A Deep Variational Approach to Clustering Survival Data
Authors: Laura Manduchi, Ričards Marcinkevičs, Michela C. Massi, Thomas Weikert, Alexander Sauter, Verena Gotta, Timothy Müller, Flavio Vasella, Marian C. Neidert, Marc Pfister, Bram Stieltjes, Julia E Vogt
ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We compare our model to the related work on clustering and mixture models for survival data in comprehensive experiments on a wide range of synthetic, semi-synthetic, and real-world datasets, including medical imaging data. |
| Researcher Affiliation | Academia | 1ETH Z urich; 2Politecnico di Milano; 3CADS, Human Technopole; 4University Hospital Basel; 5University Children s Hospital Basel; 6University of Z urich; 7St. Gallen Cantonal Hospital |
| Pseudocode | No | The paper describes its methodology using mathematical formulations and descriptive text, but it does not include any explicitly labeled "Pseudocode" or "Algorithm" blocks. |
| Open Source Code | Yes | The code is publicly available at https: //github.com/i6092467/vadesc. |
| Open Datasets | Yes | We evaluate Va De SC on a range of synthetic, semi-synthetic (surv MNIST; P olsterl (2019)), and real-world survival datasets with varying numbers of data points, explanatory variables, and fractions of censored observations (see Table 2). In particular, real-world clinical datasets include two benchmarks common in the survival analysis literature, namely SUPPORT (Knaus et al., 1995) and FLChain (Kyle et al., 2006; Dispenzieri et al., 2012); an observational cohort of pediatric patients undergoing chronic hemodialysis (Hemodialysis; Gotta et al. (2021)); an observational cohort of high-grade glioma patients (HGG); and an aggregation of several computed tomography (CT) image datasets acquired from non-small cell lung cancer (NSCLC; Aerts et al. (2019); Bakr et al. (2017); Clark et al. (2013); Weikert et al. (2019)). |
| Dataset Splits | Yes | XGBoost was trained for 5,000 boosting rounds with a maximum tree depth of 6, a learning rate equal to 0.01, a subsampling rate of 0.5, and early stopping after 200 rounds without an improvement on the validation set (set to 10% of the training data). |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU or CPU models, processor types, or memory amounts used for running the experiments. It generally discusses computational aspects without specifying the underlying hardware. |
| Software Dependencies | Yes | Va De SC: we implemented our model in Tensor Flow 2.4 (Abadi et al., 2015). |
| Experiment Setup | Yes | Va De SC hyperparameters across all datasets are reported in Table 11. We use L = 1 Monte Carlo samples for the SGVB estimator. |