Abstract: In the present work, we propose a new technique to
enhance the learning capabilities and reduce the computation
intensity of a competitive learning multi-layered neural network
using the K-means clustering algorithm. The proposed model use
multi-layered network architecture with a back propagation learning
mechanism. The K-means algorithm is first applied to the training
dataset to reduce the amount of samples to be presented to the neural
network, by automatically selecting an optimal set of samples. The
obtained results demonstrate that the proposed technique performs
exceptionally in terms of both accuracy and computation time when
applied to the KDD99 dataset compared to a standard learning
schema that use the full dataset.
Abstract: Diagnosis can be achieved by building a model of a
certain organ under surveillance and comparing it with the real time
physiological measurements taken from the patient. This paper deals
with the presentation of the benefits of using Data Mining techniques
in the computer-aided diagnosis (CAD), focusing on the cancer
detection, in order to help doctors to make optimal decisions quickly
and accurately. In the field of the noninvasive diagnosis techniques,
the endoscopic ultrasound elastography (EUSE) is a recent elasticity
imaging technique, allowing characterizing the difference between
malignant and benign tumors. Digitalizing and summarizing the main
EUSE sample movies features in a vector form concern with the use
of the exploratory data analysis (EDA). Neural networks are then
trained on the corresponding EUSE sample movies vector input in
such a way that these intelligent systems are able to offer a very
precise and objective diagnosis, discriminating between benign and
malignant tumors. A concrete application of these Data Mining
techniques illustrates the suitability and the reliability of this
methodology in CAD.
Abstract: In this paper, the problem of stability analysis for a class of impulsive stochastic fuzzy neural networks with timevarying delays and reaction-diffusion is considered. By utilizing suitable Lyapunov-Krasovskii funcational, the inequality technique and stochastic analysis technique, some sufficient conditions ensuring global exponential stability of equilibrium point for impulsive stochastic fuzzy cellular neural networks with time-varying delays and diffusion are obtained. In particular, the estimate of the exponential convergence rate is also provided, which depends on system parameters, diffusion effect and impulsive disturbed intention. It is believed that these results are significant and useful for the design and applications of fuzzy neural networks. An example is given to show the effectiveness of the obtained results.
Abstract: This paper presents a new method of fault detection and isolation (FDI) for polymer electrolyte membrane (PEM) fuel cell (FC) dynamic systems under an open-loop scheme. This method uses a radial basis function (RBF) neural network to perform fault identification, classification and isolation. The novelty is that the RBF model of independent mode is used to predict the future outputs of the FC stack. One actuator fault, one component fault and three sensor faults have been introduced to the PEMFC systems experience faults between -7% to +10% of fault size in real-time operation. To validate the results, a benchmark model developed by Michigan University is used in the simulation to investigate the effect of these five faults. The developed independent RBF model is tested on MATLAB R2009a/Simulink environment. The simulation results confirm the effectiveness of the proposed method for FDI under an open-loop condition. By using this method, the RBF networks able to detect and isolate all five faults accordingly and accurately.
Abstract: The neural network's performance can be measured by efficiency and accuracy. The major disadvantages of neural network approach are that the generalization capability of neural networks is often significantly low, and it may take a very long time to tune the weights in the net to generate an accurate model for a highly complex and nonlinear systems. This paper presents a novel Neuro-fuzzy architecture based on Extended Kalman filter. To test the performance and applicability of the proposed neuro-fuzzy model, simulation study of nonlinear complex dynamic system is carried out. The proposed method can be applied to an on-line incremental adaptive learning for the prediction of financial time series. A benchmark case studie is used to demonstrate that the proposed model is a superior neuro-fuzzy modeling technique.
Abstract: This paper proposes a technique to protect against
email bombing. The technique employs a statistical approach, Naïve
Bayes (NB), and Neural Networks to show that it is possible to
differentiate between good and bad traffic to protect against email
bombing attacks. Neural networks and Naïve Bayes can be trained
by utilizing many email messages that include both input and output
data for legitimate and non-legitimate emails. The input to the model
includes the contents of the body of the messages, the subject, and
the headers. This information will be used to determine if the email
is normal or an attack email. Preliminary tests suggest that Naïve
Bayes can be trained to produce an accurate response to confirm
which email represents an attack.
Abstract: In this paper, Differential Evolution (DE) algorithm, a new promising evolutionary algorithm, is proposed to train Radial Basis Function (RBF) network related to automatic configuration of network architecture. Classification tasks on data sets: Iris, Wine, New-thyroid, and Glass are conducted to measure the performance of neural networks. Compared with a standard RBF training algorithm in Matlab neural network toolbox, DE achieves more rational architecture for RBF networks. The resulting networks hence obtain strong generalization abilities.
