Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HFMCA: Orthonormal Feature Learning for EEG-based Brain Decoding

Yinghao Wang, Lintao Xu, Shujian Yu, Enzo Tartaglione, Van-Tam Nguyen

TL;DR

The paper tackles the difficulty of EEG decoding under noise, high dimensionality, and limited labeled data by proposing HierarchicalFMCA (HFMCA), a self-supervised framework that learns orthonormal EEG representations by maximizing dependence between low-level augmented views and a high-level fused representation while enforcing encoder orthogonality. It further extends to HFMCA++ with a contrastive regularization to prevent collapse, achieving robust, discriminative embeddings. Evaluated on SEED and BCIC-2A, HFMCA and HFMCA++ outperform competitive self-supervised baselines and exhibit strong cross-subject generalization, with LOSO gains of around 2.7% on SEED and 2.6% on BCIC-2A. This approach advances EEG foundation-model development by enabling effective, label-efficient learning and improved transfer across subjects and tasks.

Abstract

Electroencephalography (EEG) analysis is critical for brain-computer interfaces and neuroscience, but the intrinsic noise and high dimensionality of EEG signals hinder effective feature learning. We propose a self-supervised framework based on the Hierarchical Functional Maximal Correlation Algorithm (HFMCA), which learns orthonormal EEG representations by enforcing feature decorrelation and reducing redundancy. This design enables robust capture of essential brain dynamics for various EEG recognition tasks. We validate HFMCA on two benchmark datasets, SEED and BCIC-2A, where pretraining with HFMCA consistently outperforms competitive self-supervised baselines, achieving notable gains in classification accuracy. Across diverse EEG tasks, our method demonstrates superior cross-subject generalization under leave-one-subject-out validation, advancing state-of-the-art by 2.71\% on SEED emotion recognition and 2.57\% on BCIC-2A motor imagery classification.

HFMCA: Orthonormal Feature Learning for EEG-based Brain Decoding

TL;DR

The paper tackles the difficulty of EEG decoding under noise, high dimensionality, and limited labeled data by proposing HierarchicalFMCA (HFMCA), a self-supervised framework that learns orthonormal EEG representations by maximizing dependence between low-level augmented views and a high-level fused representation while enforcing encoder orthogonality. It further extends to HFMCA++ with a contrastive regularization to prevent collapse, achieving robust, discriminative embeddings. Evaluated on SEED and BCIC-2A, HFMCA and HFMCA++ outperform competitive self-supervised baselines and exhibit strong cross-subject generalization, with LOSO gains of around 2.7% on SEED and 2.6% on BCIC-2A. This approach advances EEG foundation-model development by enabling effective, label-efficient learning and improved transfer across subjects and tasks.

Abstract

Electroencephalography (EEG) analysis is critical for brain-computer interfaces and neuroscience, but the intrinsic noise and high dimensionality of EEG signals hinder effective feature learning. We propose a self-supervised framework based on the Hierarchical Functional Maximal Correlation Algorithm (HFMCA), which learns orthonormal EEG representations by enforcing feature decorrelation and reducing redundancy. This design enables robust capture of essential brain dynamics for various EEG recognition tasks. We validate HFMCA on two benchmark datasets, SEED and BCIC-2A, where pretraining with HFMCA consistently outperforms competitive self-supervised baselines, achieving notable gains in classification accuracy. Across diverse EEG tasks, our method demonstrates superior cross-subject generalization under leave-one-subject-out validation, advancing state-of-the-art by 2.71\% on SEED emotion recognition and 2.57\% on BCIC-2A motor imagery classification.
Paper Structure (7 sections, 9 equations, 1 figure, 2 tables)

This paper contains 7 sections, 9 equations, 1 figure, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Hierarchical FMCA for Self-Supervised Learning
  4. HFMCA++: Preventing Trivial Solutions via Contrastive Regularization
  5. Experiments
  6. Conclusion And Future Work
  7. Acknowledgement

Figures (1)

  • Figure 1: (a) Architecture diagram of the HFMCA framework. The raw EEG signal $\mathcal{X}$ is transformed by functions $T_i$ to generate augmented inputs $\mathcal{Y}$. The encoder $f_\theta$ extracts low-level features $\mathcal{S}$, which are concatenated and fed into $g_\omega$ to produce the high-level instance representation $\mathcal{Z}$. (b) Loss function structure: HFMCA++ employs a hybrid loss $\mathcal{L}_{\text{HFMCA++}}$ composed of the FMCA-based density ratio loss $\mathcal{L}_{\text{logdet}}$ and the cross-instance contrastive loss $\mathcal{L}_{\text{cont}}$.