Enhancing EEG Foundation Models via Dual-Branch Self-Distillation With Bi-Pretext Tasks

Abstract

We present a dual-branch self-supervised learning framework for EEG representation learning, combining masked reconstruction and clustering-based objectives. Evaluated across five diverse downstream tasks, our method achieves state-of-the-art performance under both linear probing and fine-tuning protocols. Ablation and visualization analyses confirm the robustness and transferability of the learned features. Our approach offers a promising foundation for future advances in general-purpose EEG analysis.