TEMPO published at CHIL 2026

TEMPO: Transformers for Temporal Disease Progression from Cross-Sectional Data, led by Hongtao Hao, was published at the Conference on Health, Inference, and Learning (CHIL) 2026 and appears in the Proceedings of Machine Learning Research.

Event-based models infer biomarker progression from cross-sectional data, but typically recover only an ordinal sequence and rely on rigid assumptions. TEMPO trains a transformer on data simulated from the probabilistic model — keeping the generative model where it is indispensable, while learning inference that is both faster and more accurate than our original method.

More on the disease progression project.