Showing papers on "Hybrid neural network published in 1998"

Accurate tissue injury assessment using hybrid neural network analysis

Abstract: Systems and methods using a neural network based portable absorption spectrometer system for real-time automatic evaluation of tissue injury are described. An apparatus includes an electromagnetic signal generator; an optical fiber (520) connected to the electromagnetic signal generator; a fiber optic probe (530) connected to the optical fiber; a broad band spectrometer (90) connected to the fiber optic probe (530); and a hybrid neural network connected to the broad band spectrometer (90). The hybrid neural network includes a principal component analyzer (200) of broad band spectral data obtained from said broad band spectrometer (90). The systems and methods provide unexpected advantages in that the accuracy of tissue injury analysis is improved.

Face Recognition: A Hybrid Neural Network Approach

Abstract: Faces represent complex, multidimensional, meaningful visual stimuli and developing a computa- tional model for face recognition is difficult (Turk and Pentland, 1991). We present a hybrid neural network solution which compares favorably with other methods. The system combines local image sam- pling, a self-organizing map neural network, and a convolutional neural network. The self-organizing map provides a quantization of the image samples into a topological space where inputs that are nearby in the original space are also nearby in the output space, thereby providing dimensionality reduction and invariance to minor changes in the image sample, and the convolutional neural network provides for partial invariance to translation, rotation, scale, and deformation. The convolutional network extracts successively larger features in a hierarchical set of layers. We present results using the Karhunen-Loe transform in place of the self-organizing map, and a multilayer perceptron in place of the convolu- tional network. The Karhunen-Lo` eve transform performs almost as well (5.3% error versus 3.8%). The multilayer perceptron performs very poorly (40% error versus 3.8%). The method is capable of rapid classification, requires only fast, approximate normalization and preprocessing, and consistently exhibits better classification performance than the eigenfaces approach (Turk and Pentland, 1991) on the database considered as the number of images per person in the training database is varied from 1 to 5. With 5 images per person the proposed method and eigenfaces result in 3.8% and 10.5% error respectively. The recognizer provides a measure of confidence in its output and classification error approaches zero when rejecting as few as 10% of the examples. We use a database of 400 images of 40 individuals which con- tains quite a high degree of variability in expression, pose, and facial details. We analyze computational complexity and discuss how new classes could be added to the trained recognizer.

Centroid based Multilayer Perceptron Networks

Abstract: In this study we investigate a hybrid neural network architecture for modelling purposes. The proposed network is based on the multilayer perceptron (MLP) network. However, in addition to the usual hidden layers the first hidden layer is selected to be a centroid layer. Each unit in this new layer incorporates a centroid that is located somewhere in the input space. The output of these units is the Euclidean distance between the centroid and the input. The centroid layer clearly resembles the hidden layer of the radial basis function (RBF) networks. Therefore the centroid based multilayer perceptron (CMLP) networks can be regarded as a hybrid of MLP and RBF networks. The presented benchmark experiments show that the proposed hybrid architecture is able to combine the good properties of MLP and RBF networks resulting fast and efficient learning, and compact network structure.

Recognition of paper currencies by hybrid neural network

Abstract: For the recognition of paper currencies by image processing, the two steps data processing approach can yield high performance. The two steps include "recognition" and "verification" steps. In the current recognition machine, a simple statistical test is used as the verification step, where univariate Gaussian distribution is employed. Here we propose the use of the probability density formed by a multivariable Gaussian function, where the input data space is transferred to a lower dimensional subspace. Due to the structure of this model, we refer the total processing system as a hybrid neural network. Since the computation of the verification model only needs the inner product and square, the computational load is very small. In this paper, the method and numerical experimental results are shown by using the real data and the recognition machine.

