Large-scale Agent-Based Models: Perspectives and Requirements

Tuesday, November 4, 2003 - 10:50am - 11:25am
Keller 3-180
Filippo Castiglione (Istituto Applicazioni del Calcolo (IAC) M. Picone)
Agent Based Models (ABM) are the natural extension of the Ising or Cellular Automata-like models which have been used in the past decades to simulate various physical phenomena.

One important characteristic of ABMs, which distinguishes them from the relatively simple Ising-like models, is the complexity of the internal dynamics of each agent together with the asynchrony of the interactions among agents (and between agents and their environment).

The richness of details one can take into account in its ABM, makes such paradigm very powerful, hence appealing for the simulation of complex phenomena where the behavior of the interacting components are not safely reducible to some stylized or simple mechanism.

In my talk, I will first draw the attention to the need of a standardized method to handle the description of the agent (e.g. finite state automata) and the state-determined interaction rules.

Nowadays, the growth of interest in ABMs is strictly related to the availability of powerful computers. The large number of agents and the complexity of the internal representation of the agent in a typical simulation, dictate the use of high-performance computer-science techniques. With this respect, it is important to review some concepts related to code-optimization. Hence, I will discuss some considerations based on the experience maturated by developing and using two different ABMs, one for the simulation of the immune system activity, the second to simulate the dynamics of the price of commodities in a virtual stock market.