基于径向基网络与交互式遗传算法的服装设计

针对将交互式遗传算法应用到服装设计中产生的人的疲劳问题,提出利用神经网络来逼近适应度函数.给出了以GA操作产生的每代最佳个体初步作为神经网络径向基网络函数的中心值并结合相似距离值,利用K-Means求出径向基网络的各参数以逼近适应度函数.在服装设计系统应用中取得了良好的效果.     

基于小生境技术的神经网络进化集成

针对目前神经网络集成方法中生成个体网络差异度小、集成泛化能力较差等缺点,提出一种基于小生境技术的神经网络进化集成方法。利用小生境技术在增加进化群体的多样性、提高进化局部搜索能力方面的良好性能,通过个体间相似程度的共享函数来调整神经网络集成中个体网络的适应度,再依据调整后的新适应度进行选择,以维护群体的多样性,得到多样性的个体网络。理论分析和实验结果表明,该方法能有效生成差异度较大的个体网络,提高神经网络集成系统的泛化能力与计算精度。     

Surrogate model-assisted interactive genetic algorithms with individual’s fuzzy and stochastic fitness

We propose a surrogate model-assisted algorithm by using a directed fuzzy graph to extract a user’s cognition on evaluated individuals in order to alleviate user fatigue in interactive genetic algorithms with an individual’s fuzzy and stochastic fitness. We firstly present an approach to construct a directed fuzzy graph of an evolutionary population according to individuals’ dominance relations, cut-set levels and interval dominance probabilities, and then calculate an individual’s crisp fitness based on the out-degree and in-degree of the fuzzy graph. The approach to obtain training data is achieved using the fuzzy entropy of the evolutionary system to guarantee the credibilities of the samples which are used to train the surrogate model. We adopt a support vector regression machine as the surrogate model and train it using the sampled individuals and their crisp fitness. Then the surrogate model is optimized using the traditional genetic algorithm for some generations, and some good individuals are submitted to the user for the subsequent evolutions so as to guide and accelerate the evolution. Finally, we quantitatively analyze the performance of the presented algorithm in alleviating user fatigue and increasing more opportunities to find the satisfactory individuals, and also apply our algorithm to a fashion evolutionary design system to demonstrate its efficiency.     

Negatively correlated neural networks for classification

This paper presents a new algorithm for designing neural network ensembles for classification problems with noise. The idea behind this new algorithm is to encourage different individual networks in an ensemble to learn different parts or aspects of the training data so that the whole ensemble can learn the whole training data better. Negatively correlated neural networks are trained with a novel correlation penalty term in the error function to encourage such specialization. In our algorithm, individual networks are trained simultaneously rather than independently or sequentially. This provides an opportunity for different networks to interact with each other and to specialize. Experiments on two real-world problems demonstrate that the new algorithm can produce neural network ensembles with good generalization ability. This work was presented, in part, at the Third International Symposium on Artificial Life and Robotics, Oita, Japan January 19–21, 1998     

Autonomous and adaptive procedure for cumulative failure prediction

An autonomous adaptive reliability prediction model using evolutionary connectionist approach based on Recurrent Radial Basis Function architecture is proposed. Based on the currently available failure time data, Fuzzy Min–Max algorithm is used to globally optimize the number of the k Gaussian nodes. This technique allows determining and initializing the k-centers of the neural network architecture in an iterative way. The user does not have to define arbitrary some parameters. The optimized neural network architecture is then iteratively and dynamically reconfigured as new failure occurs. The performance of the proposed approach has been tested using sixteen real-time software failure data.     

Supervised and Reinforcement Evolutionary Learning for Wavelet-based Neuro-fuzzy Networks

This study presents a wavelet-based neuro-fuzzy network (WNFN). The proposed WNFN model combines the traditional Takagi–Sugeno–Kang (TSK) fuzzy model and the wavelet neural networks (WNN). This study adopts the non-orthogonal and compactly supported functions as wavelet neural network bases. A novel supervised evolutionary learning, called WNFN-S, is proposed to tune the adjustable parameters of the WNFN model. The proposed WNFN-S learning scheme is based on dynamic symbiotic evolution (DSE). The proposed DSE uses the sequential-search-based dynamic evolutionary (SSDE) method. In some real-world applications, exact training data may be expensive or even impossible to obtain. To solve this problem, the reinforcement evolutionary learning, called WNFN-R, is proposed. Computer simulations have been conducted to illustrate the performance and applicability of the proposed WNFN-S and WNFN-R learning algorithms.     

进化神经网络在机床工具损耗预测中的应用

设计了一个基于实数编码的改进进化算法优化神经网络的连接权和网络结构.该算法 可以根据种群停止进化代数自适应调节变异率、根据个体适应度调节变异量.加工实验表明采用 进化神经网络可以较准确预测出电火花铣削加工工具损耗,所提出的进化算法是有效的,预测结 果较标准BP神经网络高.该预测模型为电火花铣削加工工具在线自动补偿打下基础.     

