Uni-NTFM: A Unified Foundation Model for EEG Signal Representation Learning

Researchers developed Uni-NTFM, a new foundation model for EEG signal analysis that incorporates biological neural mechanisms and achieved record-breaking 1.9 billion parameters. The model was pre-trained on 28,000 hours of EEG data and outperformed existing models across nine downstream tasks by aligning architecture with actual brain functionality.

• →Uni-NTFM introduces a novel approach to EEG analysis by incorporating three core neuroscience principles instead of adapting computer vision or NLP architectures.
• →The model features a Heterogeneous Feature Projection Module that encodes both time-domain and frequency-domain signals for comprehensive brain activity analysis.
• →A Topological Embedding mechanism aligns different sensor configurations onto unified latent functional topography to reconstruct brain region geometry.
• →The Mixture-of-Experts Transformer network achieves 1.9 billion parameters while preventing task interference through dynamic routing mechanisms.
• →Pre-trained on 28,000 hours of EEG data, the model outperformed existing approaches across nine distinct downstream tasks in both linear probing and fine-tuning scenarios.