Search

Abstract Functional networks (FNs) are extensions of neural networks (NNs). Unlike NNs, FNs considers general functional models instead of sigmoid-like models. Additionally, in FNs, there are no weights associated with the links that connect neurons. In this paper, we review the research progress and applications of FNs models in recent years. First, we introduce FNs architecture, three typical functional models and the learning process, and we explain the differences between NNs and FNs. Second, we discuss recent applications of FNs that have been introduced in many fields, such as time series prediction, differential and functional equations, pattern classification, detection and prediction, approximation computation, complex system modeling, computer-aided design (CAD), and linear and nonlinear regression. Finally, we present some remarks on future research directions for FNs. [ABSTRACT FROM AUTHOR]

Abstract: In this paper, a class of generalized bi-directional associative memory (BAM) neural networks with recent history distributed delays and impulses are considered. By using Lyapunov functional method and fixed point theorem, we derive several sufficient conditions for the existence and globally exponential stability of a unique periodic solution of the networks. The results of this paper are new and different from previously known results. [Copyright &y& Elsevier]

Abstract: In this paper, we consider the periodic neural networks with variable coefficients and time-varying delays, and investigate the existence of periodic solution and its global exponential stability. Various sufficient conditions ensuring the existence of periodic and its global exponential stability are given. The results obtained in this paper extend and generalize those given in previous literature. [Copyright &y& Elsevier]

In this paper, we are concerned with the anti-synchronization of master–slave neural networks with time delays. By dividing the whole anti-synchronization process into two procedures: the absolute value of error state e i ( t ) flowing from the initial state to 1, then from 1 to 0, i = 1 , … , n , it gets a new viewpoint and a clear illustration on how controller works to the systems. And combining the H o ¨ lder inequality and other techniques, rigorous analysis gives that, each component of error state e ( t ) ( t  ≥ 0) would flow to 1 in finite time, and continue to flow to 0 in fixed time. A numerical example is presented to illustrate the efficiency and effectiveness of our obtained results. [ABSTRACT FROM AUTHOR]

Artificial intelligence, including neural networks, deep learning and machine learning, has made numerous progress and offered new opportunity for academic research and applications in many fields, especially for business activities and firm development. This paper summarizes different applications of artificial intelligence technologies in several domains of business administration. Finance, retail industry, manufacturing industry, and enterprise management are all included. In spite of all the existing challenges, we conclude that the rapid development of artificial intelligence will show its great impact on more fields. [ABSTRACT FROM AUTHOR]

This paper proposes a novel neural network to treat natural language. Most of the conventional neural networks can only process sentences consisted of a few words, and their applications are very simple such as metaphor understanding. The proposed network can process many complicated sentences and can be used as an associative memory and a question-answering system for factoid questions. The proposed network is composed of three layers and one network: Sentence Layer, Knowledge Layer, Deep Case Layer and Dictionary Network. The input sentences are divided into knowledge units and stored in the Knowledge Layer. The Deep Case Layer plays an important role in processing the knowledge units properly. The Dictionary Network also plays an important role as a knowledge base. We have carried out several experiments and they have shown that the proposed neural network has superior performances as an associative memory and a question-answering system. Especially as a question-answering system, the performance is very close to the elaborated system based on artificial intelligence. [ABSTRACT FROM AUTHOR]

Abstract: In Multi-Task Learning (MTL), when several related tasks are learned at the same time considering one of them as the main task and the others as secondary ones, there is a transfer of positive information that improves the performance of the main one. However, not only does the difficulty of finding the relationship among different tasks pose a problem in real applications, but also knowing the kind of relationship among them. This paper presents a new method to generate artificial hints (subsets from the original data set) that helps the learning of the main task when all of them are learned simultaneously (as in a MTL scheme). Thus, although these hints cannot be strictly considered as secondary tasks, they will act as guides for the main one. The results obtained with toy and real problems show the advantages of the proposed method. In particular, a faster convergence, a very good performance, and a reduction in the likelihood of being trapped in a local minimum are achieved. [Copyright &y& Elsevier]