一种基于混合粒子群优化算法的深度卷积神经网络架构搜索方法

神经架构搜索（neural architecture search,NAS）技术自动寻找神经网络中各层的最佳组合和连接方式,以及各种超参数的最佳分布。该方法从搜索空间生成若干不同的卷积神经网络（CNN）,使用混合粒子群优化（hybrid particle swarm optimization,HPSO）算法,将一定数目的神经网络个体视做一个群体,将每个网络个体在评价指标下的表现值视做适应度,在给定的世代数范围内,每个神经网络个体都学习自身的历史最佳适应度个体,和整个群体的最佳适应度个体,迭代改善自身的网络架构。实验结果表明,算法运行中出现的最优网络架构,在图像分类任务的多个基准数据集上,与手工设计的神经网络和以遗传算法为基础的NAS算法相比,在网络参数数量和准确率的平衡上取得了有竞争力的结果。     

Genomic mining for complex disease traits with “random chemistry”

Our rapidly growing knowledge regarding genetic variation in the human genome offers great potential for understanding the genetic etiology of disease. This, in turn, could revolutionize detection, treatment, and in some cases prevention of disease. While genes for most of the rare monogenic diseases have already been discovered, most common diseases are complex traits, resulting from multiple gene–gene and gene-environment interactions. Detecting epistatic genetic interactions that predispose for disease is an important, but computationally daunting, task currently facing bioinformaticists. Here, we propose a new evolutionary approach that attempts to hill-climb from large sets of candidate epistatic genetic features to smaller sets, inspired by Kauffman’s “random chemistry” approach to detecting small auto-catalytic sets of molecules from within large sets. Although the algorithm is conceptually straightforward, its success hinges upon the creation of a fitness function able to discriminate large sets that contain subsets of interacting genetic features from those that don’t. Here, we employ an approximate and noisy fitness function based on the ReliefF data mining algorithm. We establish proof-of-concept using synthetic data sets, where individual features have no marginal effects. We show that the resulting algorithm can successfully detect epistatic pairs from up to 1,000 candidate single nucleotide polymorphisms in time that is linear in the size of the initial set, although success rate degrades as heritability declines. Research continues into seeking a more accurate fitness approximator for large sets and other algorithmic improvements that will enable us to extend the approach to larger data sets and to lower heritabilities.     

On Solving the Inverse Scattering Problem with RBF Neural Networks: Noise-Free Case

Neural networks are successfully used to determine small particle properties from knowledge of the scattered light – an inverse light scattering problem. This type of problem is inherently difficult to solve as it is represented by a highly ill-posed function mapping. This paper presents a technique that solves the inverse light scattering problem for spheres using Radial Basis Function (RBF) neural networks. A two-stage network architecture is arranged to enhance network approximation capability. In addition, a new approach to computing basis function parameters with respect to the inverse scattering problem is demonstrated. The technique is evaluated for noise-free data through simulations, in which a minimum 99.06% approximation accuracy is achieved. A comparison is made between the least square and the orthogonal least square training methods.     

基于VISIT算法设计模糊分类器的进化方法

介绍了在没有数据分布先验知识的情况下,用进化方法直接从训练数据中建立紧致模糊分类系统的方法。使用VISIT算法获取每个个体模糊系统,再用遗传算法从中搜索最优的模糊系统。规则和隶属函数是在进化过程中自动建立和优化的。为了同时有效地评价系统的精度和紧致性,用一个模糊专家系统作适应度函数。在2个基准分类问题上的实验结果表明了新方法的有效性。     

Self-training classifier of natural-language texts

A natural-language text classifier is developed using an artificial neural network. A model of the classifier and its implementation are proposed. The classification system consists of two main components, namely, a frequency analyzer and a neural network classifier. Before using the classifier, the user should first prepare a set of training texts and then train the classifier. __________ Translated from Kibernetika i Sistemnyi Analiz, No. 3, pp. 169–176, May–June 2007.     

Electro jet drilling using hybrid NNGA approach

This paper presents a hybrid neural network and genetic algorithm (NNGA) approach for the multi-response optimization of the electro jet drilling (EJD) process. The approach first uses a neural network model to predict the response parameters of the process. A genetic algorithm is then applied to the trained neural network model to obtain the optimal process parameters values in which desirability function approach is used to obtain the fitness function for the genetic algorithm from the network output. The simulated results are found to have a close correlation with the experimental data.     

Particle swarm optimization with genetic recombination: a hybrid evolutionary algorithm

This article presents a hybrid evolutionary algorithm (HEA) based on particle swarm optimization (PSO) and a real-coded genetic algorithm (GA). In the HEA, PSO is used to update the solution, and a genetic recombination operator is added to produce offspring individuals based on the parents, which are selected in proportion to their relative fitness. Through the recombination, new offspring enter the population, and individuals with poor fitness are eliminated. The performance of the proposed hybrid algorithm is compared with those of the original PSO and GA, and the impact of the recombination probability on the performance of the HEA is also analyzed. Various simulations of multivariable functions and neural network optimizations are carried out, showing that the proposed approach gives a superior performance to the canonical means, as well as a good balance between exploration and exploitation.     

