Search

A Stochastic Fuzzy Neural Network (SFNN) which has filtering effect on noisy input is studied. the structure of the SFNN is mended and the nodes in each layer of the SFNN are discussed. Each layer in the new structure has exact physical meaning. The number of the nodes is decreased, so is the computation amount. In the parameter learning algorithm, if noisy input data is used the LS cost function based method can cause severe biasing effects. This problem can be solved by a novel EIV cost function which contains the error variables. In this paper, the cost function is extended to multi-input single output system, and the error variables are obtained through learning algorithm to avoid repeated measurement. This method was used to train the parameters of the SFNN. The simulation results show the efficiency of this algorithm. [ABSTRACT FROM AUTHOR]

Minor component analysis (MCA) by neural networks approach has attracted many attentions in the field of neural networks. Convergent learning algorithms of MCA neural networks are very important and useful for applications. In this paper, a globally convergent learning algorithm for MCA neural network is reviewed. Rigorous mathematical proof of global convergence is given and exponential convergence rate is obtained. Comparison experiments illustrate that this algorithm has good performance on beamforming problem. [ABSTRACT FROM AUTHOR]

The minor component is the eigenvector associated with the smallest eigenvalue of the covariance matrix of the input data. The minor component analysis (MCA) is a statistical method for extracting the minor component. Many neural networks have been proposed to solve MCA. However, there exists the problem of the divergence of the norm of the weight vector in these neural networks. In this paper, a modification to the well known MCA EXIN algorithm is presented by adjusting the learning rate. The modified MCA EXIN algorithm can guarantee that the norm of the weight vector of the neural network converges to a constant. Mathematical proofs and simulation results are given to show the convergence of the algorithm. [ABSTRACT FROM AUTHOR]

A modified principal component analysis (PCA) neural network (NN) based on signal eigen-analysis is proposed to blind estimation of the pseudo noise (PN) sequence in lower signal to noise ratios (SNR) direct sequence spread spectrum (DS-SS) signals. The received signal is firstly sampled and divided into non-overlapping signal vectors according to a temporal window, which duration is two periods of PN sequence. Then an autocorrelation matrix is computed and accumulated by these signal vectors. The PN sequence can be estimated by the principal eigenvector of autocorrelation matrix in the end. Since the duration of temporal window is two periods of PN sequence, the PN sequence can be reconstructed by the first principal eigenvector only. Additionally, the eigen-analysis method becomes inefficiency when the estimated PN sequence becomes longer. We can use a PCA NN to realize the PN sequence estimation from lower SNR input DS-SS signals effectively. [ABSTRACT FROM AUTHOR]

Principle Component Analysis (PCA) technique is an important and well-developed area of image recognition and to date many linear discrimination methods have been put forward. Despite these efforts, there persist in the traditional PCA some weaknesses. In this paper, we propose a new PCA-based method that can overcome one drawback existed in the traditional PCA method. In face recognition where the training data are labeled, a projection is often required to emphasize the discrimination between the clusters. PCA may fail to accomplish this, no matter how easy the task is, as they are unsupervised techniques. The directions that maximize the scatter of the data might not be as adequate to discriminate between clusters. So we proposed a new PCA-based scheme which can straightforwardly take into consideration data labeling, and makes the performance of recognition system better. Experiment results show our method achieves better performance in comparison with the traditional PCA method. [ABSTRACT FROM AUTHOR]

This paper addresses the problem of blind source separation (BSS) based on nonlinear principal component analysis (NPCA), and presents a new Kalman filtering algorithm, which applies a different state-space representation from the one proposed recently by Lv et al. It is shown that the new Kalman filtering algorithm can be simplified greatly under certain conditions, and it includes the existing Kalman-type NPCA algorithm as a special case. Comparisons are made with several related algorithms and computer simulations on BSS are reported to demonstrate the validity. [ABSTRACT FROM AUTHOR]

