Showing papers on "Evolutionary programming published in 1991"

Meta-evolutionary programming

Abstract: A brief review of efforts is simulated evolution is given. Evolutionary programming is a stochastic optimization technique that is useful for discovering the extrema of a nonlinear function. To implement such a search, several high-level parameters must be chosen, such as the amount of mutational noise, the severity of the mutation noise, and so forth. The authors address incorporating a meta-level evolutionary programming that can simultaneously evolve optimal settings for these parameters while a search for the appropriate extrema is being conducted. The preliminary experiments reported indicate the suitability of such a procedure. Meta-evolutionary programming was able to converge to points on each of two response surfaces that were close to the global optimum. >

A Hierarchical Approach to Learning the Boolean Multiplexer Function

Abstract: This paper describes the recently developed genetic programming paradigm which genetically breeds populations of computer programs to solve problems. In genetic programming, the individuals in the population are hierarchical compositions of functions and arguments. Each of these individual computer programs is evaluated for its fitness in handling the problem environment. The size and shape of the computer program needed to solve the problem is not predetermined by the user, but instead emerges from the simulated evolutionary process driven by fitness. In this paper, the operation of the genetic programming paradigm is illustrated with the problem of learning the Boolean 11-multiplexer function.

The evolution of intelligent decision making in gaming

Abstract: Intelligent decision making requires an ability to predict one's environment and respond in an optimal manner with respect to some underlying purpose. Decision making becomes more difficult when facing an intelligently interactive player who may be always cooperative, neutral, competitive, or change as required by the circumstance. Further difficulties are encountered when the other player's behavior can be determined only through interaction. Such problems can be addressed using a technique that simulates the logic of evolution. An environment is created that presents the prisoner's dilemma. “Organisms” that optimize behavior are evolved over generations. The results indicate that this “evolutionary programming” can be useful in interactive gaming with respect to arbitrary payoff functions.

Evolutionary methods for training neural networks

Abstract: Training neural networks by the implementation of a gradient-based optimization algorithm (e.g., back-propagation) often leads to locally optimal solutions which may be far removed from the global optimum. Evolutionary optimization methods offer a procedure to stochastically search for suitable weights and bias terms given a specific network topology. The topics discussed are evolutionary programming; genetic algorithms; evolutionary function optimization experiments; background to classification problems and experimental results with evolutionary training. >

Minimax design of CMAC encoded neural network controllers using evolutionary programming

Abstract: The authors describe the use of evolutionary programming for computer-aided design and testing of cerebellar model arithmetic computer (CMAC) encoded neural network regulators. The design and testing problem is viewed as a game in that the controller parameters are to be chosen with a minimax criterion, i.e. to minimize the loss associated with their use on the worst possible plant parameters. The technique permits analysis of neural strategies against a set of plants. This gives both the best choice of control parameters and identification of the plant configuration which is most difficult for the best controller to handle. >

Using evolutionary neural networks for arterial waveform discrimination

Abstract: Summary form only given, as follows. The authors discuss evolutionary programming and details experiments using evolutionary neural networks to discriminate between valid and nonphysiologically realizable arterial waveforms as observed during a surgical procedure. Evolutionary programming has been suggested for optimizing the weights and bias terms of neural networks during supervised learning. This technique effectively avoids the tendency to become stalled in locally optimal solutions, as is common with backpropagation and other gradient-based methods. The technique differs from common 'genetic' approaches in that there is no reliance on specific mutation operations, and the representation for the solution need not be a binary string. The use of evolutionary programming reduces the required complexity of the discriminating network, allowing for more robust performance. >

Fusion-technology and the design of evolutionary machines for neural networks

Abstract: An evolutionary machine is a highly modularized genetic algorithm and as such an instrument for accelerating the process of constructing objects whose parameters are hidden in complex search spaces. The construction of neural networks requires knowledge about both ideal architectures (connectivity) and most efficient weight vectors (computational power). The fusion of neural network modeling with evolutionary strategies is therefore a natural step towards artificial neurogenetic modeling. After a brief characterization of fusion-technology,the paper introduces the central features of evolutionary machines in the spirit of J.Holland [1] and I.Rechenberg [2].

Multi-output system identification using evolutionary programming

Abstract: The use of evolutionary programming (EP) in system identification of single-input-multiple-output (SIMO) systems is studied. EP is used to identify parameters of a linear time-invariant system. Specifically, the identification of SIMO systems whose measurements can contain different amounts of noise is examined. A cost function is proposed to take into account disparate noisy observations. >