Research Talk: Learning Distance Functions through Interactive Visualization

Abstract

Scientists collect important data every day and struggle with the size and complexity. When they turn to visualization or machine learning experts, the question arises, "What are the important facets of your data?" But while data creators know their data intimately, they may not be prepared to answer that question, especially not quantitatively. The work I'm presenting attempts to help such researchers. By iteratively refining a visualization of high dimensional data, they induce a distance function, learned behind-the-scenes, that shows the relative importance of the dimensions of the data. This work also represents a first step toward answering broader questions of how we can assist humans in understanding high-dimensional data and how to quantify what they already know intuitively.