Adapting Tabular Foundation Models for Time-to-Event Prediction
September 7, 2026
Researchers evaluated adapting Tabular Foundation Models (TabFMs) to censored time-to-event tasks using CoxPH and DeepHit interfaces. Findings show that while zero-shot inference works for small datasets, supervised adaptation via survival-head tuning is necessary as dataset scale increases.
HOW THIS AFFECTS YOU
●
researcherYou can leverage frozen TabFM backbones for survival analysis using supervised adaptation.
●
healthThis provides a method for applying large-scale tabular models to clinical time-to-event forecasting.