Learning from Bilateral PPG via Cross-Site Attention and Knowledge Distillation
Thamilezai Ananthakumar, Hung Manh Pham, Dong Ma, Archan Misra, Thivya Kandappu
ICML 2026 Workshop SD4H· 2026
PPG
Summary
Dual PaPaGei encoders (one per wrist), SQI-conditioned temporal gating and cross-wrist self-attention. Bilateral sports dataset: 6 activities, 30 subjects, ECG truth. Bilateral model 2.85 bpm MAE, r = 0.96.
Key facts
- Form Factor
- wrist
- License
- Not stated
Reported benchmarks
Heart rate estimation
bilateral-sports-30 · MAE (bpm) ↓
#1 of 1
2.79
Own 30 subj, 6 sports, ECG truth; non-dominant wrist; dominant 3.72; bilateral teacher 2.85
Updated: Oct 11, 2026
Related Research
- Heart rate estimation leaderboard
Benchmark results for heart rate estimation
- LIMA-Mamba: A lightweight and distribution-robust Mamba network for PPG-based biometrics
Research paper
- A Comparative Study of Signal Representations Methods and Deep Learning Architectures for PPG-Based Cuffless Blood Pressure Estimation
Seven 1D→2D transforms × several CNN variants on the UCI BP database. Best: CWT, errors reported as
- A Decodability-Grounded Audit of Clinical Prediction Claims in PPG Foundation Models
Audits frozen representations of four encoders on 37,784 MC-MED encounters via concept decodability,