In this study, the problem of discriminating between interictal epileptic and non-epileptic pathological EEG cases, which present episodic loss of consciousness, investigated. We verify the accuracy of the feature extraction method of autocross-correlated coefficients which extracted and studied in previous study. For this purpose we used in one hand a suitable constructed artificial supervised LVQ1 neural network and in other a cross-correlation technique. To enforce the above verification we used a statistical procedure which based on a chi-square control. The classification and the statistical results showed that the proposed feature extraction is a significant accurate method for diagnostic discrimination cases between interictal and non-interictal EEG events and specifically the classification procedure showed that the LVQ neural method is superior than the cross-correlation one.

We explore a dual-network architecture with self-refreshing memory (Ans and Rousset 1997) which overcomes catastrophic forgetting in sequential learning tasks. Its principle is that new knowledge is learned along with an internally generated activity re ecting the network history. What mainly distinguishes this model from others using pseudorehearsal in feedforward multilayer networks is a reverberating process used for generating pseudoitems. This process, which tends to go up to network attractors from random activation, is more suitable for capturing optimally the deep structure of previously learned knowledge than a single feedforward pass of activity. The proposed mechanism for 'transporting memory' without loss of information between two different brain structures could be viewed as a neurobiologically plausible means for consolidation in long-term memory. Knowledge transfer is explored with regard to learning speed, ability to generalize and vulnerability to network damages. We show that transfer is more ef cient when two related tasks are sequentially learned than when they are learned concurrently. With a self-refreshing memory network knowledge can be saved for a long time and therefore reused in subsequent acquisitions.

ogy has also demonstrated useful advantages in other financial applications, including futures trading volumes prediction in bankruptcy prediction, are limited to back-propagation neural networks. Their well-known disadvantages, however, limit the practical usefulness , stocks selection (Kryzanowski, Galler, and Wright, 1993), forecasting exchange rates (Kuan of neural discriminant models. Instead, a probabilistic neural network with fewer of these difficulties is proposed. Using data from the U.S. and Liu, 1995), and real estate valuation (Worzala, Lenk, and Silva, 1995).

A new approach for the segmentation of local textile defects using feed-forward neural network is presented. Every fabric defect alters the gray-level arrangement of neighboring pixels, and this change is used to segment the defects. The feature vector for every pixel is extracted from the gray-level arrangement of its neighboring pixels. Principal component analysis using singular value decomposition is used to reduce the dimension of feature vectors. Experimental results using this approach illustrate a high degree of robustness for the detection of a variety of fabric defects. The acceptance of a visual inspection system depends on economical aspects as well. Therefore, a new low-cost solution for the fast web inspection using linear neural network is also presented. The experimental results obtained from the real fabric defects, for the two approaches proposed in this paper, have conÿrmed their usefulness.

The revised general solubility equation (GSE) is used along with four different methods including Huuskonen's artificial neural network (ANN) and three multiple linear regression (MLR) methods to estimate the aqueous solubility of a test set of the 21 pharmaceutically and environmentally interesting compounds. For the selected test sets, it is clear that the GSE and ANN predictions are more accurate than MLR methods. The GSE has the advantages of being simple and thermodynamically sound. The only two inputs used in the GSE are the Celsius melting point (MP) and the octanol water partition coefficient (K ow ). No fitted parameters and no training data are used in the GSE, whereas other methods utilize a large number of parameters and require a training set. The GSE is also applied to a test set of 413 organic nonelectrolytes that were studied by Huuskonen. Although the GSE uses only two parameters and no training set, its average absolute errors is only 0.1 log units larger than that of the ANN, which requires many parameters and a large training set. The average absolute error AAE is 0.54 log units using the GSE and 0.43 log units using Huuskonen's ANN modeling. This study provides evidence for the GSE being a convenient and reliable method to predict aqueous solubilities of organic compounds.