A robust hybrid neural architecture for an industrial sensor application

Abstract: A programmable hybrid neural network architecture has been used to implement a smart optical sensor with focal-plane pattern classification for a flexible manufacturing cell environment. The network contains an integrated photosensitive array based on modified photo BJTs as input to a fully-connected multilayer feedforward (MLFF) neural classifier. The architecture features a distributed neuron realization that employs a number of active nonlinear resistor circuits operating in parallel. It minimizes the effect of parameter variations due to non-uniform device fabrication over the die surface. Moreover, due to the modularity of the architecture and locality of interconnections, synaptic density has been doubled in comparison with a conventional realization. A photosensor-classifier chip consisting of a 2-D array of 64 neural-based smart pixels and additional neural network circuits has been fabricated. The proposed architecture has been implemented in both CMOS and BiCMOS process technologies as part of a sensor optimization study.

Hybrid identification of unlabeled nuclear power plant transients with artificial neural networks

Abstract: Proper and rapid identification of malfunctions (transients) is of premier importance for the safe operation of nuclear power plants. Feedforward neural networks trained with the backpropagation algorithm are frequently applied to model simulated nuclear power plant malfunctions. The correct identification of unlabeled transients-or transients of the "don't-know" type-have proven to be especially challenging. A novel hybrid neural network methodology is presented which correctly classifies unlabeled transients. From this analysis the importance for properly accommodating practical aspects such as the drift of electronics elements, numerical integration accumulating errors, and the digitization of simulated and actual plant signals became obvious. Various ANN based models were successfully applied to identify labeled and unlabeled malfunctions of the Hungarian Paks nuclear power plant simulator.

A new model in designing neural network in optimization: a hybrid neural network approach to machine scheduling

Abstract: This effort aims to make contributions in two areas of research. One is the area of neural network design. We present a new approach for neural network design in solving optimization problems, which is different from the traditional approach. Several limitations of traditional neural networks are discussed and the new hybrid neural network model is introduced to overcome those limitations. Neural network as a search technique in optimization has to prove itself not only capable of solving an optimization problem, but also be able to compete with other techniques on performance measures. We focus our study on two performance measures: efficiency and effectiveness. By definition, an efficient search technique can solve problems fast where as an effective technique will provide high quality solutions. Our research is successful in both directions. The second major contribution we seek to make is in the area of processor scheduling. We test our hybrid neural network by solving a processor scheduling problem, and we compare our neural network solutions with the solutions from the existing best heuristics from literature. The hybrid neural network outperforms the heuristics in most cases and is able to improve the solutions from starting with a heuristic solution. The solution time is fairly short, i.e., in the same order of the heuristics. The processor scheduling problem we solved here is the Flexible Flow Shop (FFS) scheduling. The problem is NP-complete, a category o f hard combinatorial optimization problems in literature. Applications of this problem include, but not limited to: computing systems with multiple processors at each phase where all tasks must be done through a series of phases; manufacturing systems with multiple machines at each stage where all jobs need to go through a series of stages. Therefore, solving the FFS problem is not only significant in the theory of optimization, but also in real world applications.

Supervised adaptive clustering: A hybrid neural network clustering algorithm

Abstract: A neural network architecture is introduced which implements a supervised clustering algorithm for the classification of feature vectors. The network is selforganising, and is able to adapt to the shape of the underlying pattern distribution as well as detect novel input vectors during training. It is also capable of determining the relative importance of the feature components for classification. The architecture is a hybrid of supervised and unsupervised networks, and combines the strengths of three wellknown architectures: learning vector quantisation, backpro-pagation and adaptive resonance theory. Network performance is compared to that of learning vector quantisation, back-propagation and cascade-correlation. It is found that performance is generally as good as or better than the performance of these other architectures, while training time is considerably shorter. However, the main advantage of the hybrid architecture is its ability to gain insight into the feature pattern space.

A hybrid system for well test analysis

Abstract: Petroleum well test analysis is a tool for estimating the average properties of the reservoir rock. It is a classic example of an inverse problem. Visual examination of the pressure response of the reservoir to an induced flow rate change at a well allows the experienced analyst to determine the most appropriate model from a library of generalized analytical solutions. Rock properties are determined by finding the model parameters that best fit the observed data. This paper describes a framework for hybrid network to assist the analyst in selecting the appropriate model and determining the solution. The hybrid network design offers significant advantages by reducing training time and allowing incorporation of both symbolic and numeric data. The network structure is described and the advantages and disadvantages compared to previous approaches are discussed.

