Survival Diffusion Probabilistic Model for Continuous-Time Survival Analysis
A new generative method for continuous-time survival analysis, known as the Survival Diffusion Probabilistic Model (SDPM), has been introduced by researchers. This model utilizes a denoising diffusion technique to represent the conditional distribution of survival outcomes, eliminating the need for parametric assumptions regarding event-time distribution and time axis discretization. When censoring is conditionally independent, the conditional samples are converted into survival function estimates using the Kaplan-Meier estimator. This innovative approach overcomes the constraints of current methods that either enforce structural assumptions on the hazard function or require time discretization.
Key facts
- SDPM stands for Survival Diffusion Probabilistic Model.
- It is a generative approach for continuous-time survival analysis.
- It models the conditional distribution of survival outcomes using a denoising diffusion model.
- It avoids parametric assumptions on the event-time distribution.
- It does not require discretization of the output time space.
- It uses the Kaplan-Meier estimator to transform conditional samples into survival function estimates.
- The model assumes conditionally independent censoring.
- The paper is available on arXiv with ID 2605.22776.
Entities
Institutions
- arXiv