Abstract: In this paper, we present an abstract argumentation framework for the support of agreement processes in agent societies. It takes into account arguments, attacks among them, and the social context of the agents that put forward arguments. Then, we define the semantics of the framework, providing a mechanism to evaluate arguments in view of other arguments posed in the argumentation process. We also provide a translation of the framework into a neural network that computes the set of acceptable arguments and can be tuned to give more or less importance to argument attacks. Finally, the framework is illustrated with an example in a real domain of a water-rights transfer market. [Copyright &y& Elsevier]

Abstract: The application of neural networks in the data mining has become wider. Although neural networks may have complex structure, long training time, and the representation of results is not comprehensible, neural networks have high acceptance ability for noisy data, high accuracy and are preferable in data mining. On the other hand, It is an open question as to what is the best way to train and extract symbolic rules from trained neural networks in domains like classification. In this paper, we train the neural networks by constructive learning and present the analysis of the convergence rate of the error in a neural network with and without threshold which have been learnt by a constructive method to obtain the simple structure of the network. The response of ANN is acquired but its result is not in understandable form or in a black box form. It is frequently desirable to use the model backwards and identify sets of input variable which results in a desired output value. The large numbers of variables and nonlinear nature of many materials models that can help finding an optimal set of difficult input variables. We will use a genetic algorithm to solve this problem. The method is evaluated on different public-domain data sets with the aim of testing the predictive ability of the method and compared with standard classifiers, results showed comparatively high accuracy. [Copyright &y& Elsevier]

Abstract: In this paper, a new artificial neural network model is proposed for visual object recognition, in which the bottom-up, sensory-driven pathway and top-down, expectation-driven pathway are fused in information processing and their corresponding weights are learned based on the fused neuron activities. During the supervised learning process, the target labels are applied to update the bottom-up synaptic weights of the neural network. Meanwhile, the hypotheses generated by the bottom-up pathway produce expectations on sensory inputs through the top-down pathway. The expectations are constrained by the real data from the sensory inputs, which can be used to update the top-down synaptic weights accordingly. To further improve the visual object recognition performance, the multi-scale histograms of oriented gradients (MS-HOG) method is proposed to extract local features of visual objects from images. Extensive experiments on different image datasets demonstrate the efficiency and robustness of the proposed neural network model with features extracted using the MS-HOG method on visual object recognition compared with other state-of-the-art methods. [Copyright &y& Elsevier]

Abstract: This paper presents the use of an intelligent hybrid stock trading system that integrates neural networks, fuzzy logic, and genetic algorithms techniques to increase the efficiency of stock trading when using a volume adjusted moving average (VAMA), a technical indicator developed from equivolume charting. For this research, a neuro–fuzzy-based genetic algorithm (NF-GA) system utilizing a VAMA membership function is introduced. The results show that the intelligent hybrid system takes advantage of the synergy among these different techniques to intelligently generate more optimal trading decisions for the VAMA, allowing investors to make better stock trading decisions. [Copyright &y& Elsevier]

Abstract: In this paper, we propose a neural network (NN)-based multi-agent classifier system (MACS) using the trust, negotiation, and communication (TNC) reasoning model. The main contribution of this work is that a novel trust measurement method, based on the recognition and rejection rates, is proposed. Besides, an auctioning procedure, based on the sealed bid, first price method, is adapted for the negotiation phase. Two agent teams are formed; each consists of three NN learning agents. The first is a fuzzy min–max (FMM) NN agent team and the second is a fuzzy ARTMAP (FAM) NN agent team. Modifications to the FMM and FAM models are also proposed so that they can be used for trust measurement in the TNC model. To assess the effectiveness of the proposed model and the bond (based on trust), five benchmark data sets are tested. The results compare favorably with those from a number of classification methods published in the literature. [Copyright &y& Elsevier]

