Learning Cluster Analysis through Experience

Abstract

The field of machine learning is constantly developing useful new techniques for data analysis, but they are often ignored by researchers outside the field due to unfamiliarity and the difficulty of keeping up with a large body of work. We propose a methodology for training researchers how algorithms work through experience, such that they gain an implicit, rather than explicit, understanding of their function. Thus we combine theory from discovery learning with advanced software and a more educated target population to foster such understanding. We have developed an open source application for exploratory data analysis called Divvy that lets users quickly and visually interact with a range of data analysis techniques. Using a simplified version of Divvy, we find that undergraduate subjects are generally able to learn machine learning concepts through experience, though they have only partial success in applying them.


Back to Table of Contents