The increasing adoption of electronic medical records offer a rich set of data for analysis and predictive modeling in hospital care settings.
Efficient learning of latent variable models
Latent variable models are notoriously difficult to learn because of the non-convexity of the likelihood objective function when latent variables are introduced. Method of moments algorithms provide can provide globally optimal solutions with consistency guarantees.
Electronic phenotyping with anchors
Certain highly informative variables, called anchors, can help learn classification functions that can be used to classify patients with less clear presentations of the disease.