Abstract: The eng-genes concept involves the use of fundamental known system functions as activation functions in a neural model to create a ‘grey-box’ neural network. One of the main issues in eng-genes modelling is to produce a parsimonious model given a model construction criterion. The challenges are that (1) the eng-genes model in most cases is a heterogenous network consisting of more than one type of nonlinear basis functions, and each basis function may have different set of parameters to be optimised; (2) the number of hidden nodes has to be chosen based on a model selection criterion. This is a mixed integer hard problem and this paper investigates the use of a forward selection algorithm to optimise both the network structure and the parameters of the system-derived activation functions. Results are included from case studies performed on a simulated continuously stirred tank reactor process, and using actual data from a pH neutralisation plant. The resulting eng-genes networks demonstrate superior simulation performance and transparency over a range of network sizes when compared to conventional neural models. [Copyright &y& Elsevier]

Abstract: This paper presents a novel intelligent multiple-controller framework incorporating a fuzzy-logic-based switching and tuning supervisor along with a generalised learning model (GLM) for an autonomous cruise control application. The proposed methodology combines the benefits of a conventional proportional-integral-derivative (PID) controller, and a PID structure-based (simultaneous) zero and pole placement controller. The switching decision between the two nonlinear fixed structure controllers is made on the basis of the required performance measure using a fuzzy-logic-based supervisor, operating at the highest level of the system. The supervisor is also employed to adaptively tune the parameters of the multiple controllers in order to achieve the desired closed-loop system performance. The intelligent multiple-controller framework is applied to the autonomous cruise control problem in order to maintain a desired vehicle speed by controlling the throttle plate angle in an electronic throttle control (ETC) system. Sample simulation results using a validated nonlinear vehicle model are used to demonstrate the effectiveness of the multiple-controller with respect to adaptively tracking the desired vehicle speed changes and achieving the desired speed of response, whilst penalising excessive control action. [Copyright &y& Elsevier]

Abstract: Recognition of emotional states of users in human–computer interaction (HCI) has been shown to be highly dependent on individual human characteristics and way of behavior. Multimodality is a key issue in achieving more accurate results; however, fusing different modalities is a difficult issue in emotion analysis. Emotion recognition systems are generally either rule-based or extensively trained through emotionally colored HCI data sets. In either case, such systems need to take into account, i.e., adapt their knowledge to, the specific user or context of interaction. Neural networks fit well with the adaptation requirement, by collecting and analyzing data from specific environments. An effective approach is presented in this paper, which uses neural network architectures to both detect the need for adaptation of their knowledge, and adapt it through an efficient adaptation procedure. An experimental study with emotion datasets generate in the framework of the EC IST Humaine Network of Excellence. [Copyright &y& Elsevier]

Abstract: In this paper, we address the problem of neural networks (NNs) stabilization and disturbance rejection for a class of nonlinear switched impulsive systems. An adaptive NN feedback control scheme and an impulsive controller for output tracking error disturbance attenuation of nonlinear switched impulsive systems are given under all admissible switched strategy based on NN. The NN is used to compensate for the nonlinear uncertainties of switched impulsive systems, and the approximation error of NN is introduced to the adaptive law in order to improve the tracking attenuation quality of the switched impulsive systems. Impulsive controller is designed to attenuate effect of switching impulse. Under all admissible switching law, impulsive controller and adaptive NN feedback controller can guarantee asymptotic stability of tracking error and improve disturbance attenuation level of tracking error for the overall nonlinear switched impulsive system. Finally, a numerical example is given to demonstrate the effectiveness of the proposed control and stabilization methods. [Copyright &y& Elsevier]

Abstract: This paper presents a novel quadratic error-counting network for pattern classification. Two computational issues namely, the network learning issue and the classification error-counting issue have been addressed. Essentially, a linear series functional approximation to network structure and a smooth quadratic error-counting cost function were proposed to resolve these two computational issues within a single framework. Our analysis shows that the quadratic error-counting objective can be related to the least-squares-error objective by adjusting the class-specific normalization factors. The binary classification network is subsequently extended to cater for multicategory problems. An extensive empirical evaluation validates the usefulness of proposed method. [Copyright &y& Elsevier]

