12 July 2018
Weakly Supervised Machine Learning in Medical Sensing: Integrating Expert Knowledge and Programming with Data
One of the most significant roadblocks to using modern machine learning models is collecting hand-labelled training data at the massive scale they require. In real-world settings such as clinical medicine, where domain expertise is needed, and modelling goals change rapidly, hand-labelling training sets are prohibitively slow, expensive, and static. For these reasons, practitioners are increasingly turning to weak supervision techniques wherein noisier, often programmatically-generated labels are used instead. In this talk, we will discuss recent developments in applying various types of weak supervision to problems in medical imaging and diagnostics, and assess future areas wherein the confluence of inadequate control and massive, unlabeled datasets could lead to new discoveries and reductions to an application.