Beyond Hearing: Learning Task-agnostic ExG Representations from Earphones via Physiology-informed Tokenization

Electrophysiological (ExG) signals offer valuable insights into human physiology, but building foundation models that generalize across everyday tasks remains challenging due to limited data diversity and task-specific model design. Beyond Hearing introduces Physiology-informed Multi-band Tokenization (PiMT), which decomposes ExG signals into physiology-aware tokens and learns robust representations via a reconstruction objective. Evaluations on the DailySense dataset and multiple public ExG benchmarks demonstrate strong and consistent gains across diverse tasks.

The Fourteenth International Conference on Learning Representations (ICLR ‘26)