Comparison of Reproduction Schemes in an Artificial
Society for Cooperative Gathering
By: Gusz Eiben, G.S. Nitschke, and
Martin Schut
Date: 15th November 2004
CPM Report No.: CPM-04-143
Presented at the "Engineering with Social Metaphors" day of the AISB Symposium on Socially Inspired Computing, University of Hertfordship, April 2005.
This paper compares reproduction schemes for adaptive behavior in an artificial society, where the collective task of the society is the gathering of resources in an artificial environment. The environment is randomly distributed with varying quantities of different resource types, where different resource types yield different fitness rewards for agents that successfully gather them. Gathering of the more valuable resource types (those yielding higher fitness rewards) requires cooperative behavior of varying degrees (a certain number of agents working collectively). We compared reproduction schemes over three dimensions. The first was a comparison of agents that could reproduce only at the end of their lifetimes (single reproduction at the end of the agent’s lifetime) and agents that could reproduce several times during their lifetime (multiple reproduction during lifetime). The second was a comparison of agents that could reproduce only with agents in adjacent positions and agents that could reproduce with agents at any position in the environment. The third compared different methods for deriving the number of offspring produced and the fitness share given to each offspring, as well as stochastic variants of these methods. Results indicate that the single reproduction at the end of the agent’s lifetime scheme afforded the artificial society a higher level of performance in its collective task, according to the evaluation criterion, comparative to artificial societies utilizing the multiple reproductions during lifetime reproduction scheme.
Presentation: