SDSU Department of Computer Science
  
Home Contact Us Undergraduate Student Info Graduate Student Info Faculty & Staff News & Events
News Calendar Problem of the Fortnight

November 2008 Calendar

CS Masters' Thesis Defense

Title: Agentcat: A Hybrid Approach to ALife Simulations Using the Starcat Framework
Speaker: Zachary Barrow
Date: Thursday, November 20, 2008
Time: 1:00 p.m.
Location: GMCS 307
Thesis advisor: Dr Joseph Lewis

Abstract:
Historically, there have been two paradigms with respect to cognitive modeling in the field of Artificial Intelligence. On one end of the spectrum is symbolic processing where the existence and serial manipulation of symbols is sufficient for cognition. On the other end of the spectrum is the connectionist model, where cognition arises from the parallel activation of many simple entities, the network of which provides cognition through its various activated connections. The Copycat program by Douglas Hofstadter is a hybrid architecture from which cognition arises through the dynamic contextual representations of the current understanding with and satisfaction of a particular problem Copycat is solving. The Starcat framework is a generalization of Copycat which attempts to make it possible to model arbitrary problem domains using similar constructs as found in Copycat. It is on the Starcat framework which Agentcat is based. Agentcat is a predator-prey ALife simulation. In Agentcat, the puzzle to be solved is optimization of the longevity and replication ability of the agents.

The first goal of Agentcat is to assess agent performance changes due to the tuning of Starcat framework parameters. The second goal is to determine if and how emergent behavior of agents as a result of the substrate, Starcat, is realized. The results of the Agentcat program have shown that tuning Starcat parameters does have a significant impact on agent performance. However, the changes in performance due to the changes in configuration are not predictable or entirely understandable. Ultimately, the results show that Agentcat in its current incarnation exhibited small, localized emergent effects, but there were no population-wide emergent effects. Starcat did not induce emergent phenomena in Agentcat agents, but it appears to be supportive of emergent computation when utilized by a suitably receptive design, while at the same time providing a framework on which to deploy applications from arbitrary problem domains.
webmaster@cs.sdsu.edu page counter College of Sciences