DeepPseudo: Pseudo Value Based Deep Learning Models for Competing Risk Analysis

Authors: Md Mahmudur Rahman, Koji Matsuo, Shinya Matsuzaki, Sanjay Purushotham479-487

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on real and synthetic datasets demonstrate that our proposed models obtain promising and statistically significant results compared to the state-of-the-art CRA approaches.
Researcher Affiliation Academia 1 Department of Information Systems, University of Maryland, Baltimore County, Baltimore, Maryland, USA 2 University of Southern California, Los Angeles, California, USA 3 Osaka University, Osaka, Japan
Pseudocode No The paper describes the models and their variants but does not include any pseudocode or algorithm blocks.
Open Source Code Yes Our codes and supplementary materials are at this link3. 3https://github.com/umbc-sanjaylab/Deep Pseudo AAAI2021
Open Datasets Yes Datasets: We conducted experiments on two real-world datasets and one synthetic dataset. SEER: The Surveillance, Epidemiology, and End Results (SEER)1 Program provides information on cancer statistics to reduce the cancer burden in the United States. 1https://seer.cancer.gov/ ... WIHS: We selected a cohort of 1164 women enrolled in WIHS (Bacon et al. 2005) study...
Dataset Splits Yes Implementation details: We performed stratified 5-fold cross-validation so that a constant censoring ratio2 is maintained in each fold. We used one-hot encoding for representing categorical variables. We obtain the (ground-truth) pseudo values for CIF using the jackknife function of R package prodlim for each cause and evaluation time point (separately for training and validation sets). For Deep Pseudo models, early stopping was performed, and the best model was chosen based on the performance on the validation fold.
Hardware Specification Yes We ran experiments on a 128GB RAM Intel Xeon dual 10-core processor with 3 GPUs.
Software Dependencies No The paper mentions 'R package prodlim' and 'python library innvestigate' but does not specify version numbers for any software dependencies.
Experiment Setup Yes We fine-tuned hyperparameters by random search over the number of layers, number of neurons, learning rate, batch size. We used dropout, Adam optimizer, and selu activation in training.