

Articles
by
A. Benyettou 
Total Records (
12 ) for
A. Benyettou 





D. Yedjour
,
H. Yedjour
and
A. Benyettou


Genetic algorithms are very efficient in the problems of exploration and seem to be able to find the single optimal solution in a huge space of possible solutions. However, they are ineffective when it comes to finding the exact value of the optimum in this space. This is precisely what the exact optimization algorithms perform best. It is therefore normal to think of combining exact and genetic algorithm to find the exact value of the optimum. McRULEGEN is a new system to extract symbolic rules from a trained neural network, based on two approaches genetic and exact. MCRULEGEN consists of three major components: A multilayer perceptron neural component, a genetic component, a simplification rules component based on Quine McCluskey algorithm. Our method is tested on breast cancer and iris databases, the computational results have shown that the performances of the rules extracted by MCRULEGEN are very high. 





Z.M. Mekkakia
,
S. Selka
and
A. Benyettou


This study aims two objectives. The first is the comparative study between the behavior of flows front of the presence (crossing) of a differentiated services domain and a classic heterogeneous network. The obtained results prove that the Diffserv model assures an equitability (same quality of service) for the treatment of flows of the same type. The second objective aim the study of the behavior of the TCP and UDP flows sharing the same domain of the differentiated services with the same priority class of AF, in order to define so equitability exists. Results show that the UDP flow is aggressive (insensible) screw to screw of the TCP flow 




A. Taleb
and
A. Benyettou


In this study, We propose a fuzzy neural system containing inferred rules which are modelled separately by a three layer perceptron neural network giving the conclusion part according to the premise of the rule. Such a system is applied to different morphology words for Arabic vowels recognition as a twodimensional fuzzy implication presented in the form of linguistic features values. The system has been implemented on a realtime minicomputer and is now operational, the results concerning a multispeaker corpus of continuous speech are also promising. 




D. Benhaddouche
and
A. Benyettou


In this study one used the dated mining to extract from knowledge biomedical, one used in the training course of training two algorithms Decision trees and SVMs. These methods of supervised classification are used to make diagnoses it for the disease hypothyroid and which gives custom the model of the extraction of data. We have studied the advantages of both methods, also by concrete results; we conclude that the method of SVM (Support Vector Machine) is better in our case. 




A. Lotfi
,
K. Mezzoug
and
A. Benyettou


This study presents the principle of operation of the Rotated Kernel Neural Network (RKNN) for radar target detection in nonGaussian noise. This classifier is based on adopting the architecture of standard probabilistic neural networks and using different kernel functions to approximate density functions. The training algorithm for this classifier is more complicated than the original PNN training algorithm but allow better generalization. Performance curves of the Rotated Kernel Neural Network are compared to those of probabilistic neural networks (original), Radial Basis Neural Networks with an expectation maximization training algorithm and Back propagation neural networks for Radar target detection in background noise in terms of probability of detection versus signaltonoise ratio (SNR). For most cases, the Rotated Kernel Neural Network classifier outperforms other conventional Radar target detection techniques and presents the advantage of resistance to background noise for values of SNR greater than 5 dB. 




N. Neggaz
,
M. Besnassi
and
A. Benyettou


Automatic facial expression analysis is an interesting and challenging problem and impacts important applications in many areas such as human–computer interaction. This study discusses the application of improved Active Appearance Model (AAM) based on evolutionary feature extraction in combination with Probabilistic Neural Network (PNN) for recognition of six different facial expressions from still pictures of the human face. Experimental results demonstrate an average expression recognition accuracy of 96% on the JAFFE database, which outperforms the rate of all other reported methods on the same database. The present study, therefore, proves the feasibility of computer vision based on facial expression recognition for practical applications like surveillance and human computer interaction. 