Structure minimization using the impact factor in neural networks

Despite many advances, the problem of determining the proper size of a neural network is important, especially for its practical implications in such issues as learning and generalization. Unfortunately, it is not usually obvious which size is best; a system that is too small will not be able to learn the data, while one that is just big enough may learn very slowly and be very sensitive to initial conditions and learning parameters. There are two types of approach to determining the network size: pruning and growing. Pruning consists of training a network which is larger than necessary, and then removing unnecessary weights/nodes. Here, a new pruning method is developed, based on the penalty-term method. This method makes the neural networks good for generalization, and reduces the retraining time needed after pruning weights/nodes. This work was presented, in part, at the 6th International Symposium on Artificial Life and Robotics, Tokyo, Japan, January 15–17, 2001.     

区间适应值交互式遗传算法神经网络代理模型

为了解决交互式遗传算法的用户疲劳问题,提出区间适应值交互式遗传算法神经网络代理模型．首先,对用户已评价个体的基因型及其适应值进行采样以训练神经网络,使其逼近区间适应值的上下限;然后,利用神经网络代理模型,评价后续的部分进化个体,并不断更新训练数据和代理模型,以保证逼近精度;最后,对算法性能进行了定量分析,并将其应用于服装进化设计系统．分析结果表明,所提算法在减轻用户疲劳的前提下,具有更多找到满意解的机会．     

A Simplified Neural Network for Linear Matrix Inequality Problems

A simplified neural network model is proposed to solve a class of linear matrix inequality problems. The stability and solvability of the proposed neural network are analyzed and discussed theoretically. In comparison with the previous neural network models (Lin and Huang, Neural Process Lett 11:153–169, 2000; Lin et al., IEEE Trans Neural Netw 11:1078–1092, 2000), the simplified one is composed of two layers rather than three layers, and the neuron array in each layer is triangular rather than square. The proposed approach can therefore reduce the complexity of the neural network architecture. In addition, the simplified neural network can also be extended to solve multiple linear matrix inequalities with specific constraints, which enlarges the application domain of the proposed approach. Finally, examples are given to illustrate the effectiveness and efficiency of the simplified neural network.     

History mechanism supported differential evolution for chess evaluation function tuning

This paper presents a differential evolution (DE) based approach to chess evaluation function tuning. DE with opposition-based optimization is employed and upgraded with a history mechanism to improve the evaluation of individuals and the tuning process. The general idea is based on individual evaluations according to played games through several generations and different environments. We introduce a new history mechanism which uses an auxiliary population containing good individuals. This new mechanism ensures that good individuals remain within the evolutionary process, even though they died several generations back and later can be brought back into the evolutionary process. In such a manner the evaluation of individuals is improved and consequently the whole tuning process.     

Hoeffding bound based evolutionary algorithm for symbolic regression

In symbolic regression area, it is difficult for evolutionary algorithms to construct a regression model when the number of sample points is very large. Much time will be spent in calculating the fitness of the individuals and in selecting the best individuals within the population. Hoeffding bound is a probability bound for sums of independent random variables. As a statistical result, it can be used to exactly decide how many samples are necessary for choosing i individuals from a population in evolutionary algorithms without calculating the fitness completely. This paper presents a Hoeffding bound based evolutionary algorithm (HEA) for regression or approximation problems when the number of the given learning samples is very large. In HEA, the original fitness function is used in every k generations to update the approximate fitness obtained by Hoeffding bound. The parameter 1?δ is the probability of correctly selecting i best individuals from population P, which can be tuned to avoid an unstable evolution process caused by a large discrepancy between the approximate model and the original fitness function. The major advantage of the proposed HEA algorithm is that it can guarantee that the solution discovered has performance matching what would be discovered with a traditional genetic programming (GP) selection operator with a determinate probability and the running time can be reduced largely. We examine the performance of the proposed algorithm with several regression problems and the results indicate that with the similar accuracy, the HEA algorithm can find the solution more efficiently than tradition EA. It is very useful for regression problems with large number of training samples.     

Adaptive sampling in measuring traffic parameters in a computer network using a fuzzy regulator and a neural network

Adaptive sampling using a neural network and a fuzzy regulator is described as applied to computer network traffics. The objective of this approach is to maximally reduce the amount of data to be processed with preservation of acceptable measurement accuracy. The results of experimental verification of sampling efficiency are also presented that are based on the traffic data archive of a real computer network. __________ Translated from Kibernetika i Sistemnyi Analiz, No. 3, pp. 46–54, May–June 2008.