This paper: 1) reviews different combinations between ANN's and evolutionary algorithms (EA's), including using EA's to evolve ANN connection weights, architectures, learning rules, and input features; 2) discusses different search operators which have been used in various EA's; and 3) points out possible future research directions. It is shown, through a considerably large literature review, that combinations between ANN's and EA's can lead to significantly better intelligent systems than relying on ANN's or EA's alone

In this paper we propose and justify a methodology for the development of the control systems, or `cognitive architectures', of autonomous mobile robots. We argue that the design by hand of such control systems becomes prohibitively difficult as complexity increases. We discuss an alternative approach, involving artificial evolution, where the basic building blocks for cognitive architectures are adaptive noise-tolerant dynamical neural networks, rather than programs. These networks may be recurrent, and should operate in real time. Evolution should be incremental, using an extended and modified version of genetic algorithms. We nally propose that, sooner rather than later, visual processing will be required in order for robots to engage in non-trivial navigation behaviours. Time constraints suggest that initial architecture evaluations should be largely done in simulation. The pitfalls of simulations compared with reality are discussed, together with the importance of incorporating noise. To support our claims and proposals, we present results from some preliminary experiments where robots which roam office-like environments are evolved.

Standard methods for inducing both the structure and weight values of recurrent neural networks fit an assumed class of architectures to every task. This simplification is necessary because the interactions between network structure and function are not well understood. Evolutionary computation, which includes genetic algorithms and evolutionary programming, is a population-based search method that has shown promise in such complex tasks. This paper argues that genetic algorithms are inappropriate for network acquisition and describes an evolutionary program, called GNARL, that simultaneously acquires both the structure and weights for recurrent networks. This algorithm's empirical acquisition method allows for the emergence of complex behaviors and topologies that are potentially excluded by the artificial architectural constraints imposed in standard network induction methods. To Appear in: IEEE Transactions on Neural Networks January The Ohio State University January 17, 1996 1 ...

It is appealing to consider hybrids of neural-network learning algorithms with evolutionary search procedures, simply because Nature has so successfully done so. In fact, computational models of learning and evolution offer theoretical biology new tools for addressing questions about Nature that have dogged that field since Darwin [Belew, 1990]. The concern of this paper, however, is strictly artificial: Can hybrids of connectionist learning algorithms and genetic algorithms produce more efficient and effective algorithms than either technique applied in isolation? The paper begins with a survey of recent work (by us and others) that combines Holland's Genetic Algorithm (GA) with connectionist techniques and delineates some of the basic design problems these hybrids share. This analysis suggests the dangers of overly literal representations of the network on the genome (e.g., encoding each weight explicitly). A preliminary set of experiments that use the GA to find unusual but successf...

This paper presents a new evolutionary system, i.e., EPNet, for evolving artificial neural networks (ANNs). The evolutionary algorithm used in EPNet is based on Fogel's evolutionary programming (EP) [1], [2], [3]. Unlike most previous studies on evolving ANNs, this paper puts its emphasis on evolving ANN's behaviours. This is one of the primary reasons why EP is adopted. Five mutation operators proposed in EPNet reflect such an emphasis on evolving behaviours. Close behavioural links between parents and their offspring are maintained by various mutations, such as partial training and node splitting. EPNet evolves ANN's architectures and connection weights (including biases 1 ) simultaneously in order to reduce the noise in fitness evaluation. The parsimony of evolved ANNs is encouraged by preferring node/connection deletion to addition. EPNet has been tested on a number of benchmark problems in machine learning and ANNs, such as the parity problem, the medical diagnosis problems (bre...

Research on potential interactions between connectionist learning systems, i.e., artificial neural networks (ANNs), and evolutionary search procedures, like genetic algorithms (GAs), has attracted a lot of attention recently. Evolutionary ANNs (EANNs) can be considered as the combination of ANNs and evolutionary search procedures. This paper first distinguishes among three kinds of evolution in EANNs, i.e., the evolution of connection weights, of architectures and of learning rules. Then it reviews each kind of evolution in detail and analyses critical issues related to different evolutions. The review shows that although a lot of work has been done on the evolution of connection weights and of architectures, few attempts have been made to understand the evolution of learning rules. Interactions among different evolutions are seldom mentioned in current research. However, the evolution of learning rules and its interactions with other kinds of evolution play a vital role in EANNs. As t...

Evolution and learning are two forms of adaptation that operate on different time scales. Evolution is capable of capturing relatively slow environmental changes that might encompass several generations. Learning, instead, allows an individual to adapt to environmental changes that are unpredictable at the generational level. Moreover, while evolution operates on the genotype, learning affects the phenotype and phenotypic changes cannot directly modify the genotype. Recently, the study of artificial neural networks that are subjected both to an evolutionary and a lifetime learning process received an increasing attention. These studies have been conducted with two different purposes: (a) looking at the advantages, in terms of performance, of combining two different adaptation techniques; (b) understanding the role of the interaction between learning and evolution in natural organisms (for a review see, Nolfi and Floreano, 1999). The general picture that emerge fro

We present results from experiments in which a genetic algorithm (GA) was used to evolve cellular automata (CAs) to perform a particular computational task—one-dimensional density classification. We look in detail at the evolutionary mechanisms producing the GA’s behavior on this task and the impediments faced by the GA. In particular, we identify four “epochs of innovation ” in which new CA strategies for solving the problem are discovered by the GA, describe how these strategies are implemented in CA rule tables, and identify the GA mechanisms underlying their discovery. The epochs are characterized by a breaking of the task’s symmetries on the part of the GA. The symmetry breaking results in a short-term fitness gain but ultimately prevents the discovery of the most highly fit strategies. We discuss the extent to which symmetry breaking and other impediments are general phenomena in any GA search. 1.

Abstract. What makes a problem easy or hard for a genetic algorithm (GA)? This question has become increas-ingly important as people have tried to apply the GA to ever more diverse types of problems. Much previous work on this question has studied the relationship between GA performance and the structure of a given fitness function when it is expressed as a Walsh polynomial. The work of Bethke, Goldberg, and others has produced certain theoretical results about this relationship. In this article we review these theoretical results, and then dis-cuss a number of seemingly anomalous experimental results reported by Tanese concerning the performance of the GA on a subclass of Walsh polynomials, some members of which were expected to be easy for the GA to optimize. Tanese found that the GA was poor at optimizing all functions in this subclass, that a partitioning of a single large population into a number of smaller independent populations seemed to improve performance, and that hillclimbing outperformed both the original and partitioned forms of the GA on these functions. These results seemed to contradict several commonly held expectations about GAs. We begin by reviewing schema processing in GAs. We then give an informal description of how Walsh analysis and Bethke's Walsh-schema transform relate to GA performance, and we discuss the relevance of this analysis for GA applications in optimization and machine learning. We then describe Tanese's surprising results, examine them experimentally and theoretically, and propose and evaluate some explanations. These explanations lead to a more fundamental question about GAs: what are the features of problems that determine the likelihood of suc-cessful GA performance?

Abstract not found