Position Paper by Jeffrey Heer, UC Berkeley

For the Workshop on "Information Visualization Software Infrastructures" at IEEE 2004 Visualization,
Organized by Katy Börner, Indiana University, USA and Jean-Daniel Fekete, INRIA, France

Jeffrey Heer
Group for User Interface Research
Berkeley Institute of Design
University of California, Berkeley

Part I

I.1) What functionality should a general InfoVis infrastructure provide?

As a baseline, a general infovis infrastructure should provide usable mechanisms for deploying any of a host of established interactive visualization techniques to new data while also serving as a platform for developing, integrating, and evaluating new visualizations. Some of the specific functionalities necessary to achieving this include:

• Data representation, import, and export. (including database support)
  • Tabular data
  • Time-series data
  • Graphs and Trees
• Common data processing and transformation algorithms
• A modular, integrated library of visualization and interaction techniques
• Common associated interaction and processing technologies (e.g., dynamic queries, brushing, search)
• Flexible rendering, animation support
• Multiple views (e.g., allowing comparison and "small multiples")
• Support for both systems-level and usability evaluations
• Tested, usable application programming interface (API)

Perhaps most importantly, the above features must cohere in a well-designed, integrated architecture. For example, components such as layout algorithms, data transformations, rendering modules and the like need to be reusable and composable across various visualization applications. By analogy to user interface toolkits, what is needed is not just visualization "widgets" that a user can select from, but a deeper decomposition of the visualization design space into reusable components that can function across varied visualization designs.

The above features will help ensure a useful platform for both researchers and developers to use and build various visualization applications. I would also, however, like to see such an infrastructure leveled at the construction of tools by which end users can craft their own custom visualizations. Everyday, millions of people use tools such as those in Microsoft Excel to generate a wide range of data graphics. For information visualization technologies to reach their true potential, tools by which to create customized visualizations in a direct manipulation fashion are sorely needed. I would hope that a general infovis architecture would be designed with these additional goals in mind, that infovis techniques and technologies become constructive tools accessible to everyday users.

I.2) What do you see as the main technical challenges for creating a central but flexible and universally useful (information) visualization software infrastructure (as opposed to 100 different ones)?

I think the main "technical" challenge is API design. With a skilled programmer at hand, recreating any existing, published visualization or algorithm is not intellectually all that difficult. However, decomposing the space of visualizations into individual, composable building blocks that can support both existing and future visualization designs while remaining usable is a hard problem. One must at one time balance systems issues such as memory footprint, scalability and response times; design issues such as flexibility and extensibility; and programmer usability of the resulting abstractions.

I thinks this challenge also perserveres when considering the earlier expressed goal of creating visual, end-user tools for designing visualizations. Building such tools will require that these goals also be considered in the design of the underlying programming interfaces of the supporting infrastructure. The conceptual model of the API design, though implemented in code, should hopefully be clear enough to also serve as the model by which to visually construct an infovis application.

Part II

Please describe the (information) visualization software infrastructure you are working on.

II.1) Project Name and Web Address

prefuse - a toolkit for interactive information visualization
http://prefuse.sourceforge.net

The website includes access to the toolkit source code, video and interactive demonstrations, and links to research papers and articles about or involving the toolkit.

II.2) Core Team Members (Please list in order, Role of Project Member, Full Name, E-mail. eg: Developer, John Doe, jdoe@univ.edu)

Designer and Developer, Jeffrey Heer
Advisor, Stuart K. Card,
Advisor, James Landay

II.3) Project Start Date

Project started in spring 2003.

II.4) Targeted User Group

Java programmers and user interface designers.

II.5) Supported User Tasks

The prefuse toolkit is a platform for building interactive visualizations, as such it can be used to build custom applications for any number of user tasks. To support common user needs, the architecture includes support for common interactions (e.g., focus selection, brushing, repositioning, panning and zooming), visualization techniques (e.g., animation, various layout and distortion algorithms), and indexing and searching of data. See the snapshots below for examples of some of the end-user applications constructed with the toolkit.

II.6) Major Features of the System Architecture

Informed by a review of existing applications, our own years of experience designing novel visualizations, and earlier toolkit evaluations, the prefuse toolkit offers:

• Data structures and I/O libraries (including database connectivity) for unstructured, graph, and tree data
• Multiple visualizations of a single data source
• Multiple views of a single visualization
• Scalability to thousands of on-screen items, and to backing data sources with millions of elements
• Batch processing of data using composable modules
• A library of provided layout and distortion techniques
• Animation and time-based processing
• Graphics transforms, including panning and zooming
• A full force simulator for physics-based interfaces
• Interactor components for common interactions
• Integrated color maps and search functionality
• Event logging to support visualization evaluation

II.7) Algorithms Provided

• General activity scheduling (supporting customized animation)
• Data / visual item separation, caching, and garbage collection (enabling increased flexibility and scalability)
  • Generalized fisheye views of large data sets
  • Supports non-topology-preserving visual transformations of source data
• Layout algorithms
  • Radial tree layout
  • Circular graph layout
  • Force directed layout (prefuse physics simulator and Fruchterman-Reingold)
  • Top-down tree layout (Reingold-Tilford Layout)
  • Balloon tree layout
  • Proprietary layouts (not included in source distribution)
  • TreeBlock (DOITree) layout
  • Hyperbolic Tree layout
• Extensible force-based physics simulation
  • Barnes-Hut N-Body solver
  • Spring forces, drag forces, and more
  • Numerical integration using Euler's and Runge-Kutta methods
• Distortion techniques
  • Graphical fisheye
  • Bifocal distortion
• Panning and Zooming
  • Speed-dependent automatic zooming
• Color maps
• Trie-based indexing and searching of data

II.8) Snapshot of the Interface

Force-directed graph layout	Animated radial graph layout
Speed-dependent automatic zooming	Graphical fisheye view of a graph
TreeMaps	Starfield Display
Data Mountain	Hyperbolic Tree
Degree-of-Interest Tree
Vizster Social Network Browser

II.9) Development Platform

II.10) Supported Operating Systems

Any operating system for which there is a working Java 1.4 virtual machine implementation (includes Windows, Mac OS X, and Linux).

II.11) Software Dependencies/Required Libraries

Java 1.4

II.12) Current License

BSD license (open source, commercial friendly)

II.13) Number of Users/Downloads

II.14) Pros and Cons

• Rich feature set of supported visualization, animation, and interaction techniques
• Modular system supporting technique reuse and composition at a fine-grained level
• Developed using iterative design process, including usability studies with programmers

• Needs more low-level data processing algorithms (focus so far has been on interaction and API design)
• Java2D rendering has performance limits
• Still in alpha stage, needs more bug-testing and quality assurance

II.15) Planned Work

Part III

Please describe your main interest in participating in the workshop

My main interests are to meet other infovis architecture developers and learn from each other's experience and "war stories". My hope is that walking away from the workshop, we might have plans for collaboration, resulting in the construction of a best-of-breed visualization environment that promotes the use, reuse, and expansion of visualization techniques while serving as a solid and flexible platform for future research.