Paul Rosero-Montalvo

Interactive Data Visualization Using Dimensionality Reduction and Dissimilarity-Based Representations

This work describes a new model for interactive data visualization followed from a dimensionality-reduction (DR)-based approach. Particularly, the mixture of the resulting spaces of DR methods is considered, which is carried out by a weighted sum. For the sake of user interaction, corresponding weighting factors are given via an intuitive color-based interface. Also, to depict the DR outcomes while showing information about the input high-dimensional data space, the low-dimensional representations reached by the mixture is conveyed using scatter plots enhanced with an interactive data-driven visualization. In this connection, a constrained dissimilarity approach define the graph to be drawn on the scatter plot.

Interactive visualization methodology of high-dimensional data with a color-based model for dimensionality reduction

Nowadays, a consequence of data overload is that worlds technology capacity to collect, communicate, and store large volumes of data is increasing faster than human analysis skills. Such an issue has motivated the development of graphic ways to visually represent and analyze high-dimensional data. Particularly, in this work, we propose a graphical interface that allow the combination of dimensionality reduction (DR) methods using a chromatic model to make data visualization more intelligible for humans. This interface is designed for an easy and interactive use, so that input parameters are given by the user via the selection of RGB values inside a given surface. Proposed interface enables (even non-expert) users to intuitively either select a concrete DR method or carry out a mixture of methods. Experimental results proves the usability of our interface making the selection or configuration of a DR-based visualization an intuitive and interactive task for the user.

Developments on Solutions of the Normalized-Cut-Clustering Problem Without Eigenvectors

Normalized-cut clustering (NCC) is a benchmark graph-based approach for unsupervised data analysis. Since its traditional formulation is a quadratic form subject to orthogonality conditions, it is often solved within an eigenvector-based framework. Nonetheless, in some cases the calculation of eigenvectors is prohibitive or unfeasible due to the involved computational cost – for instance, when dealing with high dimensional data. In this work, we present an overview of recent developments on approaches to solve the NCC problem with no requiring the calculation of eigenvectors. Particularly, heuristic-search and quadratic-formulation-based approaches are studied. Such approaches are elegantly deduced and explained, as well as simple ways to implement them are provided.

Human-sitting-pose detection using data classification and dimensionality reduction

The research area of sitting-pose analysis allows for preventing a range of physical health problems mainly physical. Despite that different systems have been proposed for sitting-pose detection, some open issues are still to be dealt with, such as: Cost, computational load, accuracy, portability, and among others. In this work, we present an alternative approach based on a sensor network to acquire the position-related variables and machine learning techniques, namely dimensionality reduction (DR) and classification. Since the information acquired by sensors is high-dimensional and therefore it might not be saved into embedded system memory, a DR stage based on principal component analysis (PCA) is performed. Subsequently, the automatic posed detection is carried out by the k-nearest neighbors (KNN) classifier. As a result, regarding using the whole data set, the computational cost is decreased by 33 % as well as the data reading is reduced by 10 ms. Then, sitting-pose detection task takes 26 ms, and reaches 75% of accuracy in a 4-trial experiment.

Intelligence in Embedded Systems: Overview and Applications

The use of electronic systems and devices has become widely spread and is reaching several fields as well as indispensable for many daily activities. Such systems and devices (here termed embedded systems) are aiming at improving human beings’ quality of life. To do so, they typically acquire users’ data to adjust themselves to different needs and environments in an adequate fashion. Consequently, they are connected to data networks to share this information and find elements that allow them to make the appropriate decisions. Then, for practical usage, their computational capabilities should be optimized to avoid issues such as: resources saturation (mainly memory and battery). In this line, machine learning offers a wide range of techniques and tools to incorporate “intelligence” into embedded systems, enabling them to make decisions by themselves. This paper reviews different data storage techniques along with machine learning algorithms for embedded systems. Its main focus is on techniques and applications (with special interest in Internet of Things) reported in literature about data analysis criteria to make decisions.

