Showing papers in "Evolutionary Programming in 1995"

A Survey of Constraint Handling Techniques in Evolutionary Computation Methods.

Abstract: One of the major components of any evolutionary system is the evaluation function. Evaluation functions are used to assign a quality measure for individuals in a population. Whereas evolutionary computation techniques assume the existence of an (e cient) evaluation function for feasible individuals, there is no uniform methodology for handling (i.e., evaluating) unfeasible ones. The simplest approach, incorporated by evolution strategies and a version of evolutionary programming (for numerical optimization problems), is to reject unfeasible solutions. But several other methods for handling unfeasible individuals have emerged recently. This paper reviews such methods (using a domain of nonlinear programming problems) and discusses their merits and drawbacks.

Adapting Crossover in Evolutionary Algorithms.

Abstract: One of the issues in evolutionary algorithms (EAs) is the relative importance of two search operators: mutation and crossover. Genetic algorithms (GAs) and genetic programming (GP) stress the role of crossover, while evolutionary programming (EP) and evolution strategies (ESs) stress the role of mutation. The existence of many different forms of crossover further complicates the issue. Despite theoretical analysis, it appears difficult to decide a priori which form of crossover to use, or even if crossover should be used at all. One possible solution to this difficulty is to have the EA be self-adaptive, i.e., to have the EA dynamically modify which forms of crossover to use and how often to use them, as it solves a problem. This paper describes an adaptive mechanism for controlling the use of crossover in an EA and explores the behavior of this mechanism in a number of different situations. An improvement to the adaptive mechanism is then presented. Surprisingly this improvement can also be used to enhance performance in a non-adaptive EA.

Using Cultural Algorithms for Constraint Handling in GENOCOP

Abstract: This research investigates the extension of a GA-based software package GENOCOP to deal with heavily constrained continuous numerical optimization problems. The presence of large numbers of constraints can produce situation where there are a number of disjoint feasible regions. Such problems are not amenable to solution with classical linear and non-linear programming techniques. GENOCOP handles constraints by dynamically adjusting the domains of its genetic operators to reflect the constraints expressed as a set of inequalities. Here GENOCOP is embedded in a Cultural Algorithm to allow the collection of information concerning the locations of feasible regions. This information is fed back to GENOCOP and used to control the direction of evolution in the population of solutions. This performance of this extended version of GENOCOP is compared with the original relative to a set of optimization problems. For these problems, a significant improvement in the rate of convergence for the extended GENOCOP system over the original was observed. The results suggest the potential benefits of storing global generalizations of individual experience when solving problems with constraints.

An Evolutionary Programming Approach To Self-Adaptation On Finite State Machines

Abstract: Evolutionary programming was first offered as an alternative method for generating artificial intelligence. Experiments were offered in which finite state machines were used to predict time series with respect to an arbitrary payoff function. Mutations were imposed on the evolving machines such that each of the possible modes of variation were given equal probability. The current study investigates the use of self-adaptive methods of evolutionary programming on finite state machines. Each machine incorporates a coding for its structure and an additional set of parameters that determine in part how it will distribute new trials. Two methods for accomplishing this self-adaptation are implemented and tested on two simple prediction problems. The results appear to favor the use of such self-adaptive methods.

Interactive Evolution of Images.

Abstract: Systems of selection and variation by recombination and/or mutation can be used to evolve images for computer graphics and animation. Interactive evolution can be used to direct the development of favorite designs in various application areas. Examples of the application of evolutionary algorithms to two-dimensional (2-D) bitmap images and the methods for three-dimensional (3-D) voxel images are indicated. We show that artificial evolution can serve as a useful tool for achieving flexibility and complexity in image design with only a moderate amount of user-input and detailed knowledge.

Evolving Complex Structures via Cooperative Coevolution.

Abstract: A cooperative coevolutionary approach to learning complex structures is presented which, although preliminary in nature, appears to have a number of advantages over non-coevolutionary approaches. The cooperative coevolutionary approach encourages the parallel evolution of substructures which interact in useful ways to form more complex higher level structures. The architecture is designed to be general enough to permit the inclusion, if appropriate, of a priori knowledge in the form of initial biases towards particular kinds of decompositions. A brief summary of initial results obtained from testing this architecture in several problem domains is presented which shows a significant speedup over more traditional non-coevolutionary approaches.

Evolution Strategies for Mixed-Integer Optimization of Optical Multilayer Systems.

Abstract: An extension of the evolution strategy for mixed-integer optimizationproblems is introduced. The resulting generalized evolution strategy is applied to the problem of optical multilayer coating design and the results are compared with results obtained by standard methods. The generalized evolution strategy as a synthesis method does not require the existence of a starting design, and it competes well with refinement methods for the optimization of starting designs. The results are very encouraging and indicate that this method is a robust and helpful algorithm for optical multilayer design. Furthermore, the generalized evolution strategy is not a tailored heuristic but can be used for arbitrary mixed-integer optimization problems.

Preliminary Investigations Into A Two-Stage Method of Evolutionary Optimization On Constrained Problems

Abstract: A hybrid of evolutionary programming and a deterministic optimization procedure is applied to a series of nonlinear and quadratic optimization problems. The results indicate that the hybrid method can outperform the exclusive use of evolutionary programming when addressing constrained optimization problems with finite penalty functions. Directions for future research are outlined.

Genetic Programming Exploratory Power and the Discovery of Functions.

