Censor Dependent Variational Inference

Authors: Chuanhui Liu, Xiao Wang

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments validate our analysis and demonstrate significant improvements in the estimation of individual survival distributions. Codes can be found at https://github.com/Chuanhui Liu/CDVI. (...) Section 5 presents empirical validation and findings through extensive experiments on various datasets. Notably, our methods achieve 5% higher C-index than state-of-the-art models on WHAS datasets.
Researcher Affiliation Academia Chuanhui Liu 1 Xiao Wang 1 (...) 1Department of Statistics, Purdue University, USA. Correspondence to: Xiao Wang <EMAIL>.
Pseudocode No The paper describes methods and proofs but does not contain a clearly labeled section or figure titled "Pseudocode" or "Algorithm", nor does it present structured, code-like procedural steps.
Open Source Code Yes Codes can be found at https://github.com/Chuanhui Liu/CDVI.
Open Datasets Yes Table 3: Summary table for benchmark clinical datasets. (...) Table 3 summarizes the real-world datasets. These models include Cox-PH (Cox, 1972)...
Dataset Splits Yes Train-validation-test split ratio is 0.6, 0.2, 0.2. Experiment repetition is 5, using the same seeds of dataset split.
Hardware Specification No The experiments are on Python 3.9 with Pytorch on the Windows 11 system. GPU is not required. No specific GPU or CPU models, memory details, or detailed computer specifications are provided.
Software Dependencies No The experiments are on Python 3.9 with Pytorch on the Windows 11 system. (...) All models were implemented using the Python package by Nagpal et al. (2022), and our implementation follows its API for ease of reproducibility. (...) C index is implemented via Python package Pycox by the authors of Kvamme et al. (2019) (...) Ctd and Brier score is implemented via Python package Scikit-Survival (P olsterl, 2020). While Python 3.9 is specified, other key software components like Pytorch, Pycox, or Scikit-Survival do not have explicit version numbers mentioned.
Experiment Setup Yes C.3. Hyper-parameters of training CD-CVAE and the variants (...) We have tuned the following hyper-parameters: Distribution Family of decoders: (...) Dropout: (...) Latent dimension: (...) Learning rate: 0.01, 0.001. batch size: 20, 100, 250, 500, 1000. Patience: (...) Temperature: (...)