| Description |
A broad range of applications capture dynamic data at an unprecedented scale. Independent of the application area, finding intuitive ways to understand the dynamic aspects of these increasingly large data sets remains an interesting and, to some extent, unsolved research problem. Generically, dynamic data sets can be described by some, often hierarchical, notion of feature of interest that exists at each moment in time, and those features evolve across time. Consequently, exploring the evolution of these features is considered to be one natural way of studying these data sets. Usually, this process entails the ability to: 1) define and extract features from each time step in the data set; 2) find their correspondences over time; and 3) analyze their evolution across time. However, due to the large data sizes, visualizing the evolution of features in a comprehensible manner and performing interactive changes are challenging. Furthermore, feature evolution details are often unmanageably large and complex, making it difficult to identify the temporal trends in the underlying data. Additionally, many existing approaches develop these components in a specialized and standalone manner, thus failing to address the general task of understanding feature evolution across time. This dissertation demonstrates that interactive exploration of feature evolution can be achieved in a non-domain-specific manner so that it can be applied across a wide variety of application domains. In particular, a novel generic visualization and analysis environment that couples a multiresolution unified spatiotemporal representation of features with progressive layout and visualization strategies for studying the feature evolution across time is introduced. This flexible framework enables on-the-fly changes to feature definitions, their correspondences, and other arbitrary attributes while providing an interactive view of the resulting feature evolution details. Furthermore, to reduce the visual complexity within the feature evolution details, several subselection-based and localized, per-feature parameter value-based strategies are also enabled. The utility and generality of this framework is demonstrated by using several large-scale dynamic data sets. |
