CBraMod represents a breakthrough foundation model framework for EEG signal decoding that significantly improves generalizability across diverse brain-related tasks.

• Overcomes limitations of traditional supervised learning approaches by leveraging a multi-dataset pre-training strategy
• Utilizes a novel criss-cross attention mechanism to better capture spatial-temporal EEG patterns
• Demonstrates superior performance on multiple downstream tasks including emotion recognition and seizure detection
• Achieves state-of-the-art results with fewer parameters than competing models

This research enables more accurate and reliable brain-computer interfaces for medical applications including neurological diagnosis, patient monitoring, and assistive technologies for disabled individuals.

CBraMod: A Criss-Cross Brain Foundation Model for EEG Decoding