Skip to main content
To KTH's start page To KTH's start page

Static Visualizations for Dynamic Hierarchies

Time: Wed 2022-12-07 10.00

Location: Visualization Studio VIC, floor 4, Lindstedtsvägen 5

Video link: zoom link for online defense

Language: English

Subject area: Computer Science

Doctoral student: Wiebke Köpp , Beräkningsvetenskap och beräkningsteknik (CST)

Opponent: Professor Vijay Natarajan, Indian Institute of Science

Supervisor: Professor Tino Weinkauf, Beräkningsvetenskap och beräkningsteknik (CST)

Export to calendar

QC 20221111


Dynamic hierarchical data describes phenomena in a wide range of domains, from file management to demographics and business administration, as well as feature extraction results in spatial scientific data. As with all data, visualization is an integral step for gaining understanding about dynamic hierarchical data. In contrast to the visualization of individual static hierarchies, the visualization of dynamic hierarchies comes with additional challenges since many different aspects of a hierarchy may be subject to change.

This thesis presents novel visualizations, compact data structures, and layout optimizations for dynamic hierarchies where both topology and data may change. Particular focus is placed on the type of hierarchies that stem from features in scalar fields, namely merge trees and derived discretized feature hierarchies. We propose several visualization schemes that summarize dynamic hierarchies statically by creating stacked one-dimensional representations. The stacking dimension corresponds to the data’s dynamic-inducing variable which is usually time. In contrast to animating individual visualizations for each variable setting or time step, our static overviews of the entirety of the data facilitate comparison both between multiple data sets and across the stacking dimension. For the preservation of a user’s mental map, we utilize correspondences between parts of the hierarchy to optimize the one-dimensional representations toward stability in regards to the dynamic-inducing dimension. To evaluate our proposed methods, we apply them to several real-world data sets, compare them against existing approaches, and study how the resulting visualizations are affected by method parameters.