Abstract: The challenge of predicting future values of a time series covers a variety of disciplines. The fundamental problem of selecting the order and identifying the time varying parameters of an autoregressive moving average model (ARMA) concerns many important fields of interest such as linear prediction, system identification and spectral analysis. Recent research activities in forecasting with artificial neural networks (ANNs) suggest that ANNs can be a promising alternative to the traditional ARMA structure. These linear models and ANNs are often compared with mixed conclusions in terms of the superiority in forecasting performance. This study was designed: (a) to investigate a hybrid methodology that combines ANN and ARMA models; (b) to resolve one of the most important problems in time series using ARMA structure and Box–Jenkins methodology: the identification of the model. In this paper, we present a new procedure to predict time series using paradigms such as: fuzzy systems, neural networks and evolutionary algorithms. Our goal is to obtain an expert system based on paradigms of artificial intelligence, so that the linear model can be identified automatically, without the need of human expert participation. The obtained linear model will be combined with ANN, making up an hybrid system that could outperform the forecasting result. [Copyright &y& Elsevier]

Abstract: In this paper, global exponential stability in Lagrange sense is further studied for various continuous-time delayed recurrent neural network with two different types of activation functions. Based on the parameters of the systems, detailed estimation of global exponential attractive sets and positive invariant sets are presented without any hypothesis on the existence. It is also verified that outside the global exponential attracting set; i.e., within the global attraction domain, there is no equilibrium state, periodic state, almost periodic state, and chaos attractor of the neural network. These theoretical analysis narrows the search field of optimization computation, associative memories, chaos control and synchronization and provide convenience for applications. [Copyright &y& Elsevier]

Abstract: Neural networks are widely used in many applications including astronomical physics, image processing, recognition, robotics and automated target tracking, etc. Their ability to approximate arbitrary functions is the main reason for this popularity. The main result of this paper is a constructive proof of a formula for the upper bound of the approximation error by feedforward neural networks with one hidden layer of sigmoidal units and a linear output. The result can also be used to estimate complexity of the maximum error network. An example to demonstrate the theoretical result is given. [Copyright &y& Elsevier]

Abstract: In this paper, the problem of finite-time boundedness (FTB) for neural networks with parameter uncertainties is studied. We extend the concept of FTB for neural networks. Based on the linear matrix inequality (LMI) technique, a sufficient condition is derived to guarantee FTB for uncertain neural networks. Besides, we also simply give a sufficient condition for certain systems. Finally, illustrative examples are given to demonstrate the validity of the proposed theoretical results. [Copyright &y& Elsevier]

Abstract: Many man-made objects are composed of a number of some simple geometric shapes such as lines, circles, rectangles, etc. Therefore, the detection of rectangular objects is an important issue to some practical applications such as the detection of buildings and vehicles in aerial imagery, the detection of license plates in car images, etc. Several methods have been proposed for solving the problem of the detection of rectangular objects. While some approaches are based on the detecting lines, some approaches are based on the Hough transform. Each approach has its own advantages and disadvantages (e.g., computational load). In this paper, we propose a class of neural networks with a special type of neural junctions for the detection of rectangular objects. The proposed neural networks can be trained in either an unsupervised mode or a batch mode. In contrast to some popular clustering algorithms such as the fuzzy -means algorithm and the fuzzy -rectangular shells algorithm, our approach is not based on minimizing an objective function but based on the idea of competitive learning. Based on the idea of competitive learning, the computational load can be decreased. Several data sets were tested to illustrate the effectiveness of our proposed approach. [Copyright &y& Elsevier]

Abstract: Clustering algorithms have been successfully applied in several disciplines. One of those applications is the initialization of radial basis function (RBF) centers composing a neural network, designed to solve functional approximation problems. The Clustering for Function Approximation (CFA) algorithm was presented as a new clustering technique that provides better results than other clustering algorithms that were traditionally used to initialize RBF centers. Even though CFA improves performance against other clustering algorithms, it has some flaws that can be improved. Within those flaws, it can be mentioned the way the partition of the input data is done, the complex migration process, the algorithm''s speed, the existence of some parameters that have to be set in order to obtain good solutions, and the convergence is not guaranteed. In this paper, it is proposed an improved version of this algorithm that solves the problems that its predecessor have using fuzzy logic successfully. In the experiments section, it will be shown how the new algorithm performs better than its predecessor and how important is to make a correct initialization of the RBF centers to obtain small approximation errors. [Copyright &y& Elsevier]

