Evolvability of Multi-Agent Systems
Abstract
The goal of this research was to analyze and improve the learning process for multi-agent systems using evolutionary algorithms (EAs). In particular, we wanted to take advantage of our previous work in developing a set of tools for analyzing the evolvability of genetic operators using Price's equation (Price 1970), a theory borrowed from the population genetics community. All of our previous work with Price's equation (Potter, et.al. 2003; Bassett et.al. 2004) has been done on very traditional EA representations, specifically vectors of real values. Part of our research was to focus on how to adapt the use of Price's equation to a representation appropriate for controlling agents in a multi- agent environment. We chose to look at rule-based representations, in part because there are a large number of genetic operators defined for this representation that could be analyzed. Another part of the research was to involve building multi-agent problem domains that could be used for testing our learning algorithms.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 22, 2006
- Accession Number
- ADA444826
Entities
Organizations
- United States Naval Research Laboratory