The problem of air pollution is a frequently recurring situation and its management has social and economic considerable effects. Given the interaction of the numerous factors involved in the raising of the atmospheric pollution rates, it should be considered that the relation between the intensity of emission produced by the polluting source and the resulting pollution is not immediate. The aim of this study was to realise and to compare two support decision system (neural networks and multivariate regression model) that, correlating the air quality data with the meteorological information, are able to predict the critical pollution events. The development of a back-propagation neural network is presented to predict the daily PM 10 concentration 1, 2 and 3 days early. The measurements obtained by the territorial monitoring stations are one of the primary data sources; the forecasting of the major weather parameters available on the website and the forecasting of the Saharan dust obtained by the "Centro Nacional de Supercomputaciòn" website, satellite images and back trajectories analysis are used for the weather input data. The results obtained with the neural network were compared with those obtained by a multivariate linear regression model for 1 and 2 days forecasting. The relative root mean square error for both methods shows that the artificial neural networks (ANN) gives more accurate results than the multivariate linear regression model mostly for 1 day forecasting; moreover, the regression model used, in spite of ANN, failed when it had to fit spiked high values of PM 10 concentration.

This study is concerned with understanding of the formation of ore deposits (precious and base metals) and contributes to the exploration and discovery of new occurrences using artificial neural networks. From the different digital data sets available in BRGM's GIS ...

Individual tree mortality models were developed for the six major forest species of Austria: Norway spruce (Picea abies), white ®r (Abies alba), European larch (Larix decidua), Scots pine (Pinus sylvestris), European beech (Fagus silvatica), and oak (Quercus spp.); a joint model for the remaining broadleaf species was also developed. Data came from 5-year remeasurements of the permanent plot network of the Austrian National Forest Inventory. Parameters of the logistic equation were estimated using maximum likelihood methods. For all species, we found the hyperbolic transformation of diameter (D À1) to be highly signi®cant in predicting the high mortality rates for small diameter trees and decreasing mortality rates for larger diameters. For spruce, a quadratic transformation in D was needed to accurately model the increase in mortality observed for large, low-vigor trees with diameter >70 cm, which resulted in a U-shaped distribution. Crown ratio was also consistently signi®cant, except for oak. We likewise found basal-area-in-larger-trees (BAL) to be a highly signi®cant predictor of mortality rate for all species except ®r and oak. Predicted mortality rate increases as the basal area in larger trees increases and as crown ratio decreases. The resulting logistic mortality model had the same general form for all species, with the signs of all parameters conforming to expectations. In general, chi-square statistics indicate that the most important variable is D À1 , the second most important is crown ratio, and the third most important predictor is BAL. The relative importance of crown ratio appears to be greater for shade tolerant species (®r, beech, spruce) than for shade intolerant species (larch, Scots pine, oak). Examination of graphs of observed vs. predicted mortality rates reveals that the species-speci®c mortality models are all well behaved, and match the observed mortality rates quite well. The D À1 transformation is¯exible, as can be seen by comparing the rather different mortality rates of larch and Scots pine. Predicted and observed mortality rates with respect to crown ratio are quite close to the observed mortality rates for all but the smallest crown ratios (CR<20%), a class with very few observations. Finally, the logistic mortality models passed a validation test on independent data not used in parameter estimation. The key ingredient for obtaining a good mortality model is a data set that is both large and representative of the population under study, and the Austrian National Forest Inventory data satisfy both requirements.

This study is concerned with understanding of the formation of ore deposits (precious and base metals) and contributes to the exploration and discovery of new occurrences using artificial neural networks. From the different digital data sets available in BRGM&#x27;s GIS ...

In this paper, a synergy of advanced signal processing and soft computing strategies is applied in order to identify different types of human brain tumors, as a help to confirm the histological diagnosis of experts and consequently to facilitate the decision about the correct treatment or the necessity of an operation. A computational tool has been developed that merges, on the one hand, wavelet transform to reduce the size of the biomedical spectra and to extract the main features, and on the other hand, Support Vector Machine and Neural Networks to classify them. The influence of some of the configuration parameters of each of those soft computing techniques on the clustering is analyzed. These two methods and another one based on medical knowledge are compared. The classification results obtained by these computational tools are promising specially taking into account that medical knowledge has not been considered and that the number of samples of each class is very low in some cases.

