ML-AIM Machine Learning and Artificial Intelligence for Medicine

Research Laboratory led by Prof. Mihaela van der Schaar

AutoPrognosis: Automated Clinical Prognostic Modeling via Bayesian Optimization with Structured Kernel Learning
Ahmed M. Alaa and Mihaela van der Schaar

An AutoML approach to prognostic modeling!

Clinical prognostic models derived from largescale healthcare data can inform critical diagnostic and therapeutic decisions. To enable off-theshelf usage of machine learning (ML) in prognostic research, we developed AUTOPROGNOSIS: a system for automating the design of predictive modeling pipelines tailored for clinical prognosis. AUTOPROGNOSIS optimizes ensembles of pipeline configurations efficiently using a novel batched Bayesian optimization (BO) algorithm that learns a low-dimensional decomposition of the pipelines' high-dimensional hyperparameter space in concurrence with the BO procedure. This is achieved by modeling the pipelines' performances as a black-box function with a Gaussian process prior, and modeling the similarities between the pipelines' baseline algorithms via a sparse additive kernel with a Dirichlet prior. We demonstrate the utility of AUTOPROGNOSIS using 10 major patient cohorts representing various aspects of cardiovascular patient care.

Read our paper!

The AutoPrognosis system

The core component of AUTOPROGNOSIS is an algorithm that automatically configures ML pipelines, where every pipeline comprises algorithms for missing data imputation, feature preprocessing, prediction, and calibration. The total number of hyperparameters in AUTOPROGNOSIS is 106. We use Bayesian Optimization (BO) to configure the ML pipelines.

We cluster algorithms via kernel decomposition!

The key idea of our BO algorithm is that for a given dataset, the performance of a given group of algorithms may not be informative of the performance of another group of algorithms. Since the Gaussian process kernel encodes the correlations between the performances of the different pipeline configurations, the underlying "informativeness" structure that relates the different hyperparameters can be expressed via a sparse additive kernel decomposition.

View Project