Noisy time series are typical results of observations or technical measurements. Noise reduction and signál structure saving are contradictory but useful aims. Non-linear time series processing is a way for non-gaussian noise suppression. Many valued algebras enriched by square root are able to realize the operators close to the weighted averages. Fuzzy data processing based on Łukasiewicz algebra [3] with square root satisfies the Lipschitz condition and causes constrained sensitivity of the mapping. The paper presents a fuzzy neural network based on Modus Ponens [1] with fuzzy logic function [6] preprocessing in the hidden layer. AU the fuzzy algorithms were realized in the Matlab systém and in C++. The fuzzy processing is applied to prediction of sunspot numbers. The systematic approach based on filter selection is combined with weight optimization.
Face recognition has a wide range of applications such as personal identification and authentication, criminal identification, security and surveillance, image and film processing, and human-computer interaction. Although there exist many methods, this paper proposes recent face recognition using a dynamic programming algorithm for image recognition and classification. A method based on a new mapping network called wavelet-network, namely Wavenet transform (WN). WN was employed to make approximation to the images before passing through the discrete wavelet transform decomposition to extract the image descriptive features. These features are used in the proposed image identification algorithm for enhancing the accuracy of recognition at pixel level and to minimize the additive cost function.
The proposed hybrid transform is based on the combination of the Wavenet (WN) and the Inverse Discrete Wavelet Transform (IDWT) followed by a Neural Network (NN) to be considered as feature extractor for the given image. In this paper the neural network (NN) classifier is combined with the wavelet transform. A reference set of 100 images are used and collected from different data images. This method gave an excellent and successful identification rate of 99%. Gaussian noise was added for further testing, the proposed algorithm for the same collected images and identification rate of 95% was achieved with level of up to 0.10.
The algorithm was implemented using MATLAB programming languages version 7.
Neural network based on back-propagation (BP) algorithm is a widely used prediction model. However, the nodes number of the first hidden layer, the learning rate and momentum factor are usually determined manually, which affects the forecast accuracy of network. Therefore, in this paper, to improve the forecast accuracy, firstly, the nodes number of the first hidden layer is selected adaptively based on minimizing mean square error (MSE). Secondly, improved genetic algorithm (GA) is proposed to train the learning rate and momentum factor dynamically, which includes multi-point crossover and single point mutation. Thirdly, we construct a new neural network model based on the adaptively selected nodes number of the first hidden layer, the dynamically selected learning rate and momentum factor, which is called HN-GA-BP neural network model. Finally, the proposed neural network model is used to forecast the carbon dioxide contents in China for fifty years. Experimental results demonstrate the effectiveness of the proposed HN-GA-BP neural network model.
Intrusion detection systems (IDSs) are designed to distinguish normal and intrusive activities. A critical part of the IDS design depends on the selection of informative features and the appropriate machine learning technique. In this paper, we investigated the problem of IDS from these two perspectives and constructed a misuse based neurotree classiffier capable of detecting anomalies in networks. The major implications of this paper are a) Employing weighted sum genetic feature extraction process which provides better discrimination ability for detecting anomalies in network trafic; b) Realizing the system as a rule-based model using an ensemble efficient machine learning technique, neurotree which possesses better comprehensibility and generalization ability; c) Utilizing an activation function which is targeted at minimizing the error rates in the learning algorithm. An extensive experimental evaluation on a database containing normal and anomaly trafic patterns shows that the proposed scheme with the selected features and the chosen classiffier is a state-of-the-art IDS that outperforms previous IDS methods.