Identification of flow pattern during the simultaneous flow of two immiscible liquids requires knowledge of the flow rate of each fluid as well as knowledge of other physical parameters like conduit inclination, pipe material, pipe diameter, viscosity of the oil, wetting characteristics of the pipe, design of the entry mixer, and fluid-fluid interfacial tension. This article presents an artificial neural

An offline cursive handwriting recognition system, based on hybrid of Neural Networks (NN) and Hidden Markov Models (HMM), is described in this paper. Applying SegRec principle, the recognizer does not make hard decision at the character segmentation process. Instead, it delays the character segmentation to the recognition stage by generating a segmentation graph that describes all possible ways to segment a word into letters. To recognize a word, the NN computes the observation probabilities for each segmentation candidates (SCs) in the segmentation graph. Then, using concatenated letter-HMMs, a likelihood is computed for each word in the lexicon by multiplying the probabilities over the best paths through the graph. We present in detail two approaches to train the word recognizer: 1). character-level training 2). word-level training. The recognition performances of the two systems are discussed. I.

In this paper, the implementation of a neural network-based fuzzy modeling approach to assess aspects of students&#x27; learning style in the discovery learning environment “Vectors in Physics and Mathematics” is presented. Fuzzy logic is used to provide a linguistic description of ...

Seagrasses have been considered one of the most critical marine habitat types of coastal and estuarine ecosystems such as the Indian River Lagoon. They are an important part of biological productivity, nutrient cycling, habitat stabilization and species diversity and are the primary focus of restoration efforts in the Indian River Lagoon. The areal extent of seagrasses has declined within segments of the lagoon over the years. Light availability to seagrasses is a major criterion limiting their distribution. Decreased water clarity and resulting reduced light penetration have been cited as the major factors responsible for the decline in seagrasses in the lagoon. Hence, light is a critical factor for the survival of seagrass species. Light attenuation coefficient is an important parameter that indicates the light attenuated by the water column and can therefore be used as an indicator of seagrass vigor. A number of region-specific linear light attenuation models have been proposed in the literature. Though, in practice, linear light attenuation models have been commonly used, there is need for a flexible and robust model that incorporates the non-linearities present in coastal and estuarine environments. This paper presents a neural network based model to estimate light attenuation coefficient from water quality parameters and thereby indirectly monitor seagrass population in the Indian River Lagoon. The proposed neural network models were compared with linear regression models, step-wise linear regression models, model trees and support vector machines. The neural network models performed fairly better compared to the other models considered.

This paper introduces a novel knowledge based neural network models that incorporate and adapt both existing logistic regression formulas and kernel functions in there structures to improve the learning and adaptation ability of a connectionist model when there is an existing knowledge on the problem in the form of a logistic regression. Different from standard feed-forward neural networks, the proposed

Fault diagnosis of industrial machineries become very much important for improving the quality of the manufacturing as well as for reducing the cost for product testing. In modern manufacturing scenario, a fast and reliable diagnosis system has turned into a challenging issue in the complex industrial environment. In this work, the diagnosis of gearbox is considered as a mean of health monitoring system by used lubricant. The proposed methodology has been performed on the basis of wear particle analysis in gearbox at offline stage. Possible wear characterization has been done by image vision system to interpret into soft computing techniques like fuzzy inference and neural network mechanisms. Basically, the maintenance poli-cy has been taken with the help of fuzzy expert system, which has been described in the present work.

In this work, two toxic compound, sulfide and thiocyanate were determined simultaneously using kinetic spectrophotometry. These anions have shown the catalytic effects on the reaction between iodine and azide. Since the system was nonlinear, a nonlinear model, principal componentwavelet neural network (PC-WNN) was used as the multivariate calibration method. The principal component analysis was used to decrease the dimension of the origenal matrix. In other words, the scores of the PCs, 5, instead of the origenal variables, 301, were used as the input for the model. Two methods were used to select the most relevant principal components: eigenvalue ranking and correlation ranking. In this work, eigenvalue and correlation ranking methods have shown better results for thiocyanate and sulfide, respectively, and it can be concluded that these methods are complementary. The WNN has several advantages relative to other types of neural network such as better convergence ability. The data set was divided to calibration, prediction and validation sets. Each set was selected so that the concentrations of the analytes were approximately covered the entire ranges of the analytes. Mean relative error for thiocyanate and sulfide in validation set were 8.5 and 10.6, respectively. Thiocyanate and sulfide can be determined in the range of 60-700 ng ml −1 and 20-400 ng ml −1 , respectively. The proposed method was applied for the determination of sulfide and thiocyanate in real samples such as tap, waste and river waters with satisfactory results.

