The RG-Flow Transformer: Encoding Scale-Free Dynamics in Scarce EEG

arXiv cs.AI · 6 hours ago

Researchers developed the RG-Flow Transformer, a transformer model incorporating renormalization-group principles to process EEG data by explicitly modeling the scale-free dynamics of brain signals. Testing on the PhysioNet Sleep-EDF corpus with 5 subjects showed RG-Flow achieved 77.3% accuracy on 5-class sleep staging compared to 77.0% for a vanilla transformer, with no statistically significant difference (p=0.294). The key distinction is that RG-Flow can recover the spectral exponent of EEG signals with an R² of 0.416, providing interpretability about brain state that standard transformers cannot capture.