DARE-EEG: A Foundation Model for Dual-Aligned EEG Representation
Researchers propose DARE-EEG, a self-supervised foundation model that addresses the challenge of mask-invariant representation learning in EEG data. Existing pre-trained models fail to constrain representations from different masked views of the same signal to a consistent latent subspace, degrading transferability. DARE-EEG enforces mask-invariance through dual-aligned representation learning, introducing mask alignment via contrastive learning and anchor alignment to align masked representations to a moment. The model aims to improve generalizable neural representations across brain-computer interface applications.
Key facts
- DARE-EEG is a self-supervised foundation model for EEG data.
- It addresses mask-invariance in EEG encoders.
- Existing methods fail to constrain representations from different masked views.
- Mask alignment uses contrastive learning on multiple masked views.
- Anchor alignment aligns masked representations to a moment.
- The model is designed for brain-computer interface applications.
- Pre-training uses large-scale EEG data.
- The paper is on arXiv with ID 2605.18298.
Entities
Institutions
- arXiv