Patricia Crossno is a Senior Member of Technical Staff in the Scalable Analysis and Visualization group at Sandia National Laboratories in Albuquerque, NM. Her current research interests include visual representations for abstract data, time series analysis and visualization, visualizing ensembles of simulations, and model comparison and evaluation.
Currently, my research is focused on ensemble analysis, which incorporates many components of my research interests.
TopicView: Visual Analysis of Topic Models and their Impact on Document Clustering
Patricia J. Crossno, Andrew T. Wilson, Timothy M. Shead, Warren L. Davis, IV, and Daniel M. Dunlavy
International Journal on Artificial Intelligence Tools, to appear
We present a new approach for analyzing topic models using visual analytics. We have developed TopicView, an application for visually comparing and exploring multiple models of text corpora, as a prototype for this type of analysis tool. TopicView uses multiple linked views to visually analyze conceptual or topical content, document relationships identified by the models, and the impact of the models on the results of document clustering. As case studies, we examine models created using two standard approaches: Latent Semantic Analysis (LSA) and Latent Dirichlet Allocation (LDA). Conceptual content is compared through the combination of (i) a bipartite graph matching LSA concepts with LDA topics based on the cosine similarities of model factors and (ii) a table containing the terms for each LSA concept and LDA topic listed in decreasing order of importance. Document relationships are examined through the combination of (i) side-by-side document similarity graphs, (ii) a table listing the weights for each document's contribution to each concept/topic, and (iii) a full text reader for documents selected in either of the graphs or the table. The impact of LSA and LDA models on document clustering applications is explored through similar means, using proximities between documents and cluster exemplars for graph layout edge weighting and table entries. We demonstrate the utility of TopicView's visual approach to model assessment by comparing LSA and LDA models of several example corpora.
TopicView: Visually Comparing Topic Models of Text Collections
Patricia J. Crossno, Andrew T. Wilson, Timothy M. Shead, and Daniel M. Dunlavy
Proceedings 23rd IEEE International Conference on Tools with Artificial Intelligence (ICTAI 2011), November 2011
We present TopicView, an application for visually comparing and exploring multiple models of text corpora. TopicView uses multiple linked views to visually analyze both the conceptual content and the document relationships in models generated using different algorithms. To illustrate TopicView, we apply it to models created using two standard approaches: Latent Semantic Analysis (LSA) and Latent Dirichlet Allocation (LDA). Conceptual content is compared through the combination of (i) a bipartite graph matching LSA concepts with LDA topics based on the cosine similarities of model factors and (ii) a table containing the terms for each LSA concept and LDA topic listed in decreasing order of importance. Document relationships are examined through the combination of (i) side-by-side document similarity graphs, (ii) a table listing the weights for each document’s contribution to each concept/topic, and (iii) a full text reader for documents selected in either of the graphs or the table. We demonstrate the utility of TopicView’s visual approach to model assessment by comparing LSA and LDA models of two example corpora.
TopicView: Understanding Document Relationships Using Latent Dirichlet Allocation Models
Patricia J. Crossno, Andrew T. Wilson, Daniel M. Dunlavy, and Timothy M. Shead
IEEE Workshop on Interactive Visual Text Analytics for Decision Making, October 2011
Document similarity graphs are a useful visual metaphor for assessing the conceptual content of a corpus. Algorithms such as Latent Dirichlet Allocation (LDA) provide a means for constructing such graphs by extracting topics and their associated term lists, which can be converted into similarity measures. Given that users' understanding of the corpus content (and therefore their decision-making) depends upon the outputs provided by LDA as well as how those outputs are translated into a visual representation, an examination of how the LDA algorithm behaves and an understanding of the impact of this behavior on the final visualization is critical. We examine some puzzling relationships between documents with seemingly disparate topics that are linked in LDA graphs. We use TopicView, a visual analytics tool, to uncover the source of these unexpected connections.
