标题:Self-supervised representation learning from 12-lead ECG data
作者:Temesgen Mehari, Nils Strodthoff
期刊Computers in Biology and Medicine, 2022
DOI: 10.1016/j.compbiomed.2021.105114
关键词:自监督学习、对比学习、心电图(ECG)、时间序列分析、深度学习

12-lead ECG 是常见生物信号,但高质量标注数据稀缺,标注依赖专家,成本高,尤其对罕见病或睡眠相关心血管事件更难获取标签,然而,无标签 ECG 数据海量存在,因此利用自监督学习从大量无标签 ECG 中预训练模型,学习通用特征表示,再用于下游任务微调,提升性能与标签效率。

本文采用 xresnet1d50 作为主干网络,对比了不同的自监督学习方法。

# xresnet1d

https://timeseriesai.github.io/tsai/models.xresnet1d.html

# Instance Discrimination

# SimCLR

T. Chen, S. Kornblith, M. Norouzi, G. Hinton, A simple framework for contrastive learning of visual representations, in: International Conference on Machine Learning, 2020.

# BYOL

# SwAV

# CPC

# Latent Forecasting

Edited on

Give me a cup of [coffee]~( ̄▽ ̄)~*

NoResponse WeChat Pay

WeChat Pay

NoResponse Alipay

Alipay

NoResponse PayPal

PayPal