One of the useful areas of the 2D image processing is called de-noising. When both original (ideal) and noisy images are available, the quality of de-noising is measurable. Our paper is focused to local 2D image processing using Łukasiewicz algebra with the square root function. The basic operators and functions in given algebra are first defined and then analyzed. The first result of the fuzzy logic function (FLF) analysis is its decomposition and realization as Łukasiewicz network (ŁN) with three types of processing nodes. The second result of FLF decomposition is the decomposed Łukasiewicz network (DŁN) with dyadic preprocessing and two types of processing nodes. The decomposition chain, which begins with FLF and converts it to ŁN and then to DŁN, terminates as Łukasiewicz artificial neural network (\lann) with dyadic preprocessing and only one type of processing node. Then the ŁANN is able to learn its integer weights in the ANN style. We are able to realize a set of individual FLF filters as ŁANN. Their preprocessing strategies are based on the pixel neighborhood, sorted list, Walsh list, and L-estimates. The quality of de-noising can be improved via compromise filtering. Two types of compromise de-noising filters are also realizable as ŁANN. One of them is called constrained referential neural network (CRNN). The other one is called dyadic weight neural network (DWNN). The compromise filters operate with the values from the set of individual filters. Both CRNN and DWNN are able to increase the quality of image processing as demonstrated on the biomedical MR image. All the calculations are realized in the Matlab environment
The paper deals with a predictive vector quantization of an image
based on a neural network architectures, wliere a vector predictor is iniplernented by three-layer neural network with various hidden nodes and bias units, sigrnoid function as nonlinearity and where vector quantizer is inipleinented by Kohonen self-organizing feature maps, it means the codebook is obtained by neural network clustering algorithm. We have tested aíi influence of a nuinber of hidden nodes, various convergention rates of a learning algorithm and a presence of the sigrnoid function to a rnean square prediction error. Next we háve studied an influence of codebook size to a rnean sciuare quantization error, that means a performance of predictive vector quantization system for various bit rates. The image of Lena of size 512 X 512 pels was coded for various bit rates, where we háve ušed onedirnensional and two-dimensional vector prediction of the blocks of pels.
Slow fluctuating radar targets have shown to be very difficult to classify by means of neural networks. This paper deals with the application of time-frequency decompositions for improving the performance of neural networks for this kind of targets. Several topics, such as dimensionality reduction of the time-frequency representations and the optimum value of SNR for training are discussed. The proposed detector is compared with a single neural network for radar detection, showing that he performance is improved for slow fluctuating radar targets, especially for low values of the probability of false alarm.
We present a neural network application to the diagnosis of vocal and voice disorders. These disorders normally cause changes in the voice signal, so we use acoustic parameters extracted from the voice as inputs for the neural network. The selected neural network structure is Multilayer Feedforward. In this paper, we focus our application on the classification between pathologic and non-pathologic voices. The performance of the neural network is very good, 100 % correct in the test. Furthermore, having used neural network techniques to reduce the initial nuniber of inputs (35), we conclude that only two acoustic parameters are needed for the classification between normal and pathological voices. The application can be a very useful diagnostic tool because it is non-invasive, makes it possible to develop an automatic computer-based diagnosis system, reduces the cost and tirne of the diagnosis, is objective and can also be useful for evaluation of surgical, pharmacological and rehabilitation processes. Finally, we discuss the lirnitation of our work and possible future research.
In this paper, for a class of the complex nonlinear system control problems, based on the two-person zero-sum game theory, combined with the idea of approximate dynamic programming(ADP), the constrained optimization control problem is solved for the nonlinear systems with unknown system functions and unknown time-varying disturbances. In order to obtain the approximate optimal solution of the zero-sum game, the multilayer neural network is used to fit the evaluation network, the execution network and the disturbance network of ADP respectively. The Lyapunov stability theory is used to prove the uniform convergence, and the system control output converges to the neighborhood of the target reference value. Finally, the simulation example verifies the effectiveness of the algorithm.
In this paper, we suggest Evolution Algorithms (EA) for development of neural network topologies to find the optimal solution of some problems. Topologies are modified in feed-forward neural networks and in special cases of recurrent neural networks.
We applied two approaches to the tuning of neural networks. One is classical, using evolution principles only. In the other approach, the adaptation phase (training phase) of the neural network is raade in two steps. In the hrst step we use the genetic algorithm to find better than random starting weights (nearly optimal values), in the second step we use the backpropagation algorithm to finish the adaptation phase. This means that the starting weights for the backpropagation algorithm are not random values, but approximately optimum values. In this context, the fitness of a chromosome (neural network) is a function of its estimated test error (its estimated generalization ability).
Some results obtained by these methods are demonstrated in a prediction of Geo-Magnetic Storms (GMS) and Handwriting Recognition (HWR).