Showing papers on "Hybrid neural network published in 1999"

Hybrid neural plausibility networks for news agents

Abstract: This paper describes a learning news agent HyNeT which uses hybrid neural network techniques for classifying news titles as they appear on an internet newswire. Recurrent plausibility networks with local memory are developed and examined for learning robust text routing. HyNeT is described for the first time in this paper. We show that a careful hybrid integration of techniques from neural network architectures, learning and information retrieval can reach consistent recall and precision rates of more than 92% on an 82 000 word corpus; this is demonstrated for 10000 unknown news titles from the Reuters newswire. This new synthesis of neural networks, learning and information retrieval techniques allows us to scale up to a real-world task and demonstrates a lot of potential for hybrid plausibility networks for semantic text routing agents on the internet.

Static, Dynamic, and Hybrid Neural Networks in Forecasting Inflation

Abstract: The back-propagation neural network (BPN) model has been the most popular form of artificial neural network model used for forecasting, particularly in economics and finance. It is a static (feed-forward) model which has a learning process in both hidden and output layers. In this paper we compare the performance of the BPN model with that of two other neural network models, viz., the radial basis function network (RBFN) model and the recurrent neural network (RNN) model, in the context of forecasting inflation. The RBFN model is a hybrid model with a learning process that is much faster than the BPN model and that is able to generate almost the same results as the BPN model. The RNN model is a dynamic model which allows feedback from other layers to the input layer, enabling it to capture the dynamic behavior of the series. The results of the ANN models are also compared with those of the econometric time series models.

Gas analysis system composed of a solid-state sensor array and hybrid neural network structure

Abstract: This paper presents the application of the hybrid neural network to the solution of the calibration problem of the solid state sensor array used for the gas analysis. The applied neural network is composed of two parts: the selforganizing Kohonen layer and multilayer perceptron (MLP). The role of the Kohonen layer is to perform the feature extraction of the data and MLP network fulfils the role of the estimator of the concentration of the gas components. The obtained results have shown that the array of partially selective sensors, cooperating with hybrid neural network, can be used to determine the individual analyte concentrations in a mixtures of gases with good accuracy. The hybrid network is a reasonably small net and as a result, it learns faster and reaches good generalization ability with a reasonably small sized training data set. The system has the two interesting features, i.e. lower calibration cost and good accuracy.

An hybrid neural network based system for optimization of coagulant dosing in a water treatment plant

Abstract: Artificial neural network techniques are applied to the control of coagulant dosing in a drinking water treatment plant. Coagulant dosing rate is nonlinearly correlated to raw water parameters such as turbidity, conductivity, pH, temperature, etc. An important requirement of the application is robustness of the system against erroneous sensor measurements or unusual water characteristics. The hybrid system developed includes raw data validation and reconstruction based on the Kohonen self-organizing feature map, and prediction of coagulant dosage using multilayer perceptrons. A key feature of the system is its ability to take into account various sources of uncertainty, such as a typical input data, measurement errors and limited information content of the training set. Experimental results with real data are presented.

Improving learning accuracy of fuzzy decision trees by hybrid neural networks

Abstract: In the process of learning from examples with fuzzy representation, the higher learning accuracy is always expected. The paper proposes using a hybrid neural network to improve the learning accuracy of the fuzzy ID3 algorithm which is a popular and powerful method of fuzzy rule extraction without much computational effort. The proposed hybrid neural network corresponds to a fuzzy reasoning method in which the concept of local weights and global weights is employed. The time to consult with domain experts to adjust the weights for improving the learning accuracy will be greatly reduced due to the learning capability of the hybrid neural network. The synergy between fuzzy decision tree induction and a hybrid neural network offers new insight into the construction of hybrid intelligent systems.

Incorporation of Phenomenological Models in a Hybrid Neural Network for Quality Control of Injection Molding

Abstract: Injection molding is characterized by complex dynamics, which makes quality difficult to control. This is because the exact relations among the machine inputs, material properties, and molded part quality are not known precisely. Hence, the existing models for quality prediction have a limited accuracy and difficulty in application to general molding applications. This article investigates the integration of analytical process knowledge and artificial neural networks as a solution for quality prediction of molded parts, with accuracy increased toward quality control targets of three defects per million (60). This article describes the hybrid system based on the neural network and process knowledge, then compares its performances with conventional neural models for the prediction of the injection pressure.