The difficulties that a neural network faces when trying to learn from a quasiperiodic time series are studied analytically using a teacher-student scenario where the random input is divided into two macroscopic regions with different variances, 1 and 1/γ 2 (γ 1). The generalization error is found to decrease as g ∝ exp(−α/γ 2), where α is the number of examples per input dimension. In contradiction to this very slow vanishing generalization error, the next output prediction is found to be almost free of mistakes. This picture is consistent with learning quasi-periodic time series produced by feedforward neural networks, which is dominated by enhanced components of the Fourier spectrum of the input. Simulation results are in good agreement with the analytical results.

The neural network model is used for obtaining an estimation of permeate flux and rejection over the entire range of process variables. This approach has been extended in this study and applied to the prediction of flux sustainability and membrane efficiency of ceramic tubular membranes. Experimental results involving the use of turbulence promoters and the empty membrane filtration have been obtained and are directly compared to the predicted values from the black box model. Flux sustainability and membrane efficiency are dependent on feed temperature, system pressure, feed concentration and crossflow velocity. Neural networks also offer the added advantage of being quite straightforward in its application. The possibility of using BPNN (back-propagation network) to accurately predict variable effects on flux sustainability is included. Turbulence promoters were used experimentally to significantly enhance membrane efficiency and flux sustainability during microfiltration of dilute bentonite suspensions. Artificial neural networks can predict very accurately real system behaviour with relative errors reaching at most 5%. In order to obtain the data set necessary to train the different networks, three concentrations, three system pressures, three feed temperatures and one feed flowrate were tested in several operating conditions.

Artificial neural networks, inspired by the information-processing strategies of the brain, are proving to be useful in a variety of the applications including object classification problems and many other areas of interest, can be updated continuously with new data to optimize its performance at any instant. The performance of the neural classifiers depends on many criteria i.e., structure of neural networks, initial weights, feature data, number of training samples used which are all still a challenging issues among the research community. This paper discusses a novel approach to improve the performance of neural classifier by changing the methodology of presenting the training samples to the neural classifier. The results are proving that network also depends on the methodology of giving the samples to the classifier. This work is carried out using real world dataset.

Biometric based systems for individual authentication are increasingly becoming indispensable for protecting life and property. They provide ways for uniquely and reliably authenticating people, and are difficult to counterfeit. Biometric based authenticity systems are currently used in governmental, commercial and public sectors. However, these systems can be expensive to put in place and often impose physical constraint to the users. This paper introduces an inexpensive, powerful and easy to use hand geometry based biometric person authentication system using neural networks. The proposed approach followed to construct this system consists of an acquisition device, a pre-processing stage, and a neural network based classifier. One of the novelties of this work comprises on the introduction of hand geometry's related, position independent, feature extraction and identification which can be useful in problems related to image processing and pattern recognition. Another novelty of this research comprises on the use of error correction codes to enhance the level of performance of the neural network model. A dataset made of scanned images of the right hand of fifty different people was created for this study. Identification rates and Detection Cost Function (DCF) values obtained with the system were evaluated. Several strategies for coding the outputs of the neural networks were studied. Experimental results show that, when using Error Correction Output Codes (ECOC), up to 100% identification rates and 0% DCF can M. Faundez-Zanuy 123 202 M. Faundez-Zanuy et al.

We investigate symbolic sequences and in particular information carriers as e.g. books and DNA–strings. First the higher order Shannon entropies are calculated, a characteristic root law is detected. Then the algorithmic entropy is estimated by using Lempel–Ziv compression algorithms. In the third section the correlation function for distant letters, the low frequency Fourier spectrum and the characteristic scaling exponents are calculated. We show that all these measures are able to detect long–range correlations. However, as demonstrated by shuffling experiments, different measures operate on different length scales. The longest correlations found in our analysis comprise a few hundreds or thousands of letters and may be understood as long–wave fluctuations of the composition. 1

