Selective Evolutionary Generation Systems: Theory and Applications.

Self-Reproducing Systems;Stochastic Optimization;Evolutionary Computation;Rational Behavior;Complex Systems;Flight Mechanics;Aerospace Engineering;Computer Science;Electrical Engineering;Engineering (General);Engineering;Aerospace Engineering

Menezes, Amor A.Ulsoy, A. Galip ;

University of Michigan

关键词: Self-Reproducing Systems;    Stochastic Optimization;    Evolutionary Computation;    Rational Behavior;    Complex Systems;    Flight Mechanics;    Aerospace Engineering;    Computer Science;    Electrical Engineering;    Engineering (General);    Engineering;    Aerospace Engineering;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/77855/amenezes_1.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

This dissertation is devoted to the problem of behavior design, which is a generalization of the standard global optimization problem: instead of generating the optimizer, the generalization produces, on the space of candidate optimizers, a probability density function referred to as the behavior.The generalization depends on a parameter, the level of selectivity, such that as this parameter tends to infinity, the behavior becomes a delta function at the location of the global optimizer.The motivation for this generalization is that traditional off-line global optimization is non-resilient and non-opportunistic. That is, traditional global optimization is unresponsive to perturbations of the objective function.On-line optimization methods that are more resilient and opportunistic than their off-line counterparts typically consist of the computationally expensive sequential repetition of off-line techniques.A novel approach to inexpensive resilience and opportunism is to utilize the theory of Selective Evolutionary Generation Systems (SEGS), which sequentially and probabilistically selects a candidate optimizer based on the ratio of the fitness values of two candidates and the level of selectivity.Using time-homogeneous, irreducible, ergodic Markov chains to model a sequence of local, and hence inexpensive, dynamic transitions, this dissertation proves that such transitions result in behavior that is called rational; such behavior is desirable because it can lead to both efficient search for an optimizer as well as resilient and opportunistic behavior.The dissertation also identifies system-theoretic properties of the proposed scheme, including equilibria, their stability and their optimality.Moreover, this dissertation demonstrates that the canonical genetic algorithm with fitness proportional selection and the (1+1) evolutionary strategy are particular cases of the scheme.Applications in three areas illustrate the versatility of the SEGS theory: flight mechanics, control of dynamic systems, and artificial intelligence.The dissertation results touch upon several open problems in the fields of artificial life, complex systems, artificial intelligence, and robotics.