Domain Generalization

Engineers operating in the realm of medical machine learning and biomedical engineering have a non-trivial responsibility to ensure the stability and robustness of the technologies they develop.

Yet, when applied to machine learning — a discipline inherently focused on analysis and prediction in novel settings — how do we mandate the robustness of our algorithms?

A growing research area, domain generalization, is concerned with the quantification and implementation of methods designed to best achieve this desired stability.

Our applications at Medibio are tasked with training algorithms on data from certain medical centers and hospitals. However, in order to meet user demands, it is essential that we deploy our technology to geographically unique, independent environments. To this end, domain generalization forms a significant component of our current engineering architecture: ensuring model stability and reliability and offering mathematical bounds to assess the ubiquitous utility of our offerings.

This is a simple, yet illuminating, illustration of how Medibio leverages the latest developments in computational methods in our pursuit of societal changing technology.