CLOCS: Contrastive Learning of Cardiac Signals Across Space, Time, and Patients

Authors: Dani Kiyasseh, Tingting Zhu, David A Clifton

ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We conduct our experiments1 using Py Torch (Paszke et al., 2019) on four ECG datasets that include cardiac arrhythmia labels.
Researcher Affiliation Academia 1Department of Engineering Science, University of Oxford, Oxford, United Kingdom 2Oxford-Suzhou Centre for Advanced Research, Suzhou, China.
Pseudocode No The paper does not contain any pseudocode or clearly labeled algorithm blocks.
Open Source Code Yes Code can be accessed at: https://github.com/ danikiyasseh/CLOCS
Open Datasets Yes We conduct our experiments1 using Py Torch (Paszke et al., 2019) on four ECG datasets that include cardiac arrhythmia labels. Physio Net 2020 (Perez Alday et al., 2020) consists of 12-lead ECG recordings from 6,877 patients alongside 9 different classes of cardiac arrhythmia. Chapman (Zheng et al., 2020) consists of 12-lead ECG recordings from 10,646 patients alongside 11 different classes of cardiac arrhythmia. Physio Net 2017 (Clifford et al., 2017) consists of 8,528 single-lead ECG recordings alongside 4 different classes. Cardiology (Hannun et al., 2019) consists of single-lead ECG recordings from 328 patients alongside 12 different classes of cardiac arrhythmia.
Dataset Splits Yes All datasets were split into training, validation, and test sets according to patient ID using a 60, 20, 20 configuration.
Hardware Specification No The paper does not specify the hardware used (e.g., GPU/CPU models, memory) for running experiments.
Software Dependencies No The paper mentions "Py Torch (Paszke et al., 2019)" but does not specify a version number for PyTorch or any other software dependencies.
Experiment Setup Yes During self-supervised pre-training, we chose the temperature parameter, τ = 0.1, as per (Chen et al., 2020). For BYOL, we chose the decay rate, τd = 0.90, after experimenting with various alternatives (see Appendix F). For all experiments, we use a neural architecture composed of three 1D convolutional layers followed by two fully connected layers.