Statistics Colloquium by Selin Gumustop '20

Exploring Minimax Linkage: Hierarchial Clustering With Prototypes by Selin Gumustop ’20, Statistics Colloquium, Wednesday, October 23, 1:10 – 1:50 pm, Stetson Court Classroom 105

Abstract:  Cluster analysis is a widely used class of techniques, which groups similar objects to better understand the underlying pattern in the data. Depending on the required conditions and/or assumptions, there are several existing clustering algorithms and models.  Among all, hierarchical clustering is the most intuitive one, which groups subjects based on some “dissimilarity” measures.  Specifically, the “linkage” function defines the distance between individuals or groups of points.  In this talk, I will first briefly present the basic idea and algorithm for hierarchical clustering with some commonly used linkage functions (e.g. single, average, complete, etc). Then I will introduce a newer linkage function called “Minimax”, and explore the advantages of it in comparison to the commonly used ones.  The theory of the properties of this new linkage function will be discussed, and the application of it will be illustrated through a few real data examples.