Catálogo de publicaciones - libros

To the shortcoming of general particle filter, an improved algorithm based on neural network is proposed and is shown to be more efficient than the general algorithm in the same sample size. The improved algorithm has mainly optimized the choice of importance density. After receiving the samples drawn from prior density, and then adjust the samples with general regression neural network (GRNN), make them approximate the importance density. Apply the new method to target tracking problem, has made the result more precise than the general particle filter.

Study of algorithms and its design can be progressed in various dimensions. In this paper, we have a definite refinement of lower bound on the number of tracks required to route a channel. The attack is from a complementary viewpoint. Our algorithm succeeds to avoid all kind of approximation. The approach performs exact mapping of the problem into graphical presentation and analyzes the graph taking help of mimetic algorithm, which uses combination of sequential and GA based vertex coloring. Performance of the algorithm depends on how effectively mimetic approach can applied selecting appropriate values for the parameters to evaluate the graphical presentation of the problem. This viewpoint has immense contribution against sticking at local minima for this optimization problem. The finer result clearly exemplifies instances, which give better or at least the same lower bound in VLSI channel routing problem.

In this paper we propose a novel training algorithm for RBF networks that is based on extended kalman filter and fuzzy logic. After the user choose how many prototypes to include in the network, the extended kalman filter simultaneously solves for the prototype vectors and the weight matrix. The fuzzy logic is used to cope with the devergence problem caused by the insufficiently known a priori filter statistics. Results are presented on RBF networks as applied to the Iris classification problem. It is shown that the use of the extended Kalman filter and fuzzy logic results in faster learning and better results than conventional RBF networks.

This paper presents the MRBF network, a new algorithm adapted from the RBF network, to construct the classifiers for predicting phenotypic resistance on 6 protease inhibitors. The performance of the prediction was measured by 10-fold cross-validation, The results show that MRBF gives the lowest average mean square error (MSE) when compared with the traditional RBF network and multiple linear regression analysis (REG). Moreover, it provides the best average predictive accuracy when compared with HIVdb, REG, and Support Vector Machines (SVM).

The research into the impersonality and information of the Traditional Chinese-medicine Diagnosis (TCMD) is recognized to be a crucial work, especially for the Traditional Chinese-medicine (TCM) medicine examination system. However the tongue is like a mirror of the Viscera, and the pathological changes of them can reflect on the tongue. So according the information of the Sizhen for the syndromes of the Kidney, we design the initial structure of the PGNN (Probabilistic Genetic Neural Network), and its connection weight and the structure of the PGNN automatically will be optimized by the GA and others optimal way. It is certain that the groping research will be use for the modernization of the TCM.

A computer-based filling and emptying diesel engine simulation model has been developed, which can simulate the operational behavior of diesel engine under different performance failures and different running conditions. This paper firstly describes the simulation models and simulated results of a four-stroke, turbo-charged diesel engine. The calculated results in terms of relative deviation are analyzed which reveal the relations between thermodynamic variables, performance failures, running conditions and ship operation conditions. Further, it provides more complete understanding of failures’ behaviors under different running conditions and help to detect the failures amongst complex symptoms. Relative deviation of thermodynamic variables under different running conditions exhibits strong similarity, which induces a new information source for failure detection. Compared to experiments on board ship, this simulative modeling possesses advantages of shorter studying period, less research investment, lower risk in failure simulation and more symptomatic information. Finallyan example is introduced to verify the feasibility of relative deviation analysis in the use of diesel engine failure diagnosis with artificial neural network method.

In this paper, we propose to take structural context of image regions into account for region classification through a structural neural network. Firstly, a tree structure of each region is formed to characterize the relationship among the region and its neighbours. Such structures integrate both visual attributes of regions and their structural contexts. Then the structural representations are learned through a Back-propagation Through Structure (BPTS) training algorithm. Comprehensive experimental results demonstrate that our proposed approach has a great potential in region classification.

An auto adaptive neuro-fuzzy segmentation and edge detection architecture is presented. This system consists of a multilayer perceptron (MLP)-like network that performs image segmentation by adaptive thresholding of the input image using labels automatically pre-selected by kernel based fuzzy clustering technique. The proposed architecture is feed forward, but unlike the conventional MLP the learning is unsupervised. The output status of the network is described as a fuzzy set. Fuzzy entropy is used as a measure of the error of the segmentation system as well as a criterion for determining potential edge pixels. Exponential entropy was employed to overcome the drawbacks of using conventional logarithmic entropy. The proposed system is capable to perform automatic multilevel segmentation of images, based solely on information contained by the image itself. No a priory assumptions whatsoever are made about the image (type, features, contents, stochastic model, etc.). Such an “universal” algorithm is most useful for applications that are supposed to work with different (and possibly initially unknown) types of images. The proposed system can be readily employed, “as is,” or as a basic building block by a more sophisticated and/or application-specific image segmentation algorithm. By monitoring the fuzzy entropy relaxation process, the system is able to detect edge pixels

This paper is proposing the method for Thai handwritten character recognition. The methods are Robust C-Prototype and Back-Propagation Neural Network. The objective of experimental is recognition on Thai handwritten character. This is the result of both methods to be appearing accuracy more than 85%.

For grid-based image classification, an image is divided into blocks, and a feature vector is formed for each block. Conventional grid-based classification algorithms suffer from inability to take into account the two-dimensional neighborhood interactions of blocks. We present a classification method based on two-dimensional Conditional Random Fields which can avoid the limitation. As a discriminative approach, the proposed method offers several advantages over generative approaches, including the ability to relax the assumption of conditional independence of the observations.