Is the Genetic Algorithm a Cooperative Learner?
Abstract
This paper begins to explore an analogy between the usual competitive learning metaphor presented in the genetic algorithm (GA) literature and the cooperative learning metaphor discussed by Clearwater, Huberman, and Hogg. In a blackboard cooperative learning paradigm, agents share partial results with one another through a common blackboard. By occasionally accessing the blackboard for a partial solution, an agent can dramatically increase its speed in finding the overall solution to a problem. The study of Clearwater et al. shows that the resulting speed distribution among the agents is lognormal. The GA can also be described in terms of an analogous cooperative learning paradigm. Unlike the blackboard learner, the GA shares information by copying and recombining the solutions of the agents. This method of communication slows down the propagation of useful information to agents. The slower propagation of information is necessary because the GA cannot directly evaluate parts of a solution or partial solutions. The extent to which the GA is cooperative also depends on the choice of heuristics used to modify the canonical GA. The few test cases presented in this paper suggest that the GA may at times yield an approximately lognormal distribution or a mixture of lognormal distributions. While the results look promising, more analysis of the algorithm's overall process is required. (AN)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1995
- Accession Number
- ADA294112
Entities
People
- Helen G. Cobb