M. Hendel
,
A. Benyettou
,
F. Hendel
and
H. Khelil


In this study, we apply the wavelet transform and the fusion of two Bayesian neural networks to design an electrocardiogram (ECG) beat classification system; our main objective is to minimize at the maximum the miss diagnostic. Six ECG beat types namely: Normal beat (N), Left Bundle Branch Bloc (LBBB), Right Bundle Branch Block (RBBB), Premature Ventricular Contraction (PVC), Atrial Premature contraction (APB) and the Paced Beat (PB), obtained from the MITBIH ECG database were considered. First, five Discrete Wavelet Transform (DWT) levels were applied to decompose each ECG signal beat (360 samples centered on the peak R) into timefrequency representations. Statistical features relative to the three last decomposed signals and the last approximation, in addition to the original signal, were then calculated to characterize the ECG beat. Secondly, the fusion of RBFNN (Radial Basis Functions Neural Network) and BPNN (Back Propagation Neural Network) was employed as the classification system using as inputs, the calculated statistical features, in addition to the instantaneous RR interval. The proposed method achieves an equally well recognition rate of over 98% throughout all ECG beats type. These observations prove that the proposed beat classifier is very reliable and that it may be a useful practical tool for the automatic detection of heart diseases based on ECG signals. 




A. Ourdighi
and
A. Benyettou


In this study, we present an Adaptive Time Delay Neural Network (ATNN) training based in parallel genetic algorithms and local search method in a comparative study which applied in several problems. Usually, the ATNN training used a temporal variant of gradient descent algorithm and universal approximation function and consequently inherits problematic of parameters initialization and traps into local minimum. Besides, the algorithm was based on the peculiarity to adapt not only the synaptic weight but also each delay of interconnection which singularizes the ATNN architecture. This adaptation paradigm offers more flexibility for the network to attain the optimal timedelays and to achieve more accurate pattern mapping and recognition than is the case of using arbitrary fixed delays, as has been done previously by Time Delay Neural Network (TDNN). Also, this principal provoked instability on converging process of gradient descent rules and affected the results. So, our aim is to replace discriminated method algorithm to stochastic approach, the training will be base on parallel genetic algorithms: multipledeme parallel genetic algorithms. The important characteristics of multipledeme parallel GAs are the use of a few relatively large subpopulations and migration. The model was tested on Time series prediction of MackeyGlass a chaotic series and phonetic classification. Index Terms Adaptive TimeDelay Neural Network (ATNN), Adaptable delay, Synaptic weight, Multipledeme parallel genetic algorithms, local minimum, immigration. 




D. Yedjour
,
H. Yedjour
and
A. Benyettou


Neural networks are very efficient in solving various problems but they have no ability of explaining their answers and presenting gathered knowledge in a comprehensible way. Two main approaches are used, namely the pedagogical one that treats a network as a black box and the local one that examines its structure. Because searching rules is similar to NPhard problem it justifies an application of evolutionary algorithm to the rule extraction. Pedagogical approaches such as GA are insensitive to the number of units of neural networks as they see them as "black boxes" interested only their inputs and their outputs. In the study we describe new rule extraction method based on evolutionary algorithm called GenRGA. It uses logical rules and is composed of three (03) main parts: genetic module, neural networks module and rules simplification module. GenRGA is tested in experimental studies using different benchmark data sets from UCI repository. Comparisons with other methods show that the extracted rules are accurate and highly comprehensible. 




Y. Dahmani
and
A. Benyettou


The objective of this work tries to answer the question, in what the reinforcement
learning applied to fuzzy logic can be of interest in the field of the reactive
navigation of a mobile robot. In the first instance we have established an algorithm
applying the reinforcement learning to fuzzy limited lexicon. We have applied
it to a robot for the training of the followup of a rectilinear trajectory
of a starting point "D" at a point of unspecified arrival "A", while avoiding
with the robot butting against a possible obstacle. 




Y. Dahmani
and
A. Benyettou


This study deals with avoider obstacle behaviour of a mobile robot in unknown
environment. The use of fuzzy logic allows to provide universal rules for avoiding
obstacles, and on other hand it gives assurance for a predicted and reliable
behaviour of the robot. By lack of reference trajectory, it was urged to use
one of reinforcement learning method, fuzzy Qlearning which allows to take
into account continual state spaces and actions. In this study, an adaptation
method was proposed of fuzzy linguistic rules: after the stage of rules extraction
from a fuzzy inference system of fixed structure, we provide a methodology for
parameter tuning of the fuzzy sets in terms of the robot inertia without affecting
the conclusion rules. 





Y. Dahmani
and
A. Benyettou


In this article, we presented the QLearning training method which is a derivative of the
reinforcement learning called sometimes training by penaltyreward. We illustrate this by an application to the mobility of a mobile in an enclosure closed on the basis of a starting point towards an unspecified arrival point. The objective is to find an optimal way optimal without leaving the enclosure. 





