José M. Leiva-Murillo

Maximization of Mutual Information for Supervised Linear Feature Extraction

In this paper, we present a novel scheme for linear feature extraction in classification. The method is based on the maximization of the mutual information (MI) between the features extracted and the classes. The sum of the MI corresponding to each of the features is taken as an heuristic that approximates the MI of the whole output vector. Then, a component-by-component gradient-ascent method is proposed for the maximization of the MI, similar to the gradient-based entropy optimization used in independent component analysis (ICA). The simulation results show that not only is the method competitive when compared to existing supervised feature extraction methods in all cases studied, but it also remarkably outperform them when the data are characterized by strongly nonlinear boundaries between classes.

Distinguishing the relevant features of frequent suicide attempters.

BACKGROUND In spite of the high prevalence of suicide behaviours and the magnitude of the resultant burden, little is known about why individuals reattempt. We aim to investigate the relationships between clinical risk factors and the repetition of suicidal attempts. METHODS 1349 suicide attempters were consecutively recruited in the Emergency Room (ER) of two academic hospitals in France and Spain. Patients were extensively assessed and demographic and clinical data obtained. Data mining was used to determine the minimal number of variables that blinded the rest in relation to the number of suicide attempts. Using this set, a probabilistic graph ranking relationships with the target variable was constructed. RESULTS The most common diagnoses among suicide attempters were affective disorders, followed by anxiety disorders. Risk of frequent suicide attempt was highest among middle-aged subjects, and diminished progressively with advancing age of onset at first attempt. Anxiety disorders significantly increased the risk of presenting frequent suicide attempts. Pathway analysis also indicated that frequent suicide attempts were linked to greater odds for alcohol and substance abuse disorders and more intensive treatment. CONCLUSIONS Novel statistical methods found several clinical features that were associated with a history of frequent suicide attempts. The identified pathways may promote new hypothesis-driven studies of suicide attempts and preventive strategies.

Diagnostic stability and evolution of bipolar disorder in clinical practice: a prospective cohort study

Objective:  To evaluate the long‐term stability of International Classification of Diseases‐10th revision bipolar affective disorder (BD) in multiple settings.

Multitask Remote Sensing Data Classification

Many remote sensing data processing problems are inherently constituted by several tasks that can be solved either individually or jointly. For instance, each image in a multitemporal classification setting could be taken as an individual task. Here, the relation to previous acquisitions should be properly considered because of the nonstationary behavior of temporal, spatial, and angular image features which gives rise to distribution changes. This phenomenon is known as covariate shift. Additionally, when labeled data are scarce or expensive to obtain, the small sample-set problem arises, which makes solving the problems independently in each domain difficult. Multitask learning (MTL) aims at jointly solving a set of prediction problems by sharing information across tasks. This paper introduces MTL in remote sensing data classification. The proposed methods alleviate the data set shift by imposing cross-information in the classifiers through matrix regularization. We consider the support vector machine (SVM) as the core learner and two different regularization schemes: 1) the inclusion of relational operators between tasks and 2) the pairwise Euclidean distance of the predictors in the Hilbert space. These methods rely on simple and intuitive modifications of the kernel used in the standard SVM. Experiments are conducted in three challenging remote sensing problems: cloud screening from multispectral images, land-mine detection using radar data, and multitemporal and multisource image classification. The pairwise method consistently outperforms standard independent and aggregate approaches by about +2% to 4% in all problems at no additional cost. Also, the solutions found give us information about the distribution shift among tasks.

Evaluating the Internationalization Success of Companies Through a Hybrid Grouping Harmony Search—Extreme Learning Machine Approach

The internationalization of a company is widely understood as the corporative strategy for growing through external markets. It usually embodies a hard process, which affects diverse activities of the value chain and impacts on the organizational structure of the company. There is not a general model for a successful international company, so the success of an internationalization procedure must be estimated based on different variables addressing the status, strategy and market characteristics of the company at hand. This paper presents a novel hybrid soft-computing approach for evaluating the internationalization success of a company based on existing past data. Specifically, we propose a hybrid algorithm composed by a grouping-based harmony search (HS) approach and an extreme learning machine (ELM) ensemble. The proposed hybrid scheme further incorporates a feature selection method, which is obtained by means of a given group in the HS encoding format, whereas the ELM ensemble renders the final accuracy metric of the model. Practical results for the proposed hybrid technique are obtained in a real application based on the exporting success of Spanish manufacturing companies, which are shown to be satisfactory in comparison with alternative state-of-the-art techniques.