Hybrid neural network system for texture analysis

Abstract: Texture classification and segmentation in digital images is commonly achieved using spatial grey level dependence matrices (SGLDMs), often referred to as co-occurrence matrices. This involves the computation of many matrices over a range of different spatial separations and orientations. The approach proposed in this paper uses a hybrid neural network system, consisting of a self-organising map followed by a backpropagation network, to restrict the number of SGLDMs that need to be computed. The system is trained in two phases on images with known texture content. The trained system is able to provide information, in the form of pixel spacing and orientation, on the texture content of unseen images. This information may be used to select appropriate SGLDMs for further texture classification. Experimental results are presented which demonstrate the effective performance of the system.

Feature extraction from photographic images using a hybrid neural network

Abstract: A new method for extracting features from photographic images has been developed. The input image is through a pulse coupled neural network transformed to a set of signatures, well suited for classification by unsupervised neural networks. A strategy using multiple self-organizing feature maps in a hierarchical manner is developed. With this approach, using a certain degree of supervision, an acceptable classification is obtained when applied to test images. The method is applied to license plate recognition.

A Hybrid Neural and Circuit-Based Model Structure for Microwave Modeling

Abstract: Neural networks have recently gained attention as powerful vehicles to microwave modeling, simulation, and optimization. A hybrid neural network structure incorporating prior circuit knowledge is proposed for modeling microwave components. In the proposed structure, a sub neural network establishes the mapping between original model input space and approximate circuit model input space. The neural network can learn such complicated space-mapping by training with EM simulation data. The hybrid neural models are computationally efficient and have an accuracy that is comparable to EM simulation. The proposed methodology is demonstrated through practical microwave modeling examples.

An embedded hybrid neural network and expert system in a computer-aided design system

Abstract: An embedded hybrid neural network and expert system functioning within a computer-aided design (CAD) system is developed to provide expert advice and self-learning for the structural adaptation of a mathematical model of electrical machines. The knowledge based intelligent system is designed to incorporate an optimization search of the design parameters of electric motors, while allowing the potential to develop a custom base of knowledge for application in related areas. The intelligent system is updated through a CAD system execution as well as in a text editor. The neural network component of the system is used to link the units of the expert system, and the network is trained with the gradient descent method. Application of the intelligent system demonstrates its reliability and assistance in the adaptation process.

Hybrid neural network models for environmental process control (The 1998 Hunter Lecture)

Abstract: A model that includes both first principles differential equations and an artificial neural network is used to forecast and control an environmental process. The inclusion of the first principles knowledge in this hybrid model is shown to improve substantially the stability of the model predictions in spite of the unmeasurability of some of the key parameters. The hybrid model estimates the unobservable parameters, and because of the constraints provided by the first principles equations, provides sensible extrapolations to the model. Thus, it can be used for process optimization as well as prediction. The hybrid model is compared with both a simple neural network with no a priori information, as well as some standard modern nonparametric statistical methods. For a variety of simulated parameter values, the hybrid model is shown to be comparable in predictive ability when used for interpolation and far superior when used for extrapolation. Copyright © 1999 John Wiley & Sons, Ltd.

Hybrid neural network for modeling automotive clutches

Abstract: In this paper, artificial neural network (ANN) based models to capture the dynamic engagement torque of a wet clutch system are developed and analyzed. A two-layer recurrent ANN with output feedback is chosen as the baseline architecture since its simplicity allows easy implementation and efficient execution. Although this model exhibits good performance in capturing the overall mean level of the engagement torque as a function of time, it is unable to predict some of the important clutch dynamics behaviors. To improve the performance, additional neurons that represent the first principles of the clutch engagement process are incorporated into the neural network model. In other words, a hybrid model in which physical knowledge is implicitly embedded within the ANN architecture is derived. This hybrid model is trained and tested with experimental data. The results show that the performance of the hybrid network is much superior to that of the baseline ANN. It can successfully capture not only the trends, but also the detailed characteristics of the clutch engagement torque as a function of time.

