Catálogo de publicaciones - libros

In this paper we improve on the supervised classification method developed in Kornienko et al. (2002) by the introduction of Principal Components Analysis to the inference process. We also extend the classifier from dealing with binomial (two-class) problems only to multinomial (multi-class) problems. The application to which the MML criterion has been applied in this paper is the classification of objects via a linear hyperplane, where the objects are able to come from any multi-class distribution. The inclusion of Principal Component Analysis to the original inference scheme reduces the bias present in the classifier’s search technique. Such improvements lead to a method which, when compared against three commercial Support Vector Machine (SVM) classifiers on Binary data, was found to be as good as the most successful SVM tested. Furthermore, the new scheme is able to classify objects of a multiclass distribution with just one hyperplane, whereas SVMs require several hyperplanes.

A novel Feature-Watermarking algorithm based on wavelet packet decomposition was presented in this paper. We first propose the concept of Feature-Watermark. Dither modulation embedding scheme in wavelet packet coefficients promises the hiding of large capacity of robust information and fulfills watermark blind-extraction. Experimental results show that our method successfully fulfills the compromise between the robustness and capacity.

This paper presents a novel hybrid clustering approach that takes advantage of the efficiency of k -Means clustering and the effectiveness of hierarchical clustering. It employs the combination of geometrical information defined by k -Means and topological information formed by the Voronoi diagram to advantage. Our proposed approach is able to identify clusters of arbitrary shapes and clusters of different densities in O ( n ) time. Experimental results confirm the effectiveness and efficiency of our approach.

Inspired by the theory of multi-resolution analysis of wavelet transform, combining advantages of multi-resolution theory and support vector machine, a new regression model that is called multi-resolution support vector regression (MR-SVR) for function regression is proposed in this paper. In order to construct MR-SVR, the scaling function at some scale and wavelets with different resolution is used as kernel of support vector machine, which is called multi-resolution kernel. The MR-SVR not only has the advantages of support vector machine, but also has the capability of multi-resolution which is useful to approximate nonlinear function. Simulation examples show the feasibility and effectiveness of the method.

Clustering analysis is used to analyze the clustering phenomenon occurred to the data structure. However, there are some problems when the decision maker attempts to use clustering analysis. For solving these existing problems, this paper proposes a revised Entropy Clustering Analysis method requiring no prior setting of clusters, which is based on the mean distance between the data points and the cluster center. Through using several experiments and comparing different clustering analysis methods with proposed method, the results show that the proposed clustering method could achieve reasonable clustering effect. The experiment also proves that using the attributes with high correlation coefficient in clustering can achieve higher clustering accuracies.

Most classification algorithms suffer from an inability to detect instances of classes which are not present in the training set. A novel approach for characteristic concept rule learning called IC ^2 is proposed in this paper.

In this paper three machine learning approaches, Neural Networks (NN), Support Vector Machines (SVM) and Neural Fuzzy Networks (FuNN) are used to extract rules and assess the reliability of complex networks. For NN and SVM models the TREPAN approach is proposed as a valid tool for extracting rules whereas the Adaptive Neuro-Fuzzy Inference System (ANFIS) is used for tuning a previous set of rules derived by a fuzzy inference system and neural network approach.

Creating Petri Net domain models faces the same challenges that confront all knowledge-intensive AI performance systems: model specification, knowledge acquisition, and refinement. Thus, a fundamental question to investigate is the degree to which automation can be used. This paper formulates the learning task and presents the first machine learning method for Time Interval Petri Net (TIPN) domain models. In a preliminary evaluation within a damage control domain, the method learned a nearly perfect model of fire spread augmented with temporal and spatial data.

This paper proposes the model based detection method of abnormal acoustic source using a microphone array. General source location algorithm using a microphone array can be used to locate a dominant acoustic source, while this does not verify whether the detected source is permitted one or not on outdoor environments. It is difficult to discern it among a natural environmental sound. Thus, to cope with this problem, we propose the out-of-normal acoustic rejection method based on N-best likelihood ratio test using natural environmental sound models. In order to evaluate the proposed algorithm, a real-time DSP was constructed, and experimental evaluation is described.

Mobile phones are becoming a popular platform for a range of applications. However, due to size restrictions, the interfaces of these applications can be difficult to use. Customising an interface for a particular user offers the potential to improve an interface’s efficiency. In this paper, we propose customising a mobile phone’s Profile application. We apply a machine learning approach to discover concepts that describe a user’s profile-activations in terms of their scheduled appointments. We found that it is possible to learn useful concepts, which maybe used to improve the users interaction with mobile phone devices.