Enviar aquest missatge de text: Enhancing EEG Foundation Models via Dual-Branch Self-Distillation With Bi-Pretext Tasks