KAPLAN: New AI method improves survival analysis with Kolmogorov-Arnold networks

Classical survival analysis methods like the Cox model and generalised additive models (GAMs) require manual specification of interactions and time-varying effects—a major bottleneck for rich clinical datasets. Now, researchers from Cambridge and NeurIPS 2026 present KAPLAN-HR, a novel architecture based on Kolmogorov-Arnold Networks (KANs) that nonparametrically estimates the hazard function as a joint function of covariates and time. A single-layer KAPLAN-HR recovers a GAM, while deeper architectures automatically learn complex interactions and time-varying effects through function composition.

KAPLAN-HR also offers a theoretical guarantee: its convergence rate depends only on the smoothness of the underlying KAN representation, not on the covariate dimension—mitigating the curse of dimensionality for KAN-representable targets. In evaluations over six clinical benchmark datasets, KAPLAN-HR matches or exceeds the predictive performance of established statistical and deep learning survival methods. The paper (9 pages + 13 supplementary) is submitted to NeurIPS 2026 and available on arXiv.