Abstract: Real-world systems usually involve both continuous and discrete input variables. However, in existing learning algorithms of both neural networks and fuzzy systems, these mixed variables are usually treated as continuous without taking into account the special features of discrete variables. It is inefficient to represent each discrete input variable having only a few fixed values by one input neuron with full connection to the hidden layer. This paper proposes a novel hierarchical hybrid fuzzy neural network to represent systems with mixed input variables. The proposed model consists of two levels: the lower level are fuzzy sub-systems each of which aggregates several discrete input variables into an intermediate variable as its output; the higher level is a neural network whose input variables consist of continuous input variables and intermediate variables. For systems or function approximations with mixed variables, it is shown that the proposed hierarchical hybrid fuzzy neural networks outperform standard neural networks in accuracy with fewer parameters, and both provide greater transparency and preserve the universal approximation property (i.e., they can approximate any function with mixed input variables to any degree of accuracy). [Copyright &y& Elsevier]

Abstract: The classification ability of a single-layer perceptron could be improved by considering some enhanced features. In particular, this form of neural networks is called a functional-link net. In the output neuron''s activation function, such as the sigmoid function, an inner product of a connection weight vector with an input vector is computed. However, since the input features are not independent of each other for the enhanced pattern, an assumption of the additivity is not reasonable. This paper employs a non-additive technique, namely the fuzzy integral, to aggregate performance values for an input pattern by interpreting each of the connection weights as a fuzzy measure of the corresponding feature. A learning algorithm with the genetic algorithm is then designed to automatically find connection weights. The sample data for bankruptcy analysis obtained from Moody''s Industrial Manuals is considered to examine the classification ability of the proposed method. The results demonstrate that the proposed method performs well in comparison with traditional functional-link net and multivariate techniques. [Copyright &y& Elsevier]

Abstract: In this paper, a discrete-time bidirectional associative memory neural networks model is considered. By employing the theory of coincidence degree and using Halanay-type inequality technique we give some sufficient conditions ensuring the existence and globally exponential stability of periodic solutions for the discrete-time bidirectional neural networks. An example with the numerical simulations is provided to show the correctness of our analysis. [Copyright &y& Elsevier]

This paper presents a new algorithm for on-line artificial neural networks (ANN) training. The network topology is a standard multilayer perceptron (MLP) and the training algorithm is based on the theory of variable structure systems (VSS) and sliding mode control (SMC). The main feature of this novel procedure is the adaptability of the gain (learning rate), which is obtained from sliding mode surface so that system stability is guaranteed. [Copyright &y& Elsevier]

Abstract: In this paper, a new subset-based training and pruning (SBTP) algorithm is proposed based on the relationship between node dependence and Jacobian rank deficiency. At each training iteration, the orthogonal factorization with column permutation is applied to the output of the nodes in the same hidden layer to identify the dependent nodes. The output weights of the dependent nodes will be set as zeros, and the output weights of the independent nodes will be recalculated to maintain the original input–output behavior. Then, only the weights of the independent nodes will be trained using the Levenberg–Marquardt (LM) algorithm at this iteration, while keeping the weights of the dependent nodes unchanged. In this way, the computational cost of the LM algorithm will be reduced significantly. After the training process, a unit-based optimal brain surgeon (UB-OBS) pruning method is used to prune the insensitive hidden units to further reduce the size of the neural network, and no retraining is needed. Simulations are presented to demonstrate the effectiveness of the proposed approach. [Copyright &y& Elsevier]

Abstract: In this paper, a periodic Cohen–Grossberg neural networks with varying delay is considered. Under the generalization of dropping the boundedness and differentiability hypotheses for activation functions as well as differentiability of the varying delays with derivative not greater than 1, some novel sufficient conditions are given to guarantee the existence and global exponential stability of periodic solutions. Our results obtained here improve and extend some previously related results. In addition, two examples are also to illustrate the theory. [Copyright &y& Elsevier]

Abstract: In this paper, a genetic tracker (GT) with neural network (NN) (GT-NN) is presented for single and multiple target tracking. The data association problem formulated as an N-dimensional assignment problem is solved using genetic algorithm. The incorporation of a NN into the GT is then proposed to increase its tracking performance. Performance evaluations of the GT, the GT-NN, the probabilistic data association filter, and the joint probabilistic data association filter are presented using simulation studies. Nine different tracking scenarios are considered for this evaluation. It has been observed that the estimation results of the GT-NN are better than those of the GT, the probabilistic data association filter, and the joint probabilistic data association filter. [Copyright &y& Elsevier]

Abstract: This paper presents a new approach to incrementally constructing a neural network that is capable of learning new information without forgetting old knowledge. The proposed neural network, called hyper-spherical ARTMAP network (HS-ARTMAP network), is a synthesis of an RBF-network-like module and an ART-like module. The HS-ARTMAP network is trained via a training algorithm similar to the training algorithm for the fuzzy ARTMAP system. To demonstrate the performance of the proposed HS-ARTMAP network, several pattern recognition and function approximation problems were tested. [Copyright &y& Elsevier]

Abstract: In this paper, we first design a more generalized network model, Improved circular back propagation (CBP), based on the same structure as CBP proposed by Ridella et al. The novelty of Improved circular back propagation neural network (ICBP) lies in: (1) it substitutes the original extra added node with the isotropic quadratic form input in CBP with the one with an anisotropic quadratic form input; (2) particularly, the weights between the extra node and all the hidden nodes are endowed fixed values instead of the original changeable values. As a result, ICBP possesses better generalization and adaptability although it has less adjustable weights compared to CBP. Secondly, we propose a new kernel-based self-organizing maps (KSOM) algorithm using the kernel method. Our main motives of using the kernel method are (1) to induce a class of robust non-Euclidean distance measures for the original input space and establish a new objective functions for self-organizing maps (SOM), and thus make the newly established SOM able to cluster the non-Euclidean structures in data; (2) to enhance robustness of the SOM algorithms to noise and outliers and at the same time still retain computational simplicity. And then, with the combined BP-SOM idea of Weijters, we construct a new integrated network ICBP–KSOM. Our motivation of presenting the integration is to construct a high-performance classifier by utilizing both ICBP''s good generalization and adaptability and KSOM''s higher classification performance and robustness comparing to SOM. Finally, the experimental results on three benchmark data sets show the superiority and effectiveness of our new integration in terms of the t-test. [Copyright &y& Elsevier]

Abstract: The Kernel method is an effective and popular trick in machine learning. In this paper, by introducing it into conventional auto-associative memory models (AMs), we construct a unified framework of kernel auto-associative memory models (KAMs), which makes the existing exponential and polynomial AMs become its special cases. Further, in order to reduce KAM''s connect complexity, inspired by “small-world network” recently described by Watts and Strogatz, we propose another unified framework of small-world structure (SWS) inspired kernel auto-associative memory models (SWSI–KAMs), which, in principle, makes KAMs simpler in structure. Simulation results on the FERET face database show that, the SWSI–KAMs adopting kernels such as Exponential and Hyperbolic tangent kernels have advantages of configuration simplicity while their recognition performance is almost as good as or even better than corresponding KAMs with full connectivity. In the end, the SWSI–KAM adopting Exponential kernel with different connectivities was emphatically investigated for robustness based on those face images which were added random noises and/or partially occluded in a mosaic way, and the experiments demonstrate that the SWSI–KAM with Exponential kernel is more robust in all cases of network connectivity of 20%, 40% and 60% than both PCA and recently proposed (PC)2A algorithms for face recognition. [Copyright &y& Elsevier]