This paper devises a novel neural network model applied to finding the principal components of a N-dimensional data stream. This neural network consists of r(≤ N) neurons, where the i-th neuron has only N-i+1 weights and a N-i+1 dimensional input vector that is obtained by the multistage dimension-reduced processing (multistage decomposition) [7] for the input vector sequence and orthogonal to the space spanned by the first i-1 principal components. All the neurons are trained by the conventional Oja's learning algorithms [2] so as to get a series of dimension-reduced principal components in which the dimension number of the i-th principal component is N-i+1. By systematic reconstruction technique, we can recover all the principal components from a series of dimension-reduced ones. We study its global convergence and show its performance via some simulations. Its remarkable advantage is that its computational complexity is reduced and its weight storage is saved. [ABSTRACT FROM AUTHOR]

Locally spatiotemporal salience is defined as the combination of the local contrast salience from multiple paralleling independent spatiotemporal feature channels. The computational model proposed in this paper adopts independent component analysis (ICA) to model the spatiotemporal receptive filed of visual simple cells, then uses the learned independent filters for feature extraction. The ICA-based feature extraction for modelling locally spatiotemporal saliency representation (LSTSR) provides such benefits: (1) valid to use LSTSR directly for locally spatial saliency representation (LSSR) since it includes LSSR as one of its special case; (2) Plausible for space variant sampled dynamic scene; (3) Effective for motion-based scene segmentation. [ABSTRACT FROM AUTHOR]

Based on chaotic communication and independent component analysis, an image protection and authentication technique is proposed. We modulate a watermark by chaotic signal and embed it in watermark host vector which is established in terms of the coefficients in DWT domain. When the author and the legal user want to authenticate digital image, they can detect and extract the watermark by using ICA and private key. Experimental results show that robustness of the watermarking satisfies both of their demands. [ABSTRACT FROM AUTHOR]

In this paper we introduce a scheme to obtain affine invariant descriptors for color images using Independent Component Analysis (ICA), which is a further application of using ICA on contour-known objects. First, some feature points can be found by hue-histogram of the color images in HIS space. Then ICA is applied to extract an invariant descriptor between these corresponding points, which can be a representation of shape similarity between original image and its affine image. This proposed algorithm can also estimate affine motion parameters. Simulation results show that ICA method has better performance compared with Fourier methods. [ABSTRACT FROM AUTHOR]

In this paper, Independent Component Analysis (ICA) is presented as an efficient face feature extraction method. In a task such as face recognition, important information may be contained in the high-order relationship among pixels. ICA is sensitive to high-order statistic in the data and finds not-necessarily orthogonal bases, so it may better identify and reconstruct high-dimensional face image data than Principle Component Analysis (PCA). ICA algorithms are time-consuming and sometimes converge difficultly. A modified FastICA algorithm is developed in this paper, which only need to compute Jacobian Matrix one time in once iteration and achieves the corresponding effect of FastICA. Finally a genetic algorithm is introduced to select optimal independent components (ICs). The experiment results show that modified FastICA algorithm quickens convergence and genetic algorithm optimizes recognition performance. ICA based features extraction method is robust to variations and promising for face recognition. [ABSTRACT FROM AUTHOR]

In recent years, brain-computer interface (BCI) technology has emerged very rapidly. Brain-computer interfaces (BCIs) bring us a new communication interface technology which can translate brain activities into control signals of devices like computers, robots. The preprocessing of electroencephalographic (EEG) signal and translation algorithms play an important role in EEG-based BCIs. In this study, we employed an independent component analysis (ICA)-based preprocessing method and a committee machine-based translation algorithm for the offline analysis of a cursor control experiment. The results show that ICA is an efficient preprocessing method and the committee machine is a good choice for translation algorithm. [ABSTRACT FROM AUTHOR]

A crucial problem for on-line independent component analysis (ICA) algorithm is the choice of step-size, which reflects a tradeoff between steady-state error and convergence speed. This paper proposes a novel ICA algorithm for sub-Gaussian sources, which converges fast while maintaining low steady-state error, since it adopts some techniques, such as the introduction of innovation, usage of skewness information and variable step-size for natural gradient. Simulations have verified these approaches. [ABSTRACT FROM AUTHOR]

