TabSurv Adapts Tabular Neural Networks for Survival Analysis
TabSurv is a new approach that adapts modern tabular neural network architectures to survival analysis, using either the Weibull distribution or non-parametric survival prediction. It introduces SurvHL, a novel histogram loss function designed to handle censored data. The method includes deep ensembles of MLPs trained in parallel, optimizing survival distribution parameters before averaging to promote diversity. Empirical evaluation on 10 diverse real-world survival datasets demonstrates its effectiveness.
Key facts
- TabSurv adapts modern tabular architectures to survival analysis.
- Uses Weibull distribution or non-parametric survival prediction.
- Optimizes SurvHL, a novel histogram loss function supporting censored data.
- Includes deep ensembles of MLPs for survival analysis.
- Ensemble components are trained in parallel.
- Optimizes survival distribution parameters before averaging.
- Promotes diversity across ensemble component predictions.
- Evaluated on 10 diverse real-world survival datasets.
Entities
Institutions
- arXiv