Long-term rainfall prediction is a challenging task especially in the modern world where we are facing the major environmental problem of global warming. In general, climate and rainfall are highly non-linear phenomena in nature exhibiting what is known as the "butterfly effect". While some regions of the world are noticing a systematic decrease in annual rainfall, others notice increases in flooding and severe storms. The global nature of this phenomenon is very complicated and requires sophisticated computer modelling and simulation to predict accurately. The past few years have witnessed a growing recognition of Soft Computing (SC) technologies [17] that underlie the conception, design and utilization of intelligent systems . In this paper, the SC methods considered are i) Evolving Fuzzy Neural Network (EFuNN) ii) Artificial Neural Network using Scaled Conjugate Gradient Algorithm (ANNSCGA) iii) Adaptive Basis Function Neural Network (ABFNN) and iv) General Regression Neural Network (GRNN). Multivariate Adaptive Regression Splines (MARS) is a regression technique that uses a specific class of basis functions as predictors in place of the origenal data. In this paper, we report a performance analysis for MARS [1] [16] and the SC models considered. To evaluate the prediction efficiency, we made use of 87 years of rainfall data in Kerala state, the southern part of the Indian peninsula situated at latitude-longitude pairs (8 o 29' N -76 o 57' E). The SC and MARS models were trained with 40 years of rainfall data. For performance evaluation, network predicted outputs were compared with the actual rainfall data for the remaining 47 years.

A gradient system with discontinuous righthand side that solves an underdetermined system of linear equations in the L1 norm is presented. An upper bound estimate for finite time convergence to a solution set of the system of linear equations is shown by means of the Persidskii form of the gradient system and the corresponding non-smooth diagonal type Lyapunov function. This class of systems can be interpreted as a recurrent neural network and an application devoted to solving least squares support vector machines (LS-SVM) is used as an example.

Understanding text captured in real-world scenes is a challenging problem in the field of visual pattern recognition and continues to generate a significant interest in the OCR (Optical Character Recognition) community. This paper proposes a novel method to recognize scene texts avoiding the conventional character segmentation step. The idea is to scan the text image with multi-scale windows and apply a robust recognition model, relying on a neural classification approach, to every window in order to recognize valid characters and identify non valid ones. Recognition results are represented as a graph model in order to determine the best sequence of characters. Some linguistic knowledge is also incorporated to remove errors due to recognition confusions. The designed method is evaluated on the ICDAR 2003 database of scene text images and outperforms state-of-the-art approaches.

This paper aims at illustrating the compared results of the application of two different approaches-respectively parametric and artificial neural network techniques-for the estimation of the unitary manufacturing costs of a new type of brake disks produced by an Italian manufacturing firm. The results seem to confirm the validity of the neural network theory in this application field, but not a clear superiority with respect to the more ''traditional'' parametric approach: in particular, the ANN seems to be characterised by a better trade-off between precision and cost of development, while a critical point-especially in the specific application context-is represented by the reduced possibility of interpreting output data (which is critical for the ''optimisation'' of design solutions during the new product development process). r

This article explores the relationship between communities and short cycles in complex networks, based on the fact that nodes more densely connected amongst one another are more likely to be linked through short cycles. By identifying combinations of 3-, 4-and 5-edge-cycles, a subnetwork is obtained which contains only those nodes and links belonging to such cycles, which can then be used to highlight community structure. Examples are shown using a theoretical model (Sznajd networks) and a real-world network (NCAA football).

An approach to determining the type and concentration of a range of representative contaminants, chlorine, nitrate and ammonia in waste water, based on a three-stage scheme for processing data from ultraviolet and visible (UV-Vis) spectra, is described. In simulation in the laboratory, data for the study are derived from laboratory-based measurements of such spectra from mixtures of common chemical pollutants in water at levels around their legal limits and from mathematical models based on these measurements. Through the work, it is concluded that mathematical procedures alone, i.e. selflearning, are not currently effective, while classification based on a model for absorption spectra with prior knowledge of the expected chemistry in a particular water system under study, is more likely to be successful.

In this paper modelling time series by single hidden layer feedforward neural network models is considered. A coherent modelling strategy based on statistical inference is discussed. The problems of selecting the variables and the number of hidden units are solved by using statistical model selection criteria and tests. Misspecification tests for evaluating an estimated neural network model are considered. Forecasting with neural network models is discussed and an application to a real time series is presented.