Activation function is a crucial factor in independent component analysis (ICA) and the best one is the score function defined on the probability density function (pdf) of the source. However, in FastICA, the activation function has to be selected from several predefined choices according to the prior knowledge of the sources, and the problem of how to select or optimize activation function has not been solved yet. In this paper, self-adaptive FastICA is presented based on the generalized Gaussian model (GGM). By combining the optimization of the GGM parameter and that of the demixing vector, a general framework for self-adaptive FastICA is proposed. Convergence and stability of the proposed algorithm are also addressed. Simulation results show that self-adaptive FastICA is effective in parameter optimization and has better accuracy than traditional FastICA. [ABSTRACT FROM AUTHOR]

The independent component analysis (ICA) problem originates from many practical areas, but there has not been any mathematical theory to solve it completely. In this paper, we establish a mathematical theory to solve it under the condition that the number of super-Gaussian sources is known. According to this theory, a step by step optimization algorithm is proposed and demonstrated well on solving the ICA problem with both the super- and sub-Gaussian sources. [ABSTRACT FROM AUTHOR]

In case of application to high-dimensional pattern recognition task, Independent Component Analysis (ICA) often suffers from two challenging problems. One is the small sample size problem. The other is the choice of basis functions (or independent components). Both problems make ICA classifier unstable and biased. To address the two problems, we propose an enhanced ICA algorithm using a cascaded ensemble learning scheme, named as Random Independent Subspace (RIS). A random resampling technique is used to generate a set of low dimensional feature subspaces in the original feature space and the whiten feature space, respectively. One classifier is constructed in each feature subspace. Then these classifiers are combined into an ensemble classifier using a final decision rule. Extensive experimentations performed on the FERET database suggest that the proposed method can improve the performance of ICA classifier. [ABSTRACT FROM AUTHOR]

In this paper we present a novel GA-ICA method which converges to the optimum. The new method for blindly separating unobservable independent components from their linear mixtures, uses genetic algorithms (GA) to find the separation matrices which minimize a cumulant based contrast function. We focuss our attention on theoretical analysis of convergence including a formal prove on the convergence of the well-known GA-ICA algorithms. In addition we introduce guiding operators, a new concept in the genetic algorithms scenario, which formalize elitist strategies. This approach is very useful in many fields such as biomedical applications i.e. EEG which usually use a high number of input signals. The Guided GA (GGA) presented in this work converges to uniform populations containing just one individual, the optimum. [ABSTRACT FROM AUTHOR]

Fuzzy support vector machine (FSVM) is applied in this paper, in order to resolve problem on bringing different loss for classification error to different fault type in mechanical fault diagnosis. Based on basic principle of FSVM, a method of determining numerical value range of fuzzy coefficient is proposed. Classification performance of FSVM is tested and verified by means of simulation data samples. A fuzzy fault classifier is constructed, and applied to condition monitoring of flue-gas turbine set. The results show that fuzzy coefficient can indicate importance degree of data sample, and classification error rate of important data sample can be decreased. [ABSTRACT FROM AUTHOR]

Traditional web information service can't meet the demand of users getting personalized information timely and properly, which can be think as a kind of passive information organization method. In this paper, an adaptive and active information organization model in complex Internet environment is proposed to provide personalized information service and to automatically retrieve timely, relevant information. An SVM-based Semantic text categorization method is adopted to implement adaptive and active information retrieval. Performance experiment based on a prototype retrieval system manifests the proposed schema is efficient and effective. [ABSTRACT FROM AUTHOR]

In this work, we propose to apply support vector regression (SVR) to build software reliability growth model (SRGM). SRGM is an important aspect in software reliability engineering. Software reliability is the probability that a given software will be functioning without failure during a specified period of time in a specified environment. In order to obtain the better performance of SRGM, practical selection of parameter C for SVR is discussed in the experiments. Experimental results with the classical Sys1 and Sys3 SRGM data set show that the performance of the proposed SVR-based SRGM is better than conventional SRGMs and relative good prediction and generalization ability are achieved. [ABSTRACT FROM AUTHOR]

A controlling chaotic system method based on support vector machines (SVMs) is proposed. The method has been tested for controlling the Hénon map from arbitrary initial states to the desirable stationary point or function input without the need of an analytic model. We can see that its performance is very good in simulation studies. Even if there is additive noise, the proposed method is still effective. [ABSTRACT FROM AUTHOR]