Abstract: In this paper, a novel associative memory model will be proposed and applied to memory retrievals based on the conventional continuous time model. The conventional model presents memory capacity is very low and retrieval process easily converges to an equilibrium state which is very different from the stored patterns. Genetic Algorithms is well-known with the capability of global optimal search escaping local optimum on progress to reach a global optimum. Based on the well-known idea of Genetic Algorithms, this work proposes a heuristic rule to make a mutation when the state of the network is trapped in a spurious memory. The proposal heuristic associative memory show the stored capacity does not depend on the number of stored patterns and the retrieval ability is up to ~ 1.
Abstract: By using the method of coincidence degree theory and constructing suitable Lyapunov functional, several sufficient conditions are established for the existence and global exponential stability of anti-periodic solutions for Cohen-Grossberg shunting inhibitory neural networks with delays. An example is given to illustrate our feasible results.
Abstract: This paper presents the use of a newly created network
structure known as a Self-Delaying Dynamic Network (SDN) to
create a high resolution image from a set of time stepped input
frames. These SDNs are non-recurrent temporal neural networks
which can process time sampled data. SDNs can store input data
for a lifecycle and feature dynamic logic based connections between
layers. Several low resolution images and one high resolution image
of a scene were presented to the SDN during training by a Genetic
Algorithm. The SDN was trained to process the input frames in order
to recreate the high resolution image. The trained SDN was then used
to enhance a number of unseen noisy image sets. The quality of high
resolution images produced by the SDN is compared to that of high
resolution images generated using Bi-Cubic interpolation. The SDN
produced images are superior in several ways to the images produced
using Bi-Cubic interpolation.
Abstract: Embedded systems need to respect stringent real
time constraints. Various hardware components included in such
systems such as cache memories exhibit variability and therefore
affect execution time. Indeed, a cache memory access from an
embedded microprocessor might result in a cache hit where the
data is available or a cache miss and the data need to be fetched
with an additional delay from an external memory. It is therefore
highly desirable to predict future memory accesses during
execution in order to appropriately prefetch data without incurring
delays. In this paper, we evaluate the potential of several artificial
neural networks for the prediction of instruction memory
addresses. Neural network have the potential to tackle the nonlinear
behavior observed in memory accesses during program
execution and their demonstrated numerous hardware
implementation emphasize this choice over traditional forecasting
techniques for their inclusion in embedded systems. However,
embedded applications execute millions of instructions and
therefore millions of addresses to be predicted. This very
challenging problem of neural network based prediction of large
time series is approached in this paper by evaluating various neural
network architectures based on the recurrent neural network
paradigm with pre-processing based on the Self Organizing Map
(SOM) classification technique.
Abstract: In modern human computer interaction systems
(HCI), emotion recognition is becoming an imperative characteristic.
The quest for effective and reliable emotion recognition in HCI has
resulted in a need for better face detection, feature extraction and
classification. In this paper we present results of feature space analysis
after briefly explaining our fully automatic vision based emotion
recognition method. We demonstrate the compactness of the feature
space and show how the 2d/3d based method achieves superior features
for the purpose of emotion classification. Also it is exposed that
through feature normalization a widely person independent feature
space is created. As a consequence, the classifier architecture has
only a minor influence on the classification result. This is particularly
elucidated with the help of confusion matrices. For this purpose
advanced classification algorithms, such as Support Vector Machines
and Artificial Neural Networks are employed, as well as the simple k-
Nearest Neighbor classifier.
Abstract: This paper reports work done to improve the modeling of complex processes when only small experimental data sets are available. Neural networks are used to capture the nonlinear underlying phenomena contained in the data set and to partly eliminate the burden of having to specify completely the structure of the model. Two different types of neural networks were used for the application of Pulping of Sugar Maple problem. A three layer feed forward neural networks, using the Preconditioned Conjugate Gradient (PCG) methods were used in this investigation. Preconditioning is a method to improve convergence by lowering the condition number and increasing the eigenvalues clustering. The idea is to solve the modified problem where M is a positive-definite preconditioner that is closely related to A. We mainly focused on Preconditioned Conjugate Gradient- based training methods which originated from optimization theory, namely Preconditioned Conjugate Gradient with Fletcher-Reeves Update (PCGF), Preconditioned Conjugate Gradient with Polak-Ribiere Update (PCGP) and Preconditioned Conjugate Gradient with Powell-Beale Restarts (PCGB). The behavior of the PCG methods in the simulations proved to be robust against phenomenon such as oscillations due to large step size.
