Emulation plays an important role in engineering design. However, most emulators such as Gaussian processes (GPs) are exclusively developed for interpolation/regression and their performance significantly deteriorates...
详细信息
Emulation plays an important role in engineering design. However, most emulators such as Gaussian processes (GPs) are exclusively developed for interpolation/regression and their performance significantly deteriorates in extrapolation. To address this shortcoming, we introduce evolutionary Gaussian processes (EGPs) that aim to increase the extrapolation capabilities of GPs. An EGP differs from a GP in that its training involves automatic discovery of some free-form symbolic bases that explain the data reasonably well. In our approach, this automatic discovery is achieved via evolutionary programming (EP) which is integrated with GP modeling via maximum likelihood estimation, bootstrap sampling, and singular value decomposition. As we demonstrate via examples that include a host of analytical functions as well as an engineering problem on materials modeling, EGP can improve the performance of ordinary GPs in terms of not only extrapolation, but also interpolation/regression and numerical stability.
Evoluční techniky jsou neustále se vyvíjející a progresivní část informatiky. Evoluční algoritmy se v praxi používají k řešení mnohých druhů problémů od...
详细信息
Evoluční techniky jsou neustále se vyvíjející a progresivní část informatiky. Evoluční algoritmy se v praxi používají k řešení mnohých druhů problémů od optimalizace až k plánování. Tato práce se zabývá genetickým a kartézským genetickým programováním, které patří mezi nejčastěji používané algoritmy. Cílem práce je implementovat jednotlivé přístupy a vyhodnotit jejich účinnost v úloze symbolické regrese.
In this paper a family of new methods are proposed for constructing hierarchical-interpolative fuzzy rule bases in the frame of a fuzzy rule based supervised machine learning system modeling black box systems defined ...
详细信息
In this paper a family of new methods are proposed for constructing hierarchical-interpolative fuzzy rule bases in the frame of a fuzzy rule based supervised machine learning system modeling black box systems defined by input-output pairs. The resulting hierarchical rule base is constructed by using structure building pure evolutionary and memetic techniques, namely, Genetic and Bacterial programming Algorithms and their memetic variants containing local search steps. Applying hierarchical-interpolative fuzzy rule bases is a rather efficient way of reducing the complexity of knowledge bases, whereas evolutionary methods (including memetic techniques) ensure a relatively fast convergence in the learning process. As it is presented in the paper, by applying a newly proposed representation schema these approaches can be combined to form hierarchical-interpolative machine learning systems.
Diplomová práce se zabývá použitím zvolených evolučních algoritmů k určení a úpravě parametrů neuronové sítě. K úpravě parametrů sítě se zpětný...
详细信息
Diplomová práce se zabývá použitím zvolených evolučních algoritmů k určení a úpravě parametrů neuronové sítě. K úpravě parametrů sítě se zpětným šířením chyby byly použity genetické algoritmy, evoluční strategie a evoluční programování. Součástí práce je program vytvořený v prostředí Matlab, ve kterém byly použité metody testovány na úlohách rozpoznávání vzorů a predikci průběhu funkce. Výsledkem práce jsou grafy průběhu chyby sítě a fitness během úpravy pomocí zvolených algoritmů a průběhů chyby při následném učení.
Obrazové filtry představují část vědní disciplíny zabívající se digitálním zpracováním obrazu. Filtrace obrazu slouží ke zvýraznění určit...
详细信息
Obrazové filtry představují část vědní disciplíny zabívající se digitálním zpracováním obrazu. Filtrace obrazu slouží ke zvýraznění určité informace. Můžeme potlačit šum, vyhladit obraz, zvýraznit kontrast nebo detekovat hrany. Samotný návrh obrazových filtrů představuje časově poměrně náročný proces. Je tedy žádoucí jej v maximální míře zautomatizovat a přenechat činnost předem naprogramovanému systému. Práce se zabývá návrhem komponent pro zmíněný systém. Jedná se o část funkce expertního systému, kterému se na vstupu poskytne množina vstupních komponent, z nichž se za použití principů evolučního programování mohou generovat nové filtry a následně vyhodnocovat jejich spolehlivost pro další využití.
This paper proposes a hybrid multilogistic methodology, named logistic regression using initial and radial basis function (RBF) covariates. The process for obtaining the coefficients is carried out in three steps. Fir...