Kernel based nonlinear feature extraction approaches, kernel principal component analysis (KPCA) and kernel independent component analysis (KICA), are used for radar high range resolution profiles (HRRP) feature extraction. The time-shift uncertainty of HRRP is handled by a correlation kernel function, and the kernel basis vectors are chosen via a modified LBG algorithm. The classification performance of support vector machine (SVM) classifier based on KPCA and KICA features for measured data are evaluated, which shows that the KPCA and KICA based feature extraction approaches can achieve better classification performance and are more robust to noise as well, comparing with the adaptive Gaussian classifier (AGC). [ABSTRACT FROM AUTHOR]

To solve the low sampling efficiency problem of negative samples in object detection and information retrieval, a cascaded mixture SVM classifier along with its learning method is proposed in this paper. The classifier is constructed by cascading one-class SVC and two-class SVC. In the learning method, first, 1SVC is trained by using the cluster features of the positive samples, then the 1SVC trained is used to collect the negative samples close to the positive samples and to eliminate the outlier positive samples, finally, the 2SVC is trained by using the positive samples and effective negative samples collected. The cascaded mixture SVM classifier integrates the merits of both 1SVC and 2SVC, and has the characters of higher detection rate and lower false positive rate, and is suitable for object detection and information retrieval. Experimental results show that the cascaded SVM classifier outperforms traditional classifiers. [ABSTRACT FROM AUTHOR]

The VC dimension bound of the set of separating hyperplanes is evaluated by the ratio of squared radius of smallest sphere to squared margin. Choosing some kernel and its parameters means that the radius is fixed. In SVM with hard margin, the ratio is minimized through minimizing squared 2-norm of weight vector. In this paper, a bound for squared radius in the feature space is built, which depends on the scaling factor of RBF kernel and the squared radius bound in the input space. The squared 2-norm of weight vector is described as a quadratic form. Therefore, a simple VC dimension bound with RBF kernel is proposed for classification. Based on minimizing this bound, two constrained nonlinear programming problems are constructed for the linearly and nonlinearly separable cases. Through solving them, we can design the nonlinear classifiers with RBF kernel and determine the scaling factor of RBF kernel simultaneously. [ABSTRACT FROM AUTHOR]

In this paper, a ridgelet kernel regression model is proposed for approximation of multivariate functions, especially those with certain kinds of spatial inhomogeneities. It is based on ridgelet theory, kernel and regularization technology from which we can deduce a regularized kernel regression form. Using the objective function solved by quadratic programming to define a fitness function, we adopt particle swarm optimization algorithm to optimize the directions of ridgelets. Theoretical analysis proves the superiority of ridgelet kernel regression for multivariate functions. Experiments in regression indicate that it not only outperforms support vector machine for a wide range of multivariate functions, but also is robust and quite competitive on training of time. [ABSTRACT FROM AUTHOR]

The min-max modular support vector machine (M3-SVM) was proposed for dealing with large-scale pattern classification problems. M3-SVM divides training data to several sub-sets, and combine them to a series of independent sub-problems, which can be learned in a parallel way. In this paper, we explore the use of the geometric relation among training data in task decomposition. The experimental results show that the proposed task decomposition method leads to faster training and better generalization accuracy than random task decomposition and traditional SVMs. [ABSTRACT FROM AUTHOR]

In order to handle large-scale pattern classification problems, various sequential and parallel classification methods have been developed according to the divide-and-conquer principle. However, existing sequential methods need long training time, and some of parallel methods lead to generalization accuracy decreasing and the number of support vectors increasing. In this paper, we propose a novel hierarchical and parallel method for training support vector machines. The simulation results indicate that our method can not only speed up training but also reduce the number of support vectors while maintaining the generalization accuracy. [ABSTRACT FROM AUTHOR]

This paper proposes a method to improve generalization performance of multi-class support vector machines (SVM) based on directed acyclic graph (DAG). At first the structure of DAG is optimized according to training data and Jaakkola-Haussler bound, and then we define fuzzy membership function for each class which is obtained by using average operator in the testing stage and the final recognition result is the class with maximum membership. As a result of our experiment for similar handwritten Chinese characters recognition, the generalization ability of the novel fuzzy multi-class DAG-based SVM classifier is better than that of pair-wise SVM classifier with other combination strategies and its execution time is almost the same as the original DAG. [ABSTRACT FROM AUTHOR]

