Multitask learning and benchmarking with clinical time series data.

Abstract in brief

• The paper introduces four clinical prediction benchmarks using data from the MIMIC-III database to help measure progress in healthcare machine learning, addressing the lack of publicly available benchmark datasets.
• These benchmarks cover important healthcare problems such as predicting mortality risk, estimating hospital stay length, detecting physiological decline, and classifying patient phenotypes.
• The authors evaluate various neural network models, including techniques like deep supervision and multitask learning, to improve performance on these tasks compared to simpler linear models.]