详细信息
This paper proposes a hybrid multilogistic methodology, named logistic regression using initial and radial basis function (RBF) covariates. The process for obtaining the coefficients is carried out in three steps. First, an evolutionary programming (EP) algorithm is applied, in order to produce an RBF neural network (RBFNN) with a reduced number of RBF transformations and the simplest structure possible. Then, the initial attribute space (or, as commonly known as in logistic regression literature, the covariate space) is transformed by adding the nonlinear transformations of the input variables given by the RBFs of the best individual in the final generation. Finally, a maximum likelihood optimization method determines the coefficients associated with a multilogistic regression model built in this augmented covariate space. In this final step, two different multilogistic regression algorithms are applied: one considers all initial and RBF covariates (multilogistic initial-RBF regression) and the other one incrementally constructs the model and applies cross validation, resulting in an automatic covariate selection [simplelogistic initial-RBF regression (SLIRBF)]. Both methods include a regularization parameter, which has been also optimized. The methodology proposed is tested using 18 benchmark classification problems from well-known machine learning problems and two real agronomical problems. The results are compared with the corresponding multilogistic regression methods applied to the initial covariate space, to the RBFNNs obtained by the EP algorithm, and to other probabilistic classifiers, including different RBFNN design methods [e.g., relaxed variable kernel density estimation, support vector machines, a sparse classifier (sparse multinomial logistic regression)] and a procedure similar to SLIRBF but using product unit basis functions. The SLIRBF models are found to be competitive when compared with the corresponding multilogistic regression methods and the R
[1] Finding a strategy that allows economically efficient drinking water production at minimal environmental cost is often a complex task. A systematic trade-off among the costs and benefits of possible strategies is ...
详细信息
[1] Finding a strategy that allows economically efficient drinking water production at minimal environmental cost is often a complex task. A systematic trade-off among the costs and benefits of possible strategies is required for determining the optimal production configuration. Such a trade-off involves the handling of interdependent and nonlinear relations for drawdown-related objective categories like damage to wetland vegetation, agricultural yield depression, reduction of river base flow rates, and soil subsidence. We developed a method for multiple-objective optimization of drinking water production by combining Busacker and Gowen's [1961] "minimum cost flow'' procedure for optimal use of the transport network with a genetic algorithm (GA) for optimization of other impacts. The performance of the GA was compared with analytically determined solutions of a series of hypothetical case studies. Pareto-optimality and uniqueness of solutions proved to be effective fitness criteria for identifying trade-off curves with the GA.
Cultural Algorithms are computational self-adaptive models which consist of a population and a belief space. The problem-solving experience of individuals selected from the population space by the acceptance function ...
详细信息
Cultural Algorithms are computational self-adaptive models which consist of a population and a belief space. The problem-solving experience of individuals selected from the population space by the acceptance function is generalized and stored in the belief space. This knowledge can then control the evolution of the population component by means of the influence function. Here, we examine the role that different forms of knowledge can play in the self-adaptation process within cultural systems. In particular, we compare various approaches that use normative and situational knowledge in different ways to guide the function optimization process. The results in this study demonstrate that Cultural Algorithms are a naturally useful framework for self-adaptation and that the use of a cultural framework to support self-adaptation in evolutionary programming can produce substantial performance improvements over population-only systems as expressed in terms of (1) systems success ratio, (2) execution CPU time, and (3) convergence (mean best solution) for a given set of 34 function minimization problems. The nature of these improvements and the type of knowledge that is most effective in producing them depend on the problem's functional landscape. In addition, it was found that the same held true for the population-only self-adaptive EP systems. Each level of self-adaptation (component, individual, and population) outperformed the others for problems with particular landscape features.
Conventional training methods for diagonal recurrent neural networks identifier are limited to the first and second derivative methods. In this paper, a novel training algorithm based on evolutionary programming (EP) ...
详细信息
Conventional training methods for diagonal recurrent neural networks identifier are limited to the first and second derivative methods. In this paper, a novel training algorithm based on evolutionary programming (EP) and particle swarm optimization (PSO) for evolutionary diagonal recurrent neural network (EDRNN) is proposed. Mean-while, a new select mode is given for improving the premature convergence for PSO. Compared with conventional methods, EDRNN has prominent advantage in identifying nonlinear dynamic systems because the structure and weight of EDRNN can be evolved simultaneously. Experimental results of identifying the classical nonlinear dynamic systems confirm that EDRNN-based method is a promising tool for identifier.
暂无评论