Abstract: Hierarchical genetic programming (HGP) approaches rely on the discovery, modification, and use of new functions to accelerate evolution. This paper provides a qualitative explanation of the improved behavior of HGP, based on an analysis of the evolution process from the dual perspective of diversity and causality. From a static point of view, the use of an HGP approach enables the manipulation of a population of higher diversity programs. Higher diversity increases the exploratory ability of the genetic search process, as demonstrated by theoretical and experimental fitness distributions and expanded structural complexity of individuals. From a dynamic point of view, an analysis of the causality of the crossover operator suggests that HGP discovers and exploits useful structures in a bottom-up, hierarchical manner. Diversity and causality are complementary, affecting exploration and exploitation in genetic search. Unlike other machine learning techniques that need extra machinery to control the tradeoff between them, HGP automatically trades off exploration and exploitation.

Evolving the Architecture of A Multi- Part Program in Genetic Programming Using Architecture-Altering Operations

Abstract: This paper describes six new architecture-altering operations that provide a way to dynamically determine the architecture of a multipart program during a run of genetic programming. The new operations are patterned after the naturally occurring operations of gene duplication and gene deletion and are motivated by Ohno's provocative book Evolution by Means of Gene Duplication. The new operations are branch duplication, argument duplication, branch creation, argument creation, branch deletion, and argument deletion. These operations dynamically change the architecture of various programs during a run of genetic programming. The new operations can also be interpreted as providing an automated way to specialize and generalize programs. The paper demonstrates that problems can be solved while the architecture is being evolved.

Evolutionary Stable Strategies Are Not Always Stable Under Evolutionary Dynamics

Abstract: Explanations of the observed behaviors of individuals and species often rely on evolutionary stable strategies (ESSs). Essentially, analysis of ESSs requires a description of the differential costs and benefits of alternative behaviors within a population of organisms and then determines which, if any, combinations of behaviors cannot be invaded by alternative strategies. Although such procedures have been used to predict specific naturally evolved behaviors, the assumptions required to generate ESSs (specifically, an infinite population size and no mutation) do not hold under natural conditions. Experiments presented in this paper indicate that under more realistic conditions of finite populations, random payoffs, and random mutation, populations consistently tend to evolve in trajectories that are unrelated to any ESS.

Wavelet-based signal approximation with genetic algorithms

Abstract: In this article, the usability of genetic algorithms for signal approximation is discussed. Due to recent developments in the field of signal approximation by wavelets, this work concentrates on signal approximation by wavelet-like functions. Signals are approximated by a finite linear combination of elementary functions and a genetic algorithm is employed to find the coefficients to such an approximation. The algorithm maintains a population of different approximations, encoded in the form of 'chromosomes'. From this population 'parents' are selected according to their 'fitness', and the 'children' that constitute the next generation are produced from these parents using mutation and crossover operators.Fitness functions employed to evaluate different approximations are the L1, L2, L4, and L∞ norms. Experiments are carried out on several test signals, using Gabor and spline wavelets, both to evaluate the quality of different fitness functions, encoding schemes, and operators, and to assess the usefulness of genetic algorithms in the realm of signal approximation.Although other existing methods are faster while providing comparable approximation quality, the algorithm offers a great deal of flexibility in terms of different elementary functions, fitness criteria, etc.

Racial harmony and Function Optimization in Genetic Algorithms - The Races Genetic Algorithm.

Abstract: This paper introduces a modiied genetic algorithm, RGA, which, through the use of races, maintains a very diverse population. RGA is shown to be particularly good at solving mulitmodal functions and outperforms current methods, both in discovering solutions and maintaining them.

Evolutionary Chromatographic Law Identification by Recurrent Neural Nets.

Abstract: Analytic chromatography is a physical process whose aim is the separation of the components of a chemical mixture, based on their different aanities for some porous medium through which they are percolated. This paper presents an application of evolutionary recurrent neural nets optimization to the identiication of the internal law of chromatography. New mutation operators involving the parameters of a single neuron are introduced. Furthermore, the strategy for using of the diierent kind of mutation takes into account the past history of the neural net at hand. The rst results for one-and two-component mixtures demonstrate the basic feasibility of the recurrent neural net approach. A strategy to improve the robustness of the results is presented .

An Iterated Function systems Approach to Emergence.

Abstract: An approach to action selection in autonomous agents is presented. This approach is motivated by biological examples and the operation of the Random Iteration Algorithm on an Iterated Function System. The approach is illustrated with a simple three-mode behavior that allows a swarm of 200 computational agents to function as a global optimizer on a 30-dimensional multimodal cost function. The behavior of the swarm has functional similarities to the behavior of an evolutionary computation (EC).

PANIC: A parallel evolutionary rule based system

Abstract: PANIC (Parallelism And Neural networks In Classifier systems) is a parallel system to evolve behavioral strategies codified by sets of rules. It integrates several adaptive techniques and computational paradigms, such as genetic algorithms, neural networks, temporal difference methods and classifier systems, to define a powerful and robust learning system. To allocate credit to rules, we propose a new mechanism, QCredit Assignment (QCA), based on the temporal difference method Qlearning. To overcome the sharing rule problem, posed by traditional credit assignment strategies in rule based systems, QCA evaluates a rule depending on the context where it is applied. The mechanism is implemented through a multi-layer, feed-forward neural network. To overcome the heavy computational load of this approach, a decentralized and asynchronous parallel model of the genetic algorithm for a massive parallel architecture has been devised.

An extended evolutionary programming algorithm for VLSI channel routing

Abstract: Channel routing is an important part of circuit layout in VLSI design This paper presents an evolutionary programming-based two-layer channel router (EPCHR) that uses an integer representation Adaptive mutation, solution refinement, and relabeling schemes are proposed The combination of these schemes reduces the number of generations required to find the global optimum solution of a channel routing problem EPCHR is demonstrated on six standard benchmark problems