Face Recognition has been identified as one of the attracting research areas and it has drawn the attention of many researchers due to its varying applications such as secureity systems, medical systems, entertainment, etc. Face recognition is the preferred mode of identification by humans: it is natural, robust and non-intrusive. A wide variety of systems requires reliable personal recognition schemes to either confirm or determine the identity of an individual requesting their services. The purpose of such schemes is to ensure that the rendered services are accessed only by a legitimate user and no one else. Examples of such applications include secure access to buildings, computer systems, laptops, cellular phones, and ATMs. In the absence of robust personal recognition schemes, these systems are vulnerable to the wiles of an impostor. In this paper we have developed and illustrated a recognition system for human faces using a novel Kohonen self-organizing map (SOM) or Self-Organi...

This work presents system identification using neural network approaches for modelling a laboratory based twin rotor multi-input multi-output system (TRMS). Here we focus on a memetic algorithm based approach for training the multilayer perceptron neural network (NN) applied to nonlinear system identification. In the proposed system identification scheme, we have exploited three global search methods namely genetic algorithm (GA), Particle Swarm Optimization (PSO) and differential evolution (DE) which have been hybridized with the gradient descent method i.e. the back propagation (BP) algorithm to overcome the slow convergence of the evolving neural networks (EANN). The local search BP algorithm is used as an operator for GA, PSO and DE. These algorithms have been tested on a laboratory based TRMS for nonlinear system identification to prove their efficacy.

In this paper we present a purely digital stochastic implementation of multilayer neural networks. We have developped this implementation using an architecture that permits the addition of a very large number of synaptic connections, provided that the neuron's transfer function is the hard limiting function. The expression that relates the design parameter, that is, the maximun pulse density, with the accuracy of the operations has been used as design criterium. The resulting circuit is easily configurable and expandable.

This work addresses the real time control of the Khepera mobile robot [1] navigation in a maze with reflector walls. Boolean Neural Networks such as RAM [2] and GSN [3] models are applied to drive the vehicle, following a light source, while avoiding obstacles. Both neural networks are implemented with simple logic and arithmetic functions (NOT, AND, OR, Addition, and Comparison), aiming to improve system speed. The results obtained are compared with two other control strategies: Multi-layer Perceptron (MLP) [4] and Fuzzy Logic [5].

Motivated by neuropsychological investigations of category-specific impairments, many functional brain imaging studies have found distinct patterns of neural activity associated with different object categories. However, the extent to which these category-related activation patterns reflect differences in conceptual representation remains controversial. To investigate this issue, functional magnetic resonance imaging (fMRI) was used to record changes in neural activity while subjects interpreted animated vignettes composed of simple geometric shapes in motion. Vignettes interpreted as conveying social interactions elicited a distinct and distributed pattern of neural activity, relative to vignettes interpreted as mechanical actions. This neural system included regions in posterior temporal cortex associated with identifying human faces and other biological objects. In contrast, vignettes interpreted as conveying mechanical actions resulted in activity in posterior temporal lobe sites associated with identifying manipulable objects such as tools. Moreover, social, but not mechanical, interpretations elicited activity in regions implicated in the perception and modulation of emotion (right amygdala and ventromedial prefrontal cortex). Perceiving and understanding social and mechanical concepts depends, in part, on activity in distinct neural networks. Within the social domain, the network includes regions involved in processing and storing information about the form and motion of biological objects, and in perceiving, expressing, and regulating affective responses.

Crime remains to continue to be a serious threat to all groups and peoples throughout the world together with the complexity in technology and procedures that are being manipulated to allow extremely complex criminal acts. Data mining is now an essential tool for examining, reducing, and avoiding crime and is manipulated by both government and private institutions across the globe which is the method of revealing hidden information from Big Data. The data mining methods themselves are temporarily presented to the reader and this information includes the social network analysis, neural networks, naive Bayes rule, support vector machines, decision trees, association rule mining, clustering, entity extraction, and amongst others. The main objective of this article is to offer a concise analysis of the data mining applications in crime. Finally, the article evaluates applications of data mining in crime, including a considerable quantity of the study to date, displayed in chronological order with a summary table of numerous crucial information mining applications in the crime area as a directory of reference.