IAS Fellow's Public Lecture - Desktops to Supercomputers: the wide computational spectrum in simulating emergence
Computer simulation is an invaluable tool in studying emergence, particularly when abstractions of aggregate behaviors arising out of composition of component behaviors are very hard to propose. An important insight Dr Perumalla puts forward is that computer simulation-based emergence studies must be viewed as falling into distinct classes, each class being characterized by its specific combination of scale and accuracy.
The purpose (also known as “use-case”) behind using a particular class of simulation models becomes important to articulate, since the purpose defines both the way in which results from the simulations are to be interpreted, as well as the computational burden that is to be expected from using that class of models. For example, population models intended to serve as reasonable situational surrogates for the masses (e.g., in order to test a detailed model of the effects of an antagonist group leader in an uprising) must be capable of sustaining millions of subject individuals, yet be computationally fast to allow for multi-scenario experimentation; consequently, a high degree of fidelity may not be an appropriate expectation for the masses in such a usage. On the other hand, specialized individual models for very slow hysteresis accumulation and rare events necessarily require large computational time per scenario.
This implies that emergence studies may need computational capacities as widely ranging as from desktops to supercomputers, depending on the use case. The challenge, then, is to either automatically find the right level of fidelity for a specific usage, or be able to sustain as high a fidelity level as possible at any scale that may be presented by the modeler to the simulation system. This is a grand challenge, which perhaps will remain unsolved in the near future. An intermediate step is to become aware of the issues and realize the distinctions so that the computational needs are correctly understood. Dr Perumalla will present the spectrum in terms of the principal computational dimensions of speed, scalability, fidelity, and usability, and provide illustrations of use cases.
This lecture is free and open to all.
Details about Dr Kalyan Perumalla
Directions to Josephine Butler College
Map - Josephine Butler College is denoted as building No: 55
Pay and display car parking is available at the Park & Ride.
Contact firstname.lastname@example.org for more information about this event.