A clinical deep learning framework for continually learning from cardiac signals across diseases, time, modalities, and institutions
oleh: Dani Kiyasseh, Tingting Zhu, David Clifton
| Format: | Article |
|---|---|
| Diterbitkan: | Nature Portfolio 2021-07-01 |
Deskripsi
Deep learning algorithms trained on data streamed temporally from different clinical sites and from a multitude of physiological sensors are generally affected by a degradation in performance. To mitigate this, the authors propose a continual learning strategy that employs a replay buffer.