Oral Candidacy - Yi Gu
|Start:||12/1/2014 at 2:00PM|
|End:||12/1/2014 at 5:00PM|
|Location:||258 Fitzpatrick Hall|
December 1, 2014 258 Fitzpatrick Hall 2:00pm
Dr. Chaoli Wang
Dr. Tom Peterka Dr. Aaron Striegel Dr. Dong Wang
Students/Faculty are welcome to attend the presentation portion of the defense
Data Visualization using Graph-based Representations
Data visualization consists of scientific visualization and information visualization. In this dissertation, we focus on time-varying multivariate volume visualization in scientific visualization and visualization of image and text collections in information visualization.
Time-varying multivariate volume visualization plays an essential role in many scientific, engineering and medical disciplines. Many works have been done in developing novel algorithms and techniques for processing, managing and rendering scientific datasets. However, several challenges still remain. The first challenge is solving the visual occlusion and clutter problem when visualizing a 3D volumetric data. The occlusion is inevitable due to the projection of the 3D data to a 2D screen during the rendering. This problem is exacerbated when the time and variable dimensions are considered. The second challenge is the lack of capability to help users analyze and track data change or evolution in an occlusion-free, controllable and adaptive fashion. As the size of data keeps increasing, these challenges become more and more severe. To solve these problems, several isolated works that utilized information visualization techniques such as parallel coordinates and treemaps were introduced.
This dissertation proposes a series of advanced techniques that leverage the more generalized, understandable and familiar form of graphs to address a wider range of scientific visualization problems. The underlying relations of the data are extracted for analyzing the time-varying volumetric data, followed by careful and novel designs of the graph representations to illustrate these relations in an occlusion-free and controllable fashion. First, we will present our graph-based representation TransGraph that maps the time evolution of a 4D data set to a 2D graph space and guides data exploration and tracking. Second, we will introduce IVAG which incorporates graph analytics to extract deep meanings from a graph-based representation for exploring time-varying volumetric data. Third, we will present our approach to construct an indexable tree named iTree for time-varying data which integrates efficient data compacting, indexing, querying and classification into a single framework.
Visualization of image and text collections is very practical in our daily lives. A significant challenge is to design an interface that visualizes the relations among images and texts clearly and provides a dynamic layout as users explore different images and texts in the collection for sense-making. This dissertation also presents a framework called iGraph that incorporates a graph-based representation to show the relations among images and texts in a dynamic fashion with progressive drawing capability.
Future work includes the development of TrendGraph, a dynamic graph representation that not only helps users understand the underlying relations in each time step, but also the evolution over time. Trend and detrends are utilized to classify the volume regions and graph analysis techniques such as community detection, community matching and anomaly detection are utilized. Community detection organizes nodes with similar or close relations into different groups, allowing visual comparison between groups of nodes instead of individual nodes. Community matching connects communities with close relations, allowing visual tracking for better understanding the time evolution. Anomaly detection identifies the nodes that behave differently from the others, allowing identifying abnormal events in the data.