Generalization: The Key to Building Reusable Healthcare Rules

One of the biggest challenges in AI today is the fact that most machine-learning based systems don’t learn how to generalize well. They can’t take two similar concepts and connect them together to form a named concept that can be reused in many areas. In this talk, we will discuss the current limits of machine learning and how new knowledge representations can be better at generalization. We will discuss the work of Jeff Hawkins, and Francos Chollet on generalization and then describe how we can structure Healthcare Concepts in the form of graphs to build composable rules that can be reused in many context. This presentation will focus on the practical work for doing analysis and surveillance on integrated healthcare data by Mark Megerian and others within the Healthcare Graph team and OptumHealth.