Abstract: Many constructive learning algorithms have been proposed to find an appropriate network structure for a classification problem automatically. Constructive learning algorithms have drawbacks especially when used for complex tasks and modular approaches have been devised to solve these drawbacks. At the same time, parallel training for neural networks with fixed configurations has also been proposed to accelerate the training process. A new approach that combines advantages of constructive learning and parallelism, output partitioning, is presented in this paper. Classification error is used to guide the proposed incremental-partitioning algorithm, which divides the original data set into several smaller sub-data sets with distinct classes. Each sub-data set is then handled in parallel, by a smaller constructively trained sub-network which uses the whole input vector and produces a portion of the final output vector where each class is represented by one unit. Three classification data sets are used to test the validity of this method, and results show that this method reduces the classification test error. [Copyright &y& Elsevier]

Abstract: In this paper, the Cohen–Grossberg neural networks (CGNNs) with variable and unbounded delays are studied. By using the Lipschitzian Hadamard Theorem and a property of homeomorphism mapping, some new sufficient conditions are obtained to ensure the existence, uniqueness and stability of the equilibrium point. The activation functions need only to be partially Lipschitz continuous and monotone nondecreasing, but do not require to be bounded or differentiable. [Copyright &y& Elsevier]

Abstract: While people compare images using semantic concepts, computers compare images using low-level visual features that sometimes have little to do with these semantics. To reduce the gap between the high-level semantics of visual objects and the low-level features extracted from them, in this paper we develop a framework of learning similarity (LS) using neural networks for semantic image classification, where a LS-based k-nearest neighbors (k-NNL) classifier is employed to assign a label to an unknown image according to the majority of k most similar features. Experimental results on an image database show that the k-NNL classifier outperforms the Euclidean distance-based k-NN (k-NNE) classifier and back-propagation network classifiers (BPNC). [Copyright &y& Elsevier]

Abstract: Ensemble algorithms can improve the performance of a given learning algorithm through the combination of multiple base classifiers into an ensemble. In this paper, we attempt to train and combine the base classifiers using an adaptive policy. This policy is learnt through a Q-learning inspired technique. Its effectiveness for an essentially supervised task is demonstrated by experimental results on several UCI benchmark databases. [Copyright &y& Elsevier]

This paper presents a study of the influence of perturbations in the parameters of a functional network. A quantitative measure is introduced, related to the change in the mean squared error when noise is applied to the network parameters. This measure, based on statistical sensitivity, provides a fault tolerance estimate for a functional network and allows the performance degradation of this kind of system to be predicted. It can be used, therefore, to evaluate performance differences between the training process carried out on a computer and its hardware implementation. The experimental results obtained for different functional network architectures and a feedforward multilayer neural network confirm the validity of the proposed model. [Copyright &y& Elsevier]

In the past, neural networks have been viewed as classification and regression systems whose internal representations were extremely difficult to interpret. It is now becoming apparent that algorithms can be designed which extract understandable representations from trained neural networks, enabling them to be used for data mining, i.e. the discovery and explanation of previously unknown relationships present in data. This paper reviews existing algorithms for extracting comprehensible representations from neural networks and describes research to generalize and extend the capabilities of one of these algorithms. The algorithm has been generalized for application to bioinformatics datasets, including the prediction of splice site junctions in Human DNA sequences. Results generated on this datasets are compared with those generated by a conventional data mining technique (C5) and conclusions drawn. [Copyright &y& Elsevier]

This paper presents a novel architecture of neural networks designed for pattern recognition. The concept of induction graphs coupled with a divide-and-conquer strategy defines a neural network induction graph (NNIG). First, the NNIG concept is described and its properties detailed. It is based on a set of several little neural networks, each one discriminating only two classes. The specialization of each neural network simplifies their structure and improves the classification. The principle used to perform the decision of classification on an input pattern is explained. The latter enables to take into account dubious decisions identified by the NNIG. The last section presents experimental results. A significant gain in the global classification rate can be obtained by using an NNIG. The discussion is illustrated by tests on databases from the UCI machine learning database repository. The experimental results show that an NNIG can achieve a better learning, simpler neural networks and an improved performance in classification. A final illustration is presented on a real microscopical imaging problem for the classification of cells in serous cytology. [Copyright &y& Elsevier]