Why biomedical signals fail across patients—and how to fix it

Biomedical signals like EEGs and ECGs have consistent patterns within a condition but shift in frequency and amplitude between patients, breaking models trained on one group. BioFormer treats this as a spectral alignment problem: it detects frequency-component shifts and adjusts them to match across subjects, like tuning a radio's interference. Combined with signal-aware normalization, the method improves cross-subject accuracy by 6% absolute F1 over existing approaches across six datasets.