A Hybrid Neural Network System for Trading Financial Markets

Abstract: In this chapter Marina Resta demonstrates experimentally the great potential of neural networks for design of systems for trading stock markets. She suggests the use of a hybrid neural network architecture that combines the approach of Self-Organizing Maps together with that of Genetic Algorithms. She shows the forecasting capabilities of this hybrid system and evidence of the performance of this system. This chapter is a nice extension of the applications of trading systems presented in “Trading on the Edge”. The novelty here is that genetic algorithms and self-organizing maps are combined.

Neural techniques for image segmentation

Abstract: We present new neural techniques including unsupervised technology and fuzzy logic foundations. We realized a hybrid neural network and applied three different unsupervised learning algorithms that we developed specially for it: fuzzy MLSOM, fuzzy hierarchical "neural gas" and fuzzy hierarchical "maximum entropy". The experiments presented deal with image segmentation. The results obtained show that neural networks are a valid instrument for image processing and shape recognition.

Active learning based on a hybrid neural network modeller

Abstract: OF THE PHD THESIS V arious m ethods are investigated for selecting training data for the purpose o f training neural netw orks. A n ew m ethod called M IQ R (M axim um Inter-Q uartile R ange) is proposed for effectively selecting a con cise set o f training data. In addition, the e n se m b le con cep t is introduced in this n ew m ethod. D ata selection is not unduly influenced by “o u t l i e r s ”, rather, it is principally dependent upon the “m a in s t r e a m ” output o f the ensem ble netw orks. E ncom passed in the n ew m ethod is a very sim ple ancient C hinese philosophical idea, i .e . “the m in o r i ty o b e y s th e m a jo r i ty ” . T h ese techniques are nonparam etric in the sense that several different neural netw orks com prise an ensem ble or com m ittee and co-operatively w ork together w ith each other to achieve a com m on goal. B eca u se th ese are different neural netw orks (hybrid m odel), they can be com plem entary in the entire learning system , and therefore effectively enhance the entire learning system ’s efficiency and accuracy. For learning, the neural netw orks attem pt to actively select the m ost inform ative and important training data. The m ethods described in this thesis pleasingly satisfy this need, and com pare favourably w ith contending m ethods. M any experim ents have been done to corroborate theoretical and empirical conjectures. The results are quite pleasing in that this n ew m ethod is not only as “active learning” m uch better than “passive learning” both in data selection and in generalisation perform ance, but also outperform s other existing contending active learning m ethods. In particular, the results are very satisfying and interesting w h en the m ethod is applied to d iscontinuous functions. A lthough th e experim ents are conducted w ith clean data selection , it should be easy to extend them to noisy data selection since the m ethod d eveloped is validated using unlabelled data. The algorithm developed for these m ethods has been rigorously tested , and proves to be highly autonom ous and robust. The m ethods develop ed here are not restricted to use on neural netw orks. M ore generally, they can be applied to other scientific research and econ om ic fields, even educational and socio log ica l behaviour.

Artificial Neural Networks for Generic Predictive Maintenance

Abstract: This paper outlines a research project to develop artificial neural networks as a diagnostic tool for the automatic identification of rotating machine faults. This work was instigated by the DTI Neural Computing Learning Solutions Campaign’s AXON Neural Projects Club. Industrial sponsors are Entek/IRD, Diagnostic Instruments Ltd., and Arjo Wiggins Paper Mills. The biggest problem encountered by developers of SMART software systems is that examples of all conditions to be identified are required. In practice this is not possible due to the routine method of plant machinery data collection, and due to the individual behaviour of the machinery. A method is required which is capable of diagnosing a previously unseen fault upon any bearing. This paper proposes a hybrid neural network approach which first determines a novel condition, then a knowledge base categorises the condition. The system is currently working off-line in support of the maintenance technician at the sponsors paper mill plant.