Hybrid neural network multivariable predictive controller for handling abnormal events in processing applications

Abstract: We describe a hybrid controller that uses neural networks and multivariable predictive control (MPC) to handle abnormal events in process applications. The controller detects abnormal situations, such as grinding mill spills or mill power excursions in mineral processing, or incipient flooding in separation columns and then reconfigures the multivariable controller to stabilize the operations. Neural networks are typically used to detect and classify the abnormal situation and knowledge of process dynamics and interactions is used to reconfigure the multivariable predictive controller parameters to stabilize the operations. Thus the MPC can be configured and tuned to provide good control around the 'normal' operating range, and when an upset occurs and is detected a new set of tuning parameters are used.

Forensic application of a novel hybrid neural network

Abstract: Trace compound identification forms an important element of forensic science Innovative instrumental designs based on Raman spectroscopy have made possible its in-situ use on fingerprint samples Recently, the pulse-coupled neural network (PCNN), an oscillatory model neural network, has been used for invariant feature extraction for object recognition and classification In this paper, we propose a novel hybrid neural network model for quick identification of trace materials from their Raman images This network consists of a PCNN preprocessor The features (icons) generated by the PCNN are then fed into a feedforward neural network for classification

Hybrid neural network for gas analysis measuring system

Abstract: The paper presents the application of the hybrid neural network to the solution of the calibration problem of the solid state sensor array used for the gas analysis The applied neural network is composed of two parts: the selforganizing Kohonen layer and multilayer perceptron (MLP). The role of the Kohonen layer is to perform the feature extraction of the data and the MLP network fulfils the role of estimator of the concentration of the gas components. The obtained results have shown that the array of partially selective sensors, cooperating with hybrid neural network, can be used to determine the individual analyte concentrations in the mixture of gases with good accuracy. The hybrid network is a reasonably small net and thanks to this if learns faster and reaches good generalization ability at reasonably small size of training data set. The system has two interesting features: lower calibration cost and good accuracy.

Hybrid Neural-Network/Computational Programs to the Analysis of Elastic-Plastic Structures

Abstract: In recent 7-8 years the Artificial Neural Networks (ANNs) have been widely introduced to structural engineering and especially to mechanics of structures and materials [1]. ANNs simulate biological neural networks very primitively. This concerns especially computer implementation of ANNs, called for short neurocomputing.

Estimate traffic control patterns using a hybrid neural network

Abstract: Many operating agencies are currently developing computerized freeway traffic management systems to support traffic operations as part of the intelligent transportation system (ITS) user service improvements. This study illustrates the importance of using simplified data analysis and presents a promising approach for improving demand prediction and traffic data modeling to support pro-active control. This study found that the approach of combining advanced neural networks and conventional error correction is promising for improved ITS applications.

A pulse-coupled-multilayer perceptron hybrid neural network for condition monitoring

Abstract: Studies of the visual cortex of cats highlight the role of temporal processing using synchronous oscillations for object identification. In this paper, a modified Eckhorn pulse coupled neural network model, the Computer-Model Cortex Neural Network, is used as a preprocessor to a neural network classifier for the condition monitoring of power transformers.

A probabilistic self-organizing classification neural network architecture

Abstract: This paper introduces a new analysis of the output map of Kohonen's self-organizing map. Using this analysis we are able to use the SOM as a supervised net. PSSOM's major advantage is its ability to assign degrees of classification certainty to unseen test data. This paper also investigates the application of this analysis as a first level in a hybrid neural network model. Our experiments show how this analysis tool can be used at the root of a hierarchical classifier model to increase considerably the overall speed of network training without loss of accuracy.

Heuristic Procedure Neural Networks for the CMST Problem

Abstract: Scope and Purpose – For solving combinatorial optimization problems, neural networks have traditionally been outperformed by traditional heuristic techniques developed specifically for the problem in question. This research is a step toward integrating the problem specific knowledge embedded in a traditional heuristic with the adaptive capabilities of neural networks. This is accomplished by creating a neural network topological design that embeds the steps of the traditional heuristic. The neural network learning then improves upon the performance of the embedded heuristic by modifying the neural weights attached to the embedded heuristic. Combinatorial optimization problems are by nature very difficult to solve, and the Capacitated Minimum Spanning Tree problem is one such problem. Much work has been done in the management sciences to develop heuristic solution procedures that suboptimally solve large instances of the Capacitated Minimum Spanning Tree problem in a reasonable amount of time. The Capacitated Minimum Spanning Tree problem is used in this paper to develop and demonstrate a hybrid neural network methodology that incorporates heuristic methods into the neural network topological design. The heuristic procedure is embedded into the neural network topological design, and an iterative improvement process is performed using the neural network. The semi-relaxed energy function of the problem is used to develop a neural network weight adjustment procedure that modifies the problem costs. In three-quarters (75%) of our experiments, the hybrid neural networks produced better results than any of the traditional procedures tested.

