Cohesion: A Social Perspective on Clustering

In making sense of the world around us, we instinctively identify clusters of similar objects and individuals.  Yet many clustering algorithms require that we describe our data with respect to a small set of variables or provide input distances. This initial step can be quite challenging for many interesting problems.  To sidestep this, one could develop algorithms built on responses to comparisons of the form: among A, B, and C, which one is the outlier?In this talk, I’ll share a new measure of relative proximity called “cohesion” and explain how it arises from a simple social story.  I’ll discuss some of its desirable properties and how it could provide a new perspective on clustering.  The talk will be equal parts ideas, results, and applications; and will be rather non-technical.