Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EEG2Rep: Enhancing Self-supervised EEG Representation Through Informative Masked Inputs

Navid Mohammadi Foumani, Geoffrey Mackellar, Soheila Ghane, Saad Irtza, Nam Nguyen, Mahsa Salehi

TL;DR

EEG2Rep tackles the core challenges of self-supervised EEG representation learning—low SNR, wide amplitude variability, and lack of explicit segmentation—by predicting latent-space representations rather than raw signals and by employing a semantic subsequence preserving masking strategy. The method uses a Transformer-based Target/Context/Predictor architecture with EMA updates and a VICReg-based regularization to avoid collapse, achieving superior performance across six EEG tasks in both linear probing and fine-tuning settings. Across datasets with substantial subject variability, EEG2Rep demonstrates strong generalization, cross-domain transfer, and robustness to common EEG noise, suggesting significant practical utility for EEG analysis. The work provides a principled approach to leverage informative masking and latent representations to improve semantic quality in EEG SSL, with potential alignment to brain time scales in future work.

Abstract

Self-supervised approaches for electroencephalography (EEG) representation learning face three specific challenges inherent to EEG data: (1) The low signal-to-noise ratio which challenges the quality of the representation learned, (2) The wide range of amplitudes from very small to relatively large due to factors such as the inter-subject variability, risks the models to be dominated by higher amplitude ranges, and (3) The absence of explicit segmentation in the continuous-valued sequences which can result in less informative representations. To address these challenges, we introduce \textit{EEG2Rep}, a self-prediction approach for self-supervised representation learning from EEG. Two core novel components of EEG2Rep are as follows: 1) Instead of learning to predict the masked input from raw EEG, EEG2Rep learns to predict masked input in latent representation space, and 2) Instead of conventional masking methods, EEG2Rep uses a new semantic subsequence preserving (SSP) method which provides informative masked inputs to guide EEG2Rep to generate rich semantic representations. In experiments on 6 diverse EEG tasks with subject variability, EEG2Rep significantly outperforms state-of-the-art methods. We show that our semantic subsequence preserving improves the existing masking methods in self-prediction literature and find that preserving 50\% of EEG recordings will result in the most accurate results on all 6 tasks on average. Finally, we show that EEG2Rep is robust to noise addressing a significant challenge that exists in EEG data. Models and code are available at:\url{https://github.com/Navidfoumani/EEG2Rep}

EEG2Rep: Enhancing Self-supervised EEG Representation Through Informative Masked Inputs

TL;DR

EEG2Rep tackles the core challenges of self-supervised EEG representation learning—low SNR, wide amplitude variability, and lack of explicit segmentation—by predicting latent-space representations rather than raw signals and by employing a semantic subsequence preserving masking strategy. The method uses a Transformer-based Target/Context/Predictor architecture with EMA updates and a VICReg-based regularization to avoid collapse, achieving superior performance across six EEG tasks in both linear probing and fine-tuning settings. Across datasets with substantial subject variability, EEG2Rep demonstrates strong generalization, cross-domain transfer, and robustness to common EEG noise, suggesting significant practical utility for EEG analysis. The work provides a principled approach to leverage informative masking and latent representations to improve semantic quality in EEG SSL, with potential alignment to brain time scales in future work.

Abstract

Self-supervised approaches for electroencephalography (EEG) representation learning face three specific challenges inherent to EEG data: (1) The low signal-to-noise ratio which challenges the quality of the representation learned, (2) The wide range of amplitudes from very small to relatively large due to factors such as the inter-subject variability, risks the models to be dominated by higher amplitude ranges, and (3) The absence of explicit segmentation in the continuous-valued sequences which can result in less informative representations. To address these challenges, we introduce \textit{EEG2Rep}, a self-prediction approach for self-supervised representation learning from EEG. Two core novel components of EEG2Rep are as follows: 1) Instead of learning to predict the masked input from raw EEG, EEG2Rep learns to predict masked input in latent representation space, and 2) Instead of conventional masking methods, EEG2Rep uses a new semantic subsequence preserving (SSP) method which provides informative masked inputs to guide EEG2Rep to generate rich semantic representations. In experiments on 6 diverse EEG tasks with subject variability, EEG2Rep significantly outperforms state-of-the-art methods. We show that our semantic subsequence preserving improves the existing masking methods in self-prediction literature and find that preserving 50\% of EEG recordings will result in the most accurate results on all 6 tasks on average. Finally, we show that EEG2Rep is robust to noise addressing a significant challenge that exists in EEG data. Models and code are available at:\url{https://github.com/Navidfoumani/EEG2Rep}
Paper Structure (25 sections, 8 equations, 5 figures, 7 tables)

This paper contains 25 sections, 8 equations, 5 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Methodology of EEG2Rep
  3. Problem Definition
  4. Model architecture
  5. Semantic Subsequence Preserving (SSP)
  6. Efficiency: Multiple masking
  7. Loss Regularisation
  8. Related Work
  9. Invariance-based self-supervised learning
  10. Self-Prediction based self-supervised learning
  11. Experimental Results
  12. Datasets
  13. Competitors and Implementation Setup
  14. Linear Probing
  15. Fine Tuning
  16. ...and 10 more sections

Figures (5)

  • Figure 1: Comparison of 2D t-SNE plots for representation learned by (a) MAEEG and (b) EEG2Rep on the Crowdsourced EEG dataset.
  • Figure 2: Architecture of EEG2Rep
  • Figure 3: Different masking strategies applied to EEG samples from Crowdsourced datasets crowdsourced with a 50% masking ratio. From top to bottom: (1) Original EEG sample, (2) Random masking, (3) Block masking, and (4) Semantic Subsequence Preserving (SSP).
  • Figure 4: Effect of preserving percentage ($1-\rho$) on average accuracy across all EEG datasets: A comparison of accuracy variation across different numbers of blocks ($\beta$), with error bars indicating standard deviation.
  • Figure 5: Model Robustness Comparison: Assessing model performance by introducing Gaussian noise, DC-shift, and amplitude changes to Crowdsourced data.