arXiv:2603.16281v1 Announce Type: new 
Abstract: Electroencephalography (EEG) is a widely used tool for studying brain function, with applications in clinical neuroscience, diagnosis, and brain-computer interfaces (BCIs). Recent EEG foundation models trained on large unlabeled corpora aim to learn transferable representations, but their effectiveness remains unclear; reported improvements over smaller task-specific models are often modest, sensitive to downstream adaptation and fine-tuning strategies, and limited under linear probing. We hypothesize that one contributing factor is the reliance on signal reconstruction as the primary self-supervised learning (SSL) objective, which biases representations toward high-variance artifacts rather than task-relevant neural structure. To address this limitation, we explore an SSL paradigm based on Joint Embedding Predictive Architectures (JEPA), which learn by predicting latent representations instead of reconstructing raw signals. While earlier JEPA-style methods often rely on additional heuristics to ensure training stability, recent advances such as LeJEPA provide a more principled and stable formulation. We introduce Laya, the first EEG foundation model based on LeJEPA. Across a range of EEG benchmarks, Laya demonstrates improved performance under linear probing compared to reconstruction-based baselines, suggesting that latent predictive objectives offer a promising direction for learning transferable, high-level EEG representations.

تم تقديم نهج جديد لتخطيط الدماغ الكهربائي (EEG) من خلال إطار عمل LeJEPA، الذي يستخدم هياكل التنبؤ بالتضمين المشترك (JEPA) لتعزيز التعلم الذاتي الإشرافي من خلال التنبؤ بالتمثيلات الكامنة بدلاً من الاعتماد على إعادة بناء الإشارات. يهدف هذا التغيير إلى تحسين فعالية نماذج EEG في علم الأعصاب السريري وواجهات الدماغ-الكمبيوتر.

Se ha introducido un nuevo enfoque para la electroencefalografía (EEG) a través del marco LeJEPA, que utiliza arquitecturas predictivas de incrustación conjunta (JEPA) para mejorar el aprendizaje auto-supervisado al predecir representaciones latentes en lugar de depender de la reconstrucción de señales. Este cambio busca mejorar la efectividad de los modelos de EEG en neurociencia clínica e interfaces cerebro-computadora.

Une nouvelle approche de l'électroencéphalographie (EEG) a été introduite grâce au cadre LeJEPA, qui utilise les architectures prédictives d'encodage conjoint (JEPA) pour améliorer l'apprentissage auto-supervisé en prédisant des représentations latentes plutôt qu'en s'appuyant sur la reconstruction du signal. Ce changement vise à améliorer l'efficacité des modèles EEG dans les neurosciences cliniques et les interfaces cerveau-ordinateur.

A new approach to electroencephalography (EEG) has been introduced through the LeJEPA framework, which utilizes Joint Embedding Predictive Architectures (JEPA) to enhance self-supervised learning by predicting latent representations rather than relying on signal reconstruction. This shift aims to improve the effectiveness of EEG models in clinical neuroscience and brain-computer interfaces.