Catálogo de publicaciones - libros

One of standard methods in vision research is measuring the psychometric functions (PFs) that are further analyzed implying the validity of traditional signal detection theory (SDT). This research paradigm contains essential inherent contradiction: in contrast to most empirical PFs the ones predicted by the SDT do not satisfy the Neyman-Pearson objective. The problem may successfully be overcome within the framework of recent binary signal detection theory (BSDT) providing PFs for which the objective required is always achieved. Here, the original BSDT theory for vision is for the first time applied to quantitative description of specific empirical PFs measured in experiments where the coding of facial identity has been studied. By fitting the data, some parameters of BSDT face recognition algorithm were extracted and it was demonstrated that the BSDT supports popular prototype face identification model. Results can be used for developing new high-performance computational methods for face recognition.

A new approach to short-term electrical load forecasting is investigated in this paper. As electrical load data are highly non-linear in nature, in the proposed approach, we first separate out the linear and the non-linear parts, and then forecast using the non-linear part only. Semi-parametric spectral estimation method is used to decompose a load data signal into a harmonic linear signal model and a non-linear trend. A support vector machine is then used to predict the non-linear trend. The final predicted signal is then found by adding the support vector machine predicted trend and the linear signal part. The performance of the proposed method seems to be more robust than using only the raw load data. This is due to the fact that the proposed method is intended to be more focused on the non-linear part rather than a diluted mixture of the linear and the non-linear parts as done usually.

Since the development of pulse compression in the mid-1950’s the concept has become an indispensable feature of modern radar systems. A matched filter is used on reception to maximize the signal to noise ratio of the received signal. The actual waveforms that are transmitted are chosen to have an autocorrelation function with a narrow peak at zero time shift and the other values, referred to as sidelobes, as low as possible at all other times. A new approach to radar pulse compression is introduced, namely the Reproducing Kernel Hilbert Space (RKHS) method. This method reduces sidelobe levels significantly. The paper compares a second degree polynomial kernel RKHS method to a least squares and -norm mismatched filter, and concludes with a presentation of the representative testing results.

The performance of many learning methods are usually influenced by the class imbalance problem, where the training data is dominated by the instances belonging to one class. In this paper, we propose a novel method which combines random forest based techniques and sampling methods for effectively learning from imbalanced data. Our method is mainly composed of two phases: data cleaning and classification based on random forest. Firstly, the training data is cleaned through the elimination of dangerous negative instances. The data cleaning process is supervised by a negative biased random forest, where the negative instances have a major proportion of the training data in each of the tree in the forest. Secondly, we develop a variant of random forest in which each tree is biased towards the positive class to classify the data set, where a major vote is provided for prediction. In the experimental test, we compared our method with other existing methods on the real data sets, and the results demonstrate the significative performance improvement of our method in terms of the area under the ROC curve(AUC).

This paper presents a system for classifying e-mails into two categories, legitimate and fraudulent. This classifier system is based on the serial application of three filters: a Bayesian filter that classifies the textual content of e-mails, a rule- based filter that classifies the non grammatical content of e-mails and, finally, a filter based on an emulator of fictitious accesses which classifies the responses from websites referenced by links contained in e-mails. This system is based on an approach that is hybrid, because it uses different classification methods, and also integrated, because it takes into account all kind of data and information contained in e-mails. This approach aims to provide an effective and efficient classification. The system first applies fast and reliable classification methods, and only when the resulting classification decision is imprecise does the system apply more complex analysis and classification methods.

In combination with efficient kernel-base learning machines such as Support Vector Machine (SVM), string kernels have proven to be significantly effective in a wide range of research areas ( bioinformatics, text analysis, voice analysis). Many of the string kernels proposed so far take advantage of simpler kernels such as trivial comparison of characters and/or substrings, and are classified into two classes: the string kernel which takes advantage of positional information of characters/substrings in their parent strings, and the string kernel which does not. Although the positive semidefiniteness of kernels is a critical prerequisite for learning machines to work properly, a little has been known about the positive semidefiniteness of the position-aware string kernel. The present paper is the first paper that presents easily checkable sufficient conditions for the positive semidefiniteness of a certain useful subclass of the position-aware string kernel: the similarity/matching of pairs of characters/substrings is evaluated with weights determined according to (the differences in the positions of characters/substrings). Such string kernels have been studied in the literature but insufficiently. In addition, by presenting a general framework for converting positive semidefinite string kernels into those for richer data structures such as trees and graphs, we generalize our results.

The paper aims to provide for the improvement of software estimation research through a new regression model. The study design of the paper is organized as follows. Evaluation of estimation methods based on historical data sets requires that these data sets be representative for current or future projects. For that reason the data set for software cost estimation model the International Software Benchmarking Standards Group (ISBSG) data set Release 9 is used. The data set records true project values in the real world, and can be used to extract information to predict new projects cost in terms of effort. As estimation method regression models are used. The main contribution of this study is the new cost production function that is used to obtain software cost estimation. The new proposed cost estimation function performance is compared with related work in the literature. In the study same calibration on the production function is made in order to obtain maximum performance. There is some important discussion on how the results can be improved and how they can be applied to other estimation models and datasets.

We review a new form of self-organizing map which is based on a nonlinear projection of latent points into data space, identical to that performed in the Generative Topographic Mapping (GTM) [1]. But whereas the GTM is an extension of a mixture of experts, this model is an extension of a product of experts [6]. We show visualisation and clustering results on a data set composed of video data of lips uttering 5 Korean vowels and show that the new mapping achieves better results than the standard Self-Organizing Map.

In this paper, the uplink multiuser code division multiple access (CDMA) communication system model is described in the form of space-time domain through antenna array and multipath fading expression. Novel suitable neural network technique is proposed as an effective signal processing method for the receiver of such an uplink multiuser CDMA system. By the appropriate choice of the channel state information for the neural network parameters, the neural network can collectively resolve the effects of both the inter-symbol interference due to the multipath fading channel and the multiple access interference in the receiver of the uplink multiuser CDMA communication system. The dynamics of the proposed neural network receiver for the uplink multiuser CDMA communication system is also studied.

There are many variants of the original self-organizing neural map algorithm proposed by Kohonen. One of the most recent is the Evolving Tree, a tree-shaped self-organizing network which has many interesting characteristics. This network builds a tree structure splitting the input dataset during learning. This paper presents a speed-up modification of the original training algorithm useful when the Evolving Tree network is used with complex data as images or video. After a measurement of the effectiveness an application of the modified algorithm in image segmentation is presented.