DeepHeart: Semi-Supervised Sequence Learning for Cardiovascular Risk Prediction

First-author paper presented at AAAI 2018 (the arXiv Comments field states 'Presented at AAAI 2018'): a semi-supervised multi-task LSTM trained on 57,675 person-weeks of wearable heart-rate data that detects diabetes, high cholesterol, high blood pressure and sleep apnea, outperforming conventional biomarkers. Co-authored with Cardiogram's Johnson Hsieh and UCSF cardiologists.