In collaboration with Dr. Michael Agar, principal investigator on NIH grant DA 10736, Redfish developed an agent-based model to show how the “buzz” around a drug generated incidence curves of different shapes. Use of Design of Experiment techniques enabled analysis of multiple runs to show that critical parameters suggest the relevance of consumer models rather than traditional models from epidemiology. This research was presented at the Agents04 conference and is now under review for publication.

An ongoing effort involves an exploration of the model to better visualize the flow of change in agent attitudes that produce the incidence curve. A second effort now in progress focuses on the characteristics of a drug market and how systems of production and distribution emerge. A third effort centers on the possible use of the models as policy and intervention tools.

An innovation in this research is the link between ethnographic and historical research and agent-based models. Translation between thick description of local social practices and abstract clarity of formal models is an enduring issue in social research with a pedigree of several centuries. Particular attention is paid in this project to value added and lost in the translation process.

Screen Shots: Screen 1, Screen 2

Visualization for the The Santa Fe Institute Business Network Value Network Optimization Project. The initiative seeks to develop agent-based models to support collaborative value network research. The developed models, thus far, are agent-based model extensions of John Sterman's BeerGame at MIT.

The models of the project are released as open source and are provided at the project page at SourceForge.

Screen Shots: Screen 1, Screen 2

An agent-based model of wildfire propagation. The model was written to validate methods for causal inference from observational data. This model was written for Frank Wimberly and his collaborators at Carnegie Mellon University. More details of the theoretical basis can be found here.

The model is designed to extend the agent-based modeling for purposes of adaptive resource scheduling which could be used for the "initial attack" and "extended attack" of wildfires. The model can also be used for more long term policy questions such as thinning.

Screen Shots: Screen 1, Screen 2

An application of agent-based simulation to policy appraisal in The Criminal Justice System in England And Wales. The Criminal Justice System spans the work of three central government departments that are responsible for the activity of the: (1) Police Service; (2) Prison Service; (3) Probation Service; (4) Crown Prosecution Service; (5) Criminal Defence Service; and, (6) Court Services. All departments are about to undergo a fundamental government spending review.

The model and visualization focused on interactive effects when one part of the justice system impacts on the workload and costs in other parts. The model was designed primarily so that Parliamentary policy-makers could assess the impact of proposed policy changes on flows of activity through the system, and hence on the workload and costs in each area.

Developed in conjunction with the London School of Economics

Screen Shots: Screen 1 - Screen 2 - Screen 3

Allows for the visualization of a time series of high-dimensional data points. Data dimensions can be mapped to six different display properties, including: X, Y, and Z locations; size, shape, and color. The time series can be run using "VCR" type controls, or through random access.

Developed in conjunction with NTF Group Australia

Screen Shots: Screen 1 - Screen 2

A prototype web application for Nasdaq exploring how stocks cluster in high-dimensional parameter space (ie Market Cap, P/E ratio, Price/Book, etc.)

Developed while at BiosGroup with Rich Harris, Manoj Gambhir and Isaac Saias.

Screen Shot: Screen 1

Developed a prototype web application that allows consumers to evolve the shape of future possible automobile shapes. Three-dimensional auto shapes were bred using genetic algorithms in a peer2peer web application that runs in a web browser and screensaver context. Car designs were traded between users without the need for intermediary servers.

Developed while at BiosGroup.

Screen Shot: Screen 1