Visualization of Geologic Stress Perturbations Using Mohr Diagrams
Patricia Crossno, David H. Rogers, Rebecca M. Brannon, David Coblentz, and Joanne T. Fredrich
IEEE Transactions on Visualization and Computer Graphics, September/October 2005
Huge salt formations, trapping large untapped oil and gas reservoirs, lie in the deepwater region of the Gulf of Mexico. Drilling in this region is high-risk and drilling failures have led to well abandonments, with each costing tens of millions of dollars. Salt tectonics plays a central role in these failures. To explore the geomechanical interactions between salt and the surrounding sand and shale formations, scientists have simulated the stresses in and around salt diapirs in the Gulf of Mexico using nonlinear finite element geomechanical modeling. In this paper, we describe novel techniques developed to visualize the simulated subsurface stress field. We present an adaptation of the Mohr diagram, a traditional paper-and-pencil graphical method long used by the material mechanics community for estimating coordinate transformations for stress tensors, as a new tensor glyph for dynamically exploring tensor variables within three-dimensional finite element models. This interactive glyph can be used as either a probe or a filter through brushing and linking.
Patricia Crossno and David H. Rogers
IEEE Computer Graphics and Applications, November/December 2002
We developed an approach that uses our innate visual pattern recognition skills as part of the debugging process. Inspired by Huang's (1996) use of color to visualize energy distributions while untangling knots, we represented the particles graphically and color-coded them by energy value. Thus far, we've applied this approach to three domains: particle systems, cluster hardware configurations, and physics codes using finite element models. This debugging paradigm differs from software or program visualization in that we don't visualize software elements such as procedures, message passing between processors, or graph-based representations of data structures. In most application domains developers that use algorithm visualization tools must make decisions about what kind of visualization would best represent their code, and they must, in effect, code this visualization in addition to their application. For many developers, the time investment is too great compared to their perceived benefit, so they return to a traditional debugging approach. We believe that restricting the application domain increases the ease of use of visual debuggers. However, we go one step further by creating a, visual tool tailored to a particular application domain that can use either captured data or simulation outputs and requires no coding effort on the part of the user.
Spiraling Edge: Fast Surface Reconstruction from Partially Organized Sample Points
Patricia Crossno and Edward Angel
Proceedings Visualization '99 (Vis '99), October 1999
Many applications produce three-dimensional points that must be further processed to generate a surface. Surface reconstruction algorithms that start with a set of unorganized points are extremely time-consuming. Sometimes however, points are generated such that there is additional information available to the reconstruction algorithm. We present Spiraling Edge, a specialized algorithm for surface reconstruction that is three orders of magnitude faster than algorithms for the general case. In addition to sample point locations, our algorithm starts with normal information and knowledge of each point's neighbors. Our algorithm produces a localized approximation to the surface by creating a star-shaped triangulation between a point and a subset of its nearest neighbors. This surface patch is extended by locally triangulating each of the points along the edge of the patch. As each edge point is triangulated, it is removed from the edge and new edge points along the patch's edge are inserted in its place. The updated edge spirals out over the surface until the edge encounters a surface boundary and stops growing in that direction, or until the edge reduces to a small hole that is filled by the final triangle.
Isosurface Extraction Using Particle Systems
Patricia Crossno and Edward Angel
Proceedings Visualization '97 (Vis '97), October 1997
Presents a new approach to isosurface extraction from volume data using particle systems. Particle behavior is dynamic and can be based on laws of physics or artificial rules. For isosurface extraction, we program particles to be attracted towards a specific surface value while simultaneously repelling adjacent particles. The repulsive forces are based on the curvature of the surface at that location. A birth-death process results in a denser concentration of particles in areas of high curvature and sparser populations in areas of lower curvature. The overall level of detail is controlled through a scaling factor that increases or decreases the repulsive forces of the particles. Once particles reach equilibrium, their locations are used as vertices in generating a triangular mesh of the surface. The advantages of our approach include: vertex densities are based on surface features rather than on the sampling rate of the volume; a single scaling factor simplifies level-of-detail control; and meshing is efficient because it uses neighbor information that has already been generated during the force calculations.