweight feature extraction and reconstruction pipeline, FusionGen ensures both data diversity and trainability under limited data constraints. Extensive experiments on multiple publicly available EEG datasets demonstrate that FusionGen significantly outperforms existing augmentation techniques, yielding notable improvements in classification accuracy.

Figures (9)

• Figure 1: Closed-loop brain-computer interface system.
• Figure 2: Scenarios of data distributions and generation strategies.
• Figure 3: Architecture of proposed FusionGen. Raw and auxiliary trials are first aligned, then encoded into multi-scale features; randomly sampled target features are matched and replaced with source features, propagated through the decoder via skip connections, and finally decoded to produce realistic generated EEG trials.
• Figure 4: Overview of the proposed EEG data generation pipeline.
• Figure 5: Visualization of generated EEG signals by various augmentation methods.