In this paper, a novel method called Twi-Map Support Vector Machines (TMSVM) for multi-classification problems is presented. Our ideas are as follows: Firstly, the training data set is mapped into a high-dimensional feature space. Secondly, we calculate the distances between the training data points and hyperplanes. Thirdly, we view the new vector consisting of the distances as new training data point. Finally, we map the new training data points into another high-dimensional feature space with the same kernel function and construct the optimal hyperplanes. In order to examine the training accuracy and the generalization performance of the proposed algorithm, One-against-One algorithm, Fuzzy Least Square Support Vector Machine (FLS-SVM) and the proposed algorithm are applied to five UCI data sets. Comparison results obtained by using three algorithms are given. The results show that the training accuracy and the testing one of the proposed algorithm are higher than those of One-against-One and FLS-SVM. [ABSTRACT FROM AUTHOR]

Multi-category classification is a most interesting problem in the fields of pattern recognition. A one-step method is presented to deal with the multi-category problem. The proposed method converts the problem of classification to the function regression and is applied to solve the converted problem by least squares support vector machines. The novel method classifies the samples in all categories simultaneously only by solving a set of linear equations. Demonstrations of computer experiments are given and good performance is achieved in the simulations. [ABSTRACT FROM AUTHOR]

Support vector machines are primarily designed for binary-class classification. Multicategory classification problems are typically solved by combining several binary machines. In this paper, we propose a novel classifier with only one machine for even multiclass data sets. The proposed method consists of two phases. The first phase builds a trained kernel radius function via the support vector domain decomposition. The second phase constructs a dynamical system corresponding to the trained kernel radius function to decompose data domain and to assign class label to each decomposed domain. Numerical results show that our method is robust and efficient for multicategory classification. [ABSTRACT FROM AUTHOR]

In least squares support vector (LS-SVM), the key challenge lies in the selection of free parameters such as kernel parameters and tradeoff parameter. However, when a large number of free parameters are involved in LS-SVM, the commonly used grid search method for model selection is intractable. In this paper, SLOO-MPS is proposed for tuning multiple parameters for LS-SVM to overcome this problem. This method is based on optimizing the smooth leave- one-out error via a gradient descent algorithm and feasible to compute. Extensive empirical comparisons confirm the feasibility and validation of the SLOO-MPS. [ABSTRACT FROM AUTHOR]

Based on the continuous wavelet transform theory and conditions of the admissible support vector kernel, a novel notion of multidimensional wavelet kernels is proposed for Least Squares Support Vector Machine (LS-WSVM) for pattern recognition and function estimation. Theoretic analysis of the wavelet kernel is discussed in detail. The good approximation property of wavelet kernel function enhances the generalization ability of LS-WSVM method and some experimental results are presented to illustrate the effectiveness and feasibility of the proposed method. [ABSTRACT FROM AUTHOR]

Support Vector Machines are finding application in pattern recognition, regression estimation, and operator inversion. To extend the using range, people have always been trying their best in finding online algorithms. But the Support Vector Machines are sensitive only to the extreme values and not to the distribution of the whole data. Ordinary algorithm can not predict which value will be sensitive and has to deal with all the data once. This paper introduces an algorithm that selects promising vectors from given vectors. Whenever a new vector is added to the training data set, unnecessary vectors are found and deleted. So we could easily get an online algorithm. We give the reason we delete unnecessary vectors, provide the computing method to find them. At last, we provide an example to illustrate the validity of algorithm. [ABSTRACT FROM AUTHOR]

The present study investigates a geometrical method for optimizing the kernel function of a support vector machine. The method is an improvement of the one proposed in [4,5]. It consists of using prior knowledge obtained from conventional SVM training to conformally rescale the initial kernel function, so that the separation between two classes of data is effectively enlarged. It turns out that the new algorithm works efficiently, has few free parameters, consumes very low computational cost, and overcomes the susceptibility of the original method. [ABSTRACT FROM AUTHOR]