Supervised-PCA and SVM classifiers for object detection in infrared images

We tackle the problem of detecting sources of combustion in high definition multispectral medium wavelength infrared (MWIR) (3-5 /spl mu/m) images. We present a novel approach to this problem consisting of processing the images block-wise using a new technique that we call supervised principal component analysis (SPCA) to get the components of these blocks. This outperforms state-of-the-art methods with a significant reduction in the complexity of the whole scheme. As a classifier, we propose the use of a support vector machine (SVM) comparing the results from both its novelty-detection and binary non-linear versions. High performance is achieved from a small set of components.

Algorithms for maximum-likelihood bandwidth selection in kernel density estimators

In machine learning and statistics, kernel density estimators are rarely used on multivariate data due to the difficulty of finding an appropriate kernel bandwidth to overcome overfitting. However, the recent advances on information-theoretic learning have revived the interest on these models. With this motivation, in this paper we revisit the classical statistical problem of data-driven bandwidth selection by cross-validation maximum likelihood for Gaussian kernels. We find a solution to the optimization problem under both the spherical and the general case where a full covariance matrix is considered for the kernel. The fixed-point algorithms proposed in this paper obtain the maximum likelihood bandwidth in few iterations, without performing an exhaustive bandwidth search, which is unfeasible in the multivariate case. The convergence of the methods proposed is proved. A set of classification experiments are performed to prove the usefulness of the obtained models in pattern recognition.

Nucleotide variation in central nervous system genes among male suicide attempters

Despite marked morbidity and mortality associated with suicidal behavior, accurate identification of individuals at risk remains elusive. The goal of this study is to identify a model based on single nucleotide polymorphisms (SNPs) that discriminates between suicide attempters and non‐attempters using data mining strategies. We examined functional SNPs (n = 840) of 312 brain function and development genes using data mining techniques. Two hundred seventy‐seven male psychiatric patients aged 18 years or older were recruited at a University hospital psychiatric emergency room or psychiatric short stay unit. The main outcome measure was history of suicide attempts. Three SNPs of three genes (rs10944288, HTR1E; hCV8953491, GABRP; and rs707216, ACTN2) correctly classified 67% of male suicide attempters and non‐attempters (0.50 sensitivity, 0.82 specificity, positive likelihood ratio = 2.80, negative likelihood ratio = 1.64). The OR for the combined three SNPs was 4.60 (95% CI: 1.31–16.10). The models accuracy suggests that in the future similar methodologies may generate simple genetic tests with diagnostic utility in identification of suicide attempters. This strategy may uncover new pathophysiological pathways regarding the neurobiology of suicidal acts.

Technical Section: Automated generation and visualization of picture-logic puzzles

A picture-logic puzzle is a game that takes the form of an NxM grid, with numbers situated on the left of its rows and on the top of its columns, which give the clues for solving the puzzle. These puzzles have gained popularity in the last years all over the world, and there are companies involved in the commercialization of products related to them, mainly magazines, on-line puzzles via the web and games for mobile phones. The main problem with selling picture-logic puzzles is the need to generate a large number of picture-logic puzzles per month, or even per week, so automating the generation process of such puzzles is an important task. In this paper we propose algorithms for the automated generation of picture-logic puzzles (both black and white and color puzzles), starting from any digital RGB color image. We also present a visualization interface which provides several useful tools for finishing the puzzle generation process, like the possibility of changing colors and displaying the process of puzzle solution. We show the performance of the algorithms for generating picture-logic puzzles with several examples.

A Gaussian Mixture Based Maximization of Mutual Information for Supervised Feature Extraction

In this paper, we propose a new method for linear feature extraction and dimensionality reduction for classification problems. The method is based on the maximization of the Mutual Information (MI) between the resulting features and the classes. A Gaussian Mixture is used for modelling the distribution of the data. By means of this model, the entropy of the data is then estimated, and so the MI at the output. A gradient descent algorithm is provided for its optimization. Some experiments are provided in which the method is compared with other popular linear feature extractors.