Campus Event Calendar

Event Entry

What and Who

Hierarchical clustering beyond heuristics: Quality guarantees without speed sacrifices

Evangelos Kipouridis
Max-Planck-Institut für Informatik - D1
Joint Lecture Series
AG 1, AG 2, AG 3, INET, AG 4, AG 5, D6, SWS, RG1, MMCI  
Public Audience

Date, Time and Location

Wednesday, 3 April 2024
60 Minutes
E1 5


The algorithmic task of constructing hierarchical representations of
data has been studied by various communities over many decades. Their
applications range from statistics and databases to the analysis of
complex networks and, more recently, machine learning, where they have
proven useful for understanding text, images, graphs and
multi-relational data. The reason why hierarchical representations are
so ubiquitous is that many data sets stemming from nature or society are
organized according to a latent hierarchy. Furthermore, in contrast to
“flat” clustering techniques, like k-means or k-median which cannot
capture fine-grained relationships among points, hierarchical clustering
reveals the structure of a data set at multiple levels of granularity

Despite of the plethora of applications, the theory behind hierarchical
clustering is underdeveloped, and popular heuristics offer little formal
guarantees. In this talk I will present my work on algorithms with near
optimal quality guarantees; in fact, in certain cases the algorithms run
in near linear time, bridging the gap between theory and practice.
Finally, I will discuss how to incorporate domain specific knowledge,
leading to semi-supervised hierarchical clustering, as opposed to the
traditional view of hierarchical clustering as an unsupervised learning


Jennifer Müller
+49 681 9325 2900
--email hidden

Virtual Meeting Details

997 1565 5535
passcode not visible
logged in users only

Jennifer Müller, 03/27/2024 13:52 -- Created document.