Abstract: Recently, a lot of attention has been devoted to
advanced techniques of system modeling. PNN(polynomial neural
network) is a GMDH-type algorithm (Group Method of Data
Handling) which is one of the useful method for modeling nonlinear
systems but PNN performance depends strongly on the number of
input variables and the order of polynomial which are determined by
trial and error. In this paper, we introduce GPNN (genetic
polynomial neural network) to improve the performance of PNN.
GPNN determines the number of input variables and the order of all
neurons with GA (genetic algorithm). We use GA to search between
all possible values for the number of input variables and the order of
polynomial. GPNN performance is obtained by two nonlinear
systems. the quadratic equation and the time series Dow Jones stock
index are two case studies for obtaining the GPNN performance.
Abstract: This paper is mainly concerned with the application of
a novel technique of data interpretation for classifying measurements
of plasma columns in Tokamak reactors for nuclear fusion
applications. The proposed method exploits several concepts derived
from soft computing theory. In particular, Artificial Neural Networks
and Multi-Class Support Vector Machines have been exploited to
classify magnetic variables useful to determine shape and position of
the plasma with a reduced computational complexity. The proposed
technique is used to analyze simulated databases of plasma equilibria
based on ITER geometry configuration. As well as demonstrating the
successful recovery of scalar equilibrium parameters, we show that
the technique can yield practical advantages compared with earlier
methods.
Abstract: This paper presents a new approach to tackle the problem of recognizing machine-printed Arabic texts. Because of the difficulty of recognizing cursive Arabic words, the text has to be normalized and segmented to be ready for the recognition stage. The new scheme for recognizing Arabic characters depends on multiple parallel neural networks classifier. The classifier has two phases. The first phase categories the input character into one of eight groups. The second phase classifies the character into one of the Arabic character classes in the group. The system achieved high recognition rate.
Abstract: In this paper, the robust exponential stability problem of discrete-time uncertain stochastic neural networks with timevarying delays is investigated. By introducing a new augmented Lyapunov function, some delay-dependent stable results are obtained in terms of linear matrix inequality (LMI) technique. Compared with some existing results in the literature, the conservatism of the new criteria is reduced notably. Three numerical examples are provided to demonstrate the less conservatism and effectiveness of the proposed method.
Abstract: Most of the commonly used blind equalization algorithms are based on the minimization of a nonconvex and nonlinear cost function and a neural network gives smaller residual error as compared to a linear structure. The efficacy of complex valued feedforward neural networks for blind equalization of linear and nonlinear communication channels has been confirmed by many studies. In this paper we present two neural network models for blind equalization of time-varying channels, for M-ary QAM and PSK signals. The complex valued activation functions, suitable for these signal constellations in time-varying environment, are introduced and the learning algorithms based on the CMA cost function are derived. The improved performance of the proposed models is confirmed through computer simulations.
Abstract: An artificial neural network (ANN) model is
presented for the prediction of kinematic viscosity of binary mixtures
of poly (ethylene glycol) (PEG) in water as a function of temperature,
number-average molecular weight and mass fraction. Kinematic
viscosities data of aqueous solutions for PEG (0.55419×10-6 –
9.875×10-6 m2/s) were obtained from the literature for a wide range
of temperatures (277.15 - 338.15 K), number-average molecular
weight (200 -10000), and mass fraction (0.0 – 1.0). A three layer
feed-forward artificial neural network was employed. This model
predicts the kinematic viscosity with a mean square error (MSE) of
0.281 and the coefficient of determination (R2) of 0.983. The results
show that the kinematic viscosity of binary mixture of PEG in water
could be successfully predicted using an artificial neural network
model.
Abstract: Research into the problem of classification of sonar signals has been taken up as a challenging task for the neural networks. This paper investigates the design of an optimal classifier using a Multi layer Perceptron Neural Network (MLP NN) and Support Vector Machines (SVM). Results obtained using sonar data sets suggest that SVM classifier perform well in comparison with well-known MLP NN classifier. An average classification accuracy of 91.974% is achieved with SVM classifier and 90.3609% with MLP NN classifier, on the test instances. The area under the Receiver Operating Characteristics (ROC) curve for the proposed SVM classifier on test data set is found as 0.981183, which is very close to unity and this clearly confirms the excellent quality of the proposed classifier. The SVM classifier employed in this paper is implemented using kernel Adatron algorithm is seen to be robust and relatively insensitive to the parameter initialization in comparison to MLP NN.