This paper presents a method of Sparse Kernel Fisher Discriminant Analysis (SKFDA) through approximating the implicit within-class scatter matrix in feature space. Traditional Kernel Fisher Discriminant Analysis (KFDA) has to use all the training samples to construct the implicit within-class scatter matrix while SKFDA needs only small part of them. Based on this idea, the aim of sparseness can be obtained. Experiments show that SKFDA can dramatically reduce the number of training samples used for constructing the implicit within-class scatter matrix. Numerical simulations on "Banana Shaped" and "Ripley and Ionosphere" data sets confirm that SKFDA has the merit of decreasing the training complexity of KFDA. [ABSTRACT FROM AUTHOR]

Fisher discriminant based Foley-Sammon Transform (FST) has great influence in the area of pattern recognition. On the basis of FST, the Generalized Foley-Sammon Transform (GFST) is presented. The main difference between the GFST and the FST is that the transformed sample set by GFST has the best discriminant ability in global sense while FST has this property only in part sense. Linear discriminants are not always optimal, so a new nonlinear feature extraction method GFST with Kernels (KGFST) based on kernel trick is proposed in this paper. Linear feature extraction in feature space corresponds to non-linear feature extraction in input space. Then, KGFST is proved to correspond to a generalized eigenvalue problem. Lastly, our method is applied to digits and images recognition problems, and the experimental results show that present method is superior to the existing methods in term of space distribution and correct classification rate. [ABSTRACT FROM AUTHOR]

Recent decades viewed increasing interests in the subject of decentralized decision-making. In this paper, three definitions of stochastic Nash equilibrium, which casts different decision criteria, are proposed for a stochastic decentralized decision system. Then the problem of how to find the stochastic Nash equilibrium is converted to an optimization problem. Lastly, a solution method combined with neural network and genetic algorithm is provided. Keywords: decentralized decision system; Nash equilibrium; stochastic programming; neural network; genetic algorithm [ABSTRACT FROM AUTHOR]

Designing optimal measurement infrastructure is a key step for network management. In this work the goal of the optimization is to identify a minimum aggregating nodes set subject to bandwidth and delay constraints on the aggregating procedure. The problem is NP-hard. In this paper, we describe the way of using Hopfield networks for finding aggregating nodes set. The simulation indicates that Hopfield networks can produce much better result than the current method of randomly picking aggregating nodes. [ABSTRACT FROM AUTHOR]

The principle of solving multiobjective optimization problems with fuzzy sets theory is studied. Membership function is the key to introduce the fuzzy sets theory to multiobjective optimization. However, it is difficult to determine membership functions in engineering applications. On the basis of rapid quadratic optimization in the learning of weights, simplification in hardware as well as in computational procedures of functional-link net, discrete membership functions are used as sample training data. When the network converges, the continuous membership functions implemented with the network. Membership functions based on functional-link net have been used in multiobjective optimization. An example is given to illustrate the method. [ABSTRACT FROM AUTHOR]

Resource-constrained project scheduling allocates scarce resources over time to perform a set of activities. Priority rule-based heuristics are the most widely used scheduling methods though their performance depends on the characteristics of the projects. To overcome this deficiency, a feed-forward neural network is designed and integrated into the scheduling scheme so as to automatically select the suitable priority rules for each stage of project scheduling. Testing on Patterson's classic test problems and comparison with other heuristics show that the proposed neural network based heuristic is able to improve the performance of project scheduling. [ABSTRACT FROM AUTHOR]

This paper addresses the problem of arranging jobs to machines in hybrid flow shop in which the setup times are dependent on job sequence. A new heuristic combined artificial neural network approach is proposed. The traditional Hopfield network formulation is modified upon theoretical analysis. Compared with the common used permutation matrix, the new construction needs fewer neurons, which makes it possible to solve large scale problems. The traditional Hopfield network running manner is also modified to make it more competitive with the proposed heuristic algorithm. The performance of the proposed algorithm is verified by randomly generated instances. Computational results of different size of data show that the proposed approach works better when compared to the individual heuristic with random initialization. [ABSTRACT FROM AUTHOR]

A job shop scheduling problem with fuzzy due dates is discussed. The membership function of a fuzzy due date assigned to each job denotes the degree of satisfaction of a decision maker for the completion time of this job. The performance criterion of proposed problem is to maximize the minimum degree of satisfaction over given jobs, and it is an NP-complete problem. Thus artificial neural network is considered to search optimal jobs schedule. [ABSTRACT FROM AUTHOR]

