As noted earlier, virtually all specific modeling techniques—from differential equations to agent-based models—are in CAM’s armamentarium. But we approach problems in an integrative fashion, fusing techniques in ways that offer insights unavailable from their isolated application. Among other examples, the following three are representative.
Unified ABM-CFD Models
Traffic modeling and computational fluid dynamics are established fields. But no field combines them. We do. We have pioneered the development of hybrid agent-based traffic and computational fluid dynamics models. In our Los Angeles Plume-Agent Hybrid (see Movies and Models page), state-of-the-art computation fluid dynamics is used to represent the airborne flow of a toxic contaminant in a full 3-dimensional replica of Los Angeles. But, we also model a mobile agent-population reacting to this hazard, and to directives to shelter-in-place, to car-pool, to avoid certain routes, among others. Our current works uses behaviorally rich Agent-Zero type agents subject to network effects and emotions like fear or distrust. Both the physical plume and human behavior affect levels of exposure, and resulting casualties and economic costs.
In collaboration with the Johns Hopkins Department of Civil Engineering, we have parallel agent-based/engineering efforts where realistic agents are subjected to the stress of Earthquakes, Hurricanes (like Sandy), and a wider range of airborne contaminant plumes.
Unified Evolutionary-Epidemiological Models
Likewise, cellular-level viral phylogenetics and population-level infectious disease dynamics are separate fields. In collaborations with Harvard University and the Applied Physics Lab, we have developed entirely new computational methods of phylogenetic forecasting, and unprecedented techniques for linking these micro-projections to viral transmissibility in humans, and in turn to population-level epidemic modeling. This unification also allows us gauge the pandemic threat inherent in potential phylogentic branches before they have actually evolved, and may further position us to divert evolution away from those branches, without exerting selection pressure in equally (or more) dangerous directions.
High-Performance Interactive Computing
User-friendly graphical user interfaces (GUIs) are common. And high performance computing (HPC) is a large field. But there are very few High Performance computer simulations that offer non-modelers a GUI through which they can explore large models, for policy and/or pedagogical purposes. CAM is developing such a GUI for its high resolution US national Scale model, which includes 300 million distinct individuals matching census data, and well as all travel between the the country’s 30,000 zip codes.
From Synapses to Societies
We are extending Agent_Zero in a number of ways. In collaborations with the Institute for Computational Medicine, we are deepening the model neurally, to take fuller account of the circuitry and reward systems implicated in addictive, obesegenic, and other adverse behaviors. At the same time, we are dramatically increasing the scale of our Agent-Zero modeling, by populating our Award winning National Scale US Model (300 million individuals), and our Global Scale Agent Model (6.5 billion agents) with advanced individuals of this class. These tools will offer new leverage on a wide range of challenges form civil violence to the resurgence of polio.