SurvICL: a Foundation Model for in-context Survival Analysis
SurvICL beat 20 survival-analysis baselines without tuning on 63 held-out datasets, the authors report.
It won 82.9% of pairwise comparisons on integrated Brier score, versus 71.3% for the runner-up, the Cox model. The authors also report it is the first model to rank above Cox on small datasets. SurvICL predicts survival curves in-context after pretraining entirely on synthetic data. Its generator supplies uncensored event times, allowing training with a plain quantile regression loss.
ArXiv · AI/CL/LG's note
It won 82.9% of pairwise comparisons on integrated Brier score, versus 71.3% for the runner-up, the Cox model. The authors also report it is the first model to rank above Cox on small datasets. SurvICL predicts survival curves in-context after pretraining entirely on synthetic data. Its generator supplies uncensored event times, allowing training with a plain quantile regression loss.
ArXiv · AI/CL/LG's note
score 5