Coevolution and Tartarus

Coevolution is the process of mutual adaptation of two populations. When a difficult optimization is performed with evolutionary computation, a population of adaptive test cases can strongly affect the progress of evolution. This study applies coevolution to the Tartarus task, a grid robot test problem. If the coevolving test cases are viewed as a form of parasite, then the question of virulence becomes an important feature of the algorithm. This study compares different types of parasites for the Tartarus problem. The impact of coevolution in this study is at odds with intuition and statistically significant. Analysis of the different types of coevolution suggests that disruptive crossover has a key effect. In the presence of disruptive crossover, coevolution may need to be modified to be effective. Examples of these modifications are presented. The key method of dealing with disruptive crossover is tracking the age of the Tartarus agents. The age of an agent is defined to be the number of selection steps the agent has survived. Using only older agents to drive coevolution of test cases substantially enhances the performance of one of the two type of coevolution studied.

[1]  Vasant Honavar,et al.  On sensor evolution in robotics , 1996 .

[2]  W. Daniel Hillis,et al.  Co-evolving parasites improve simulated evolution as an optimization procedure , 1990 .

[3]  Peter J. Angeline,et al.  Competitive Environments Evolve Better Solutions for Complex Tasks , 1993, ICGA.

[4]  Jordan B. Pollack,et al.  Coevolution of a Backgammon Player , 1996 .

[5]  Jordan B. Pollack,et al.  A Game-Theoretic Approach to the Simple Coevolutionary Algorithm , 2000, PPSN.

[6]  John H. Miller,et al.  The coevolution of automata in the repeated Prisoner's Dilemma , 1996 .

[7]  J. Pollack,et al.  Challenges in coevolutionary learning: arms-race dynamics, open-endedness, and medicocre stable states , 1998 .

[8]  K. De Jong.  Learning with Genetic Algorithms: An Overview , 1988 .

[9]  John H. Holland,et al.  Adaptation in Natural and Artificial Systems: An Introductory Analysis with Applications to Biology, Control, and Artificial Intelligence , 1992 .

[10]  K. D. Jong.  Learning with Genetic Algorithms: An Overview , 2005, Machine Learning.

[11]  James A. Hendler,et al.  Co-evolving Soccer Softbot Team Coordination with Genetic Programming , 1997, RoboCup.

[12]  J. Pollack,et al.  Coevolving High-Level Representations , 1993 .

[13]  Jordan B. Pollack,et al.  Co-Evolution in the Successful Learning of Backgammon Strategy , 1998, Machine Learning.

[14]  Seth Bullock,et al.  Learning lessons from the common cold: How reducing parasite virulence improves coevolutionary optimization , 2002, Proceedings of the 2002 Congress on Evolutionary Computation. CEC'02 (Cat. No.02TH8600).

[15]  Astro Teller,et al.  The evolution of mental models , 1994 .

[16]  David E. Goldberg,et al.  Genetic Algorithms in Search Optimization and Machine Learning , 1988 .

[17]  Daniel Ashlock,et al.  A pure finite state baseline for Tartarus , 2000, Proceedings of the 2000 Congress on Evolutionary Computation. CEC00 (Cat. No.00TH8512).

[18]  L. Tesfatsion,et al.  Preferential partner selection in an evolutionary study of Prisoner's Dilemma. , 1994, Bio Systems.

[19]  J. Pollack,et al.  Coevolutionary dynamics in a minimal substrate , 2001 .

[20]  Richard K. Belew,et al.  New Methods for Competitive Coevolution , 1997, Evolutionary Computation.