Approaching the post nonlinearity of blind mixtures by hybrid neural network

Abstract: It is very difficult to approach the post nonlinearity of blind mixtures. The recent neural networks for separating the post nonlinear blind mixtures are limited to the diagonal nonlinearity. In this paper a hybrid neural network is proposed to separate the post nonlinearly mixed blind signals with cross-channel disturbance. This hybrid network consists of a new neural blind de-mixer for approximating the post nonlinearity and a common network for separating the predicted linear mixtures. The blind de-mixer is made up of two subnets, which in total produce a "weak" nonlinear operator and can approach relatively strong nonlinearity by parameter-tuning. A six-step batch learning algorithm based on the fixed-point algorithm and information backpropagation is deduced. Preliminary results on a blind signal separation problem of two sources and four different types of post nonlinearity indicate the effectiveness of our model.

Modeling higher level processing functions inherent to the human brain

Abstract: The most significant feature of the information processing in the brain might be the autonomy based on the motivation and self reward (MSR) to form a processor sequence intending to find out the solution to a problem one is facing. In this paper, we show some preliminary ideas to incorporate the concept of MSR in designing brain-like information processing means, based on physiological and engineering points of view. We propose a hybrid neural network model as an extension of Hebb's rule, to be hypothesized for the function of association areas in the cerebral cortex. The generated neural network model is tested on the problem of segmentation of brain magnetic resonance images.

Hybrid neural network for the identification of narcotics and explosives

Abstract: Raman spectroscopy proves to be a versatile technique for forensic tasks Whilst it is very easy and fast to record a spectrum, it may be very time-consuming for the non-experts (and even the experts) to identify chemicals from the large database of Raman spectra of these materials Parallel developments in the field of neural networks have come to a stage that they can participate well in the recognition of these materials In this paper, a novel neural network-based technique is developed and presented for the quick identification of chemical compounds, in particular narcotics and explosives Preliminary results suggest this hybrid network has a great potential for forensic applications and for recognition of materials, in general

Feedforward neural network with adaptive reference pattern layer.

Abstract: A hybrid neural network architecture is investigated for modeling purposes. The proposed hybrid is based on the multilayer perceptron (MLP) network. In addition to the usual hidden layers, the first hidden layer is selected to be an adaptive reference pattern layer. Each unit in this new layer incorporates a reference pattern that is located somewhere in the space spanned by the input variables. The outputs of these units are the component wise-squared differences between the elements of a reference pattern and the inputs. The reference pattern layer has some resemblance to the hidden layer of the radial basis function (RBF) networks. Therefore the proposed design can be regarded as a sort of hybrid of MLP and RBF networks. The presented benchmark experiments show that the proposed hybrid can provide significant advantages over standard MLPs and RBFs in terms of fast and efficient learning, and compact network structure.

Hybrid neural network techniques for storm system identification and tracking

Abstract: In this paper a hybrid neural network/genetic algorithm (NN/GA) approach is presented that analyzes the behavior of storm systems from one time frame to the next. The goal of the hybrid neural network algorithm is to improve the classifier output by reducing the number of infeasible solutions using constraint optimization techniques. The input to the hybrid neural network algorithm is the output from a traditional backpropagation neural network. The hybrid NN/GA analyzes the backpropagation neural network output for logical consistencies and makes changes to the classification results based on strength of neural network classifications and satisfaction of logical constraints. The results are compared with classification results obtained using the linear discriminant analysis, k-nearest neighbor rule, and backpropagation neural network techniques.

A novel hybrid neural network for identification of trace compounds

Abstract: Developments in the field of neural networks have come to a stage that they can participate well in the recognition of these trace compounds. A novel neural network model is proposed for quick identification of trace compounds from their Raman images.