Contrastive Learning for Clinical Outcome Prediction with Partial Data Sources

Authors: Meng Xia, Jonathan Wilson, Benjamin Goldstein, Ricardo Henao

ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We present experiments on two real-world datasets demonstrating that CLOPPS consistently outperforms strong baselines in several practical scenarios.
Researcher Affiliation Academia 1Department of Electrical and Computer Engineering, Duke University, Durham, US 2Department of Biostatistics and Bioinformatics, Duke University, Durham, US 3King Abdullah University of Science and Technology, Thuwal, KSA.
Pseudocode No The paper does not contain any explicit pseudocode or algorithm blocks.
Open Source Code Yes The source code used in the experiments is available at https://github.com/mx41-m/Contrastive-Learning.git.
Open Datasets Yes Given that the Private dataset is not readily publicly accessible, we also validate CLOPPS using the MIMIC-III clinical database (Johnson et al., 2016).
Dataset Splits Yes These sequences were then divided into training, validation, and test datasets following an 8 : 1 : 1 ratio.
Hardware Specification No The paper mentions that models were trained and experiments conducted, but it does not provide any specific details about the hardware used, such as GPU models, CPU specifications, or memory.
Software Dependencies No The paper mentions using the Hugging Face's transformers library, but it does not specify any version numbers for this or any other software dependencies, which is necessary for reproducibility.
Experiment Setup Yes The encoders for CLOPPS are trained for 50, 100 and 100 epochs on MMNIST, Private and MIMIC, respectively. The classifiers for CLOPPS are trained for 10, 5 and 5 epochs on MMNIST, Private and MIMIC, respectively. In CLOPPS, the values for τ, w and d are set to 0.1, 2 and 12, respectively, based on experimental results. For all models (excluding Elastic Net), Adam W (Loshchilov & Hutter, 2017) is employed as the optimizer. The values for the learning rate, beta, weight decay and batch size are set for all models to 10 4, (0.9, 0.999), 0.01, and 64, respectively.