Intelligent System of Squat Analysis Exercise to Prevent Back Injuries

The sports ergonomics study allows a bio-mechanical analysis in order to evaluate the impact produced by different muscle conditioning exercises such as the squat. This exercise, if carried out in an erroneous way, it can cause lumbar injuries. The present electronic system acquire the data of the Smith bar and the back by means of accelerometer sensors. This is done in order to implement an intelligent algorithm that allows to recognize if the athlete performs the exercise properly. For this, a stage of prototypes selection and a comparison of classification algorithms (CA) is carried out. Finally, a quantitative measure of equilibrium between both criteria is established for its proper selection. As a result, the k-Nearest Neighbors algorithm with k = 5 achieves a 96% performance and a 50% training matrix reduction.

A Novel Color-Based Data Visualization Approach Using a Circular Interaction Model and Dimensionality Reduction

Dimensionality reduction (DR) methods are able to produce low-dimensional representations of an input data sets which may become intelligible for human perception. Nonetheless, most existing DR approaches lack the ability to naturally provide the users with the faculty of controlability and interactivity. In this connection, data visualization (DataVis) results in an ideal complement. This work presents an integration of DR and DataVis through a new approach for data visualization based on a mixture of DR resultant representations while using visualization principle. Particularly, the mixture is done through a weighted sum, whose weighting factors are defined by the user through a novel interface. The interface’s concept relies on the combination of the color-based and geometrical perception in a circular framework so that the users may have a at hand several indicators (shape, color, surface size) to make a decision on a specific data representation. Besides, pairwise similarities are plotted as a non-weighted graph to include a graphic notion of the structure of input data. Therefore, the proposed visualization approach enables the user to interactively combine DR methods, while providing information about the structure of original data, making then the selection of a DR scheme more intuitive.

Cardiac Pulse Modeling Using a Modified van der Pol Oscillator and Genetic Algorithms

This paper proposes an approach for modeling cardiac pulses from electrocardiographic signals (ECG). A modified van der Pol oscillator model (mvP) is analyzed, which, under a proper configuration, is capable of describing action potentials, and, therefore, it can be adapted for modeling a normal cardiac pulse. Adequate parameters of the mvP system response are estimated using non-linear dynamics methods, like dynamic time warping (DTW). In order to represent an adaptive response for each individual heartbeat, a parameter tuning optimization method is applied which is based on a genetic algorithm that generates responses that morphologically resemble real ECG. This feature is particularly relevant since heartbeats have intrinsically strong variability in terms of both shape and length. Experiments are performed over real ECG from MIT-BIH arrhythmias database. The application of the optimization process shows that the mvP oscillator can be used properly to model the ideal cardiac rate pulse.

Angle-Based Model for Interactive Dimensionality Reduction and Data Visualization

In recent times, an undeniable fact is that the amount of data available has increased dramatically due mainly to the advance of new technologies allowing for storage and communication of enormous volumes of information. In consequence, there is an important need for finding the relevant information within the raw data through the application of novel data visualization techniques that permit the correct manipulation of data. This issue has motivated the development of graphic forms for visually representing and analyzing high-dimensional data. Particularly, in this work, we propose a graphical approach, which, allows the combination of dimensionality reduction (DR) methods using an angle-based model, making the data visualization more intelligible. Such approach is designed for a readily use, so that the input parameters are interactively given by the user within a user-friendly environment. The proposed approach enables users (even those being non-experts) to intuitively select a particular DR method or perform a mixture of methods. The experimental results prove that the interactive manipulation enabled by the here-proposed model-due to its ability of displaying a variety of embedded spaces-makes the task of selecting a embedded space simpler and more adequately fitted for a specific need.

Automatic Motion Segmentation via a Cumulative Kernel Representation and Spectral Clustering

Dynamic or time-varying data analysis is of great interest in emerging and challenging research on automation and machine learning topics. In particular, motion segmentation is a key stage in the design of dynamic data analysis systems. Despite several studies have addressed this issue, there still does not exist a final solution highly compatible with subsequent clustering/classification tasks. In this work, we propose a motion segmentation compatible with kernel spectral clustering (KSC), here termed KSC-MS, which is based on multiple kernel learning and variable ranking approaches. Proposed KSC-MS is able to automatically segment movements within a dynamic framework while providing robustness to noisy environments.