Machine Learning on Medical Data: Methods, Needs, Ambitions

A modern, evidence-driven approach to healthcare relies on the accumulation of electronic medical records. Data sets with medical histories now cover decades and millions of patients.

Most of this data is never made available to external researchers. In this talk, Roderick discusses a common question in machine learning for healthcare: "Will patient X need resource Y due to a rare event Z?" While generic, this may require the development of specialized algorithms to deal with extremely sparse, noisy, and imbalanced medical data.

Roderick also describes some of the more ambitious projects that rely on multiple sources of medical data combined with analysis of the patient's financial behavior and lifestyle choices.

Event Topic

Data Science