Multi-task deep learning for cardiac rhythm detection in wearable devices

摘要

The proposed model architecture for DeepBeat, two tasks are shown: (top) unsupervised pretraining and (bottom) supervised learning through fine-tuning. The top represents the pretraining process on the unlabeled simulated data, and the bottom represents the multitask fine-tuning process on the labeled data. The trained encoder weights serve as the foundational layers of the multitask model.