Uncertainty quantification for machine learning models applied to photoplethysmography signals

光电体积图(PPG)信号使用廉价的设备很容易非侵入性地采集，并被用于临床和可穿戴设备中进行家庭监测。人们认识到，PPG信号包含丰富的有价值的生理信息，用于监测或诊断一系列健康状况。机器学习(ML)被应用于PPG信号，但缺乏关于可信度的工作，这在医学背景下是至关重要的。通过开发对应用于PPG信号的ML的数据和模型不确定性进行量化的方法，该项目旨在生成基准数据集以对此类模型进行基准测试，并识别具有高精度和低不确定性的模型，从而提供适合实施的可靠模型。

Photoplethysmogram (PPG) signals are easy to collect non-invasively using cheap devices and are used in the clinic and in wearable devices for home monitoring. It is recognised that PPG signals contain a wealth of valuable physiological information for monitoring or diagnosing a range of health conditions. Machine learning (ML) is applied to PPG signals but there is a lack of work on trustworthiness, which is crucial in a medical context. By developing methods to quantify both the data and model uncertainty for ML applied to PPG signals, this project aims to generate reference datasets to benchmark such models and to identify models with high accuracy and low uncertainty thus providing trustworthy models that are ripe for implementation.