Research
Carlos Torrez, C. López-Del-Alamo,"Classification of People who Suffer Schizophrenia and Healthy People by EEG Signals using Deep Learning , International Journal of Advanced Computer Science and Applications(IJACSA). 2019
Abstract—Abstract: More than 21 million people worldwide suffer from schizophrenia. This serious mental disorder exposes people to stigmatization, discrimination, and violation of their human rights. Different works on classification and diagnosis of mental illnesses use electroencephalogram signals (EEG) because it reflects brain functioning, and how these diseases affect it. Due to the information provided by the EEG signals and the perfor-mance demonstrated by Deep Learning algorithms, the present work proposes a model for the classification of schizophrenic and healthy people through EEG signals using Deep Learning methods. Considering the properties of an EEG, high-dimensional and multichannel, we applied the Pearson Correlation Coefficient (PCC) to represent the relations between the channels, this way instead of using the large amount of data that an EEG provides, we used a shorter matrix as an input of a Convolutional Neural Network (CNN). Finally, results demonstrated that the proposed EEG-based classification model achieved Accuracy, Specificity, and Sensitivity of 90%, 90%, and 90%, respectively.
M. Paco Ramos, C. López-Del-Alamo, R. Alfonte "Forecasting of Meteorological Weather Time Series Through a Feature Vector Based on Correlation , International Conference on Computer Analysis of Images and Patterns(CAIP 2019), Italia
Abstract—Nowadays, the impacts of climate change are harming many countries around the world. For this reason, the scientific community is interested in improving methods to forecast weather events, so it is possible to avoid people from being injured. One important thing in the development of time series forecasting methods is to consider the set of values over time that facilitates the prediction of future value. In this sense, we propose a new feature vector based on the correlation and autocorrelation functions. These measures reflect how the observations of a time series are related to each other. Then, univariate forecasting is performed using Multilayer Perceptron (MLP) and Long Short-Term Memory (LSTM) deep neural network. Finally, we compared the new model with linear and non-linear models. Reported results exhibit that MLP and LSTM models using the proposed feature vector, they show promising results for univariate forecasting. We tested our method on a real-world dataset from the Fisher weather station (Harvard Forest).
Ali Erkan, John Barr, Tony Clear, Cruz Izu, Cristian López, Hanan Mohammend, Mahadev Nadimpalli. "Developing a Holistic Understanding of Systems and Algorithms through Research Papers" , Innovation and Technology in Computer Science Education (ITiCSE-WGR '17), Italia
Abstract—Even though a computer science degree is unavoidably broken into semesters and courses, we always hope that our students form a holistic picture of the discipline by the time they graduate. Yet as educators, we do not have too many opportunities to make this point front and center for an extended period of time. This report es a well-defined portion of this problem: revealing conceptual connections between algorithmic courses (such as Discrete Math, Data Structures, Algorithms) and systems oriented courses (such as Organization, Computer Networks, Operating Systems, and Hardware) through the use of research papers. In particular, we provide a pedagogical framework as well as a set of carefully selected papers to crosscut our disciplinary space in a way that is orthogonal to conventional course design. This framework includes a paper taxonomy, strategies for covering topics that students are yet to encounter in upper level courses, strategies for reading and writing technical papers, three modules (one each for operating systems, networks, and architecture) that can be integrated into standard systems courses, and a new (optional) course template as a container for all of the listed elements. Since we have already tried these ideas once at the institution of the two leading authors, our report is rich with scaffolding suggestions as well.
L. Castillo, G. Dávila, C. López. "A New Graph-Based Approach for Document SimilarityUsing Concepts of Non-Rigid Shapes , The Seventh International Conference on Advances in Information Mining and Managemen (IMMM 2017), Italia- Venecia
