Two major goals in machine learning are the discovery of complex multidimensional solutions and continual improvement of existing solutions. In this paper, we argue that complexification, i.e. the incremental elaboration of solutions through adding new structure, achieves both these goals. We demonstrate the power of complexification through the NeuroEvolution of Augmenting Topologies (NEAT) method, which evolves increasingly complex neural network architectures. NEAT is applied to an open-ended coevolutionary robot duel domain where robot controllers compete head to head. Because the robot duel domain supports a wide range of sophisticated strategies, and because coevolution benefits from an escalating arms race, it serves as a suitable testbed for observing the effect of evolving increasingly complex controllers. The result is an arms race of increasingly sophisticated strategies. When compared to the evolution of networks with fixed structure, complexifying networks discover significantly more sophisticated strategies. The results suggest that in order to realize the full potential of evolution, and search in general, solutions must be allowed to complexify as well as optimize.

Genetic programming is a technique for the automatic generation of computer programs loosely based on the theory of evolution. It has produced successful solutions to a wide variety of problems and can be effective even in noisy and changing environments. However, genetic programming produces solutions with large amounts of unnecessary code. The amount of unnecessary code increases over time and is not proportional to increases in the quality of the solutions produced. Thus, this additional code seriously hinders the genetic programming processes by requiring extra resources without producing equivalent returns. This dissertation examines the causes of this "code growth." We use three test problems from very different fields of interest to confirm the generality of the results. We tested the destructive hypothesis, that code growth is a protective response to the destructiveness of crossover, as a potential cause of code growth. It is a definite cause, but is not sufficient to explai...

Co-evolution (i.e. the evolution of two or more competing populations with coupled fitness) has several features that may potentially enhance the power of adaptation of artificial evolution. In particular, as discussed by Dawkins and Krebs [3], competing populations may reciprocally drive one another to increasing levels of complexity by producing an evolutionary "arms race". In this paper we will investigate the role of co-evolution in the context of evolutionary robotics. In particular, we will try to understand in what conditions co-evolution can lead to "arms races". Moreover, we will show that in some cases artificial co-evolution has a higher adaptive power than simple evolution. Finally, by analyzing the dynamics of coevolved populations, we will show that in some circumstances well adapted individuals would be better advised to adopt simple but easily modifiable strategies suited for the current competitor strategies rather than incorporate complex and general strategies that m...

s: Jan. 1992 -- Dec. 1994 ffl CTI: Current Technology Index Jan./Feb. 1993 -- Jan./Feb. 1994 ffl DAI: Dissertation Abstracts International: Vol. 53 No. 1 -- Vol. 55 No. 4 (1994) ffl EEA: Electrical & Electronics Abstracts: Jan. 1991 -- Dec. 1994 ffl P: Index to Scientific & Technical Proceedings: Jan. 1986 -- Feb. 1995 (except Nov. 1994) ffl EI A: The Engineering Index Annual: 1987 -- 1992 ffl EI M: The Engineering Index Monthly: Jan. 1993 -- Dec. 1994 The following GA researchers have already kindly supplied their complete autobibliographies and/or proofread references to their papers: Dan Adler, Patrick Argos, Jarmo T. Alander, James E. Baker, Wolfgang Banzhaf, Ralf Bruns, I. L. Bukatova, Thomas Back, Yuval Davidor, Dipankar Dasgupta, Marco Dorigo, Bogdan Filipic, Terence C. Fogarty, David B. Fogel, Toshio Fukuda, Hugo de Garis, Robert C. Glen, David E. Goldberg, Martina Gorges-Schleuter, Jeffrey Horn, Aristides T. Hatjimihail, Mark J. Jakiela, Richard S. Judson, Akihiko Konaga...

In the last few years several researchers have resorted to artificial evolution (e.g. genetic algorithms) and learning techniques (e.g. neural networks) for studying the interaction between learning and evolution. These studies have been conducted for two different purposes: (a) looking at the performance advantages obtained by combining these two adaptive techniques

It is argued that competitive co-evolution is a viable methodology for developing truly autonomous and intelligent machines capable of setting their own goals in order to face new and continuously changing challenges. The paper starts giving an introduction to the dynamics of competitive co-evolutionary systems and reviews their relevance from a computational perspective. The method is then applied to two mobile robots, a predator and a prey, which quickly and autonomously develop efficient chase and evasion strategies. The results are then explained and put in a longterm framework resorting to a visualization of the Red Queen effect on the fitness landscape. Finally, comparative data on different selection criteria are used to indicate that co-evolution does not optimize "intuitive" objective criteria. 1. Competitive Co-Evolution In a competitive co-evolutionary system the survival probability of a species is affected by the behavior of the other species. In the simplest scenario of...

Co-evolution of competitive species provides an interesting testbed to study the role of adaptive behavior because it provides unpredictable and dynamic environments. In this paper we experimentally investigate some arguments for the co-evolution of different adaptive protean behaviors in competing species of predators and preys. Both species are implemented as simulated mobile robots (Kheperas) with infrared proximity sensors, but the predator has an additional vision module whereas the prey has a maximum speed set to twice that of the predator. Different types of variability during life for neurocontrollers with the same architecture and genetic length are compared. It is shown that simple forms of proteanism affect co-evolutionary dynamics and that preys rather exploit noisy controllers to generate random trajectories, whereas predators benefit from directional-change controllers to improve pursuit behavior.

In competitive coevolution, the goal is to establish an "arms race" that will lead to increasingly sophisticated strategies. The existing methods for monitoring progress in coevolution are designed to demonstrate that the arms race indeed occurred. However, two issues remain: (1) How can progress be monitored efficiently so that every generation champion does not need to be compared to every other generation champion? (2) How can a monitoring method determine whether strictly more sophisticated strategies are discovered as the evolution progresses? We introduce a new method for tracking progress, the dominance tournament, which provides an answer to both questions. The dominance tournament shows how different coevolution runs continue to innovate for different periods of time, reveals the precise generation in each run where stagnation occurs, and identifies the best individuals found during the runs. Such differences are difficult to detect using standard techniques but are clearly distinguished in a dominance tournament, which makes this method a highly useful tool in understanding progress in coevolution.

It is well known that individual learning can speed up artificial evolution enormously. However both supervised learning and reinforcement learning require specific learning goals which usually are not available or difficult to find. We introduce a new principle -- homeokinesis -- which is completely unspecific and yet induces specific, seemingly goal--oriented behaviors of an agent in a complex external world. The principle is based on the assumption that the agent is equipped with an adaptive model of its behavior. A learning signal for both the model and the controller is derived from the misfit between the real behavior of the agent in the world and that predicted by the model. If the structural complexity of the model is chosen adequately, this misfit is minimized if the agent exhibits a smooth controlled behavior. The principle is explicated by two examples. We moreover discuss how functional modularization emerges in a natural way in a structured system from a mechanism of competition for the best internal representation.

In competitive coevolution, the goal is to establish an “arms race ” that will lead to increasingly sophisticated strategies. However, in practice, the process often leads to idiosyncrasies rather than continual improvement. Applying the NEAT method for evolving neural networks to a competitive simulated robot duel domain, we will demonstrate that (1) as evolution progresses the networks become more complex, (2) complexification elaborates on existing strategies, and (3) if NEAT is allowed to complexify, it finds dramatically more sophisticated strategies than when it is limited to fixed-topology networks. The results suggest that in order to realize the full potential of competitive coevolution, genomes must be allowed to complexify as well as optimize over the course of evolution. 1