The job shop scheduling with release dates and due dates is considered. An effective algorithm based on GENET neural network model is presented for solving this type of problem. Two heuristics are embedded into the progressive stochastic search scheme in GENET network for the objective of minimizing the weighted number of tardy jobs. Experimental results indicate that the presented algorithm is competent to attain significant reductions in number of tardy jobs in relation to priority heuristic algorithms. [ABSTRACT FROM AUTHOR]

Job-shop scheduling is usually a strongly NP-hard problem of combinatorial optimization problems and is one of the most typical production scheduling problem. It is usually very hard to find its optimal solution. In this paper, a new hybrid approach in dealing with this job-shop scheduling problem based on artificial neural network and genetic algorithm (GA) is presented. The GA is used for optimization of sequence and neural network (NN) is used for optimization of operation start times with a fixed sequence. New type of neurons which can represent processing restrictions and resolve constraint conflict are defined to construct a constraint neural network (CNN). CNN with a gradient search algorithm is applied to the optimization of operation start times with a fixed processing sequence. Computer simulations have shown that the proposed hybrid approach is of high speed and efficiency. [ABSTRACT FROM AUTHOR]

As an extension of the classical job-shop scheduling problem, the flexible job-shop scheduling problem (FJSP) allows an operation to be performed by one machine out of a set of machines. To solve the problem in real job shops, this paper presents a method of the discrete neural network with transient chaos (TDNN). The method considers various constraints in a FJSP. Furthermore, a new computational energy function for FJSP is proposed. A production scheduling program is developed in this research for validation and implementation of the proposed method in practical engineering situations. The experimental results show that the method can converge to the global optimum or near to the global optimum in reasonable and finite time. [ABSTRACT FROM AUTHOR]

A novel sweep-based transiently chaotic neural network (STCNN) algorithm for Capacity Vehicle Routing Problem (CVRP) is proposed. CVRP can be partitioned into two kinds of decisions: the selection of vehicles among the available vehicles and the routing of the selected fleet. The proposed algorithm first generates a number of routes that are served by vehicles and then solves the routes to optimality. Simulations show that the proposed algorithm produces excellent solutions in short computing times. [ABSTRACT FROM AUTHOR]

This paper applies a new version of the transiently chaotic neural network (TCNN), the speedy convergent chaotic neural network (SCCNN), to solve the k-coloring problem, a classic NP-complete graph optimization problem, which has many real-world applications. From analyzing the chaotic states of its computational energy, we reach the conclusion that, like the TCNN, the SCCNN can avoid getting stuck in local minima and thus yield excellent solutions, which overcome the disadvantage of the Hopfield neural network (HNN). In addition, the experimental results verify that the SCCNN converges more quickly than the TCNN does in solving the k-coloring problem, which leads it to be a practical algorithm like the HNN. Therefore, the SCCNN not only adopts the advantages of the HNN as well as the TCNN but also avoids their drawbacks, thus provides an effective and efficient approach to solve the k-coloring problem. [ABSTRACT FROM AUTHOR]

We present a new chaotic neural network whose neuron activation function is hysteretic, called hysteretic transiently chaotic neural network (HTCNN) and with this network, a combinatorial optimization problem is solved. By using hysteretic activation function which is multi-valued, has memory, is adaptive, HTCNN has higher ability of overcoming drawbacks that suffered from the local minimum. We proceed to prove Lyapunov stability for this new model, and then solve a combinatorial optimization problem-Assignment problems. Numerical simulations show that HTCNN has higher ability to search for globally optimal and has higher searching efficiency. [ABSTRACT FROM AUTHOR]

This paper analyzes the behavior of Hopfield networks as a method of solving the travelling salesman problem (TSP) with an enhanced formulation of energy function, which is more available than the Hopfield-Tank (H-T) one. The analysis is based on the geometry of the subspace set up by the degenerate eigenvalues of the connection matrix. A set of criterion for parameter settings is derived. The new parameters performed well in the simulations. [ABSTRACT FROM AUTHOR]