PhysioWave-T cuts biosignal token length with wavelet scale folding
Original title: PhysioWave-T: Token-Efficient Wavelet Foundation Models for Biosignals
A 30M EEG encoder leads 8 of 9 metrics versus BENDR, BIOT, LaBraM and EEGPT, and RALF lifts seizure sensitivity from 61.4% to 74.4%.
The authors present PhysioWave-T, a wavelet foundation-model family for EEG, ECG and sEMG. Dynamic scale folding recombines wavelet subbands before tokenization, shortening sequences from (J+1)CS to CS tokens, with route-specific front ends and adapters for unseen montages. They pre-train 30M EEG and ECG encoders and a 12M sEMG encoder; code and models are released.
Unconfirmed: Accuracy and efficiency figures are the authors' reported results.
- EEG pre-training used over 15 million windows with 19 to 128 channels.
- Scale folding cut latency 4.0–7.2x and peak memory 6.8–12.7x without accuracy loss.
- On SeizeIT2, RALF raised window AUPRC from 0.321 to 0.399 over the best unimodal baseline.
- The EEG model beat LUNA-L on wearable seizure detection with fewer parameters.
From the original article
1 min read at huggingface.co
We present PhysioWave-T, a token-efficient wavelet foundation model family for EEG, ECG, and sEMG. Dynamic scale folding recombines J+1 wavelet subbands prior to tokenization, compressing sequence length from (J+1)CS to CS tokens for J levels, C channels, and S temporal patches.
By Yanlong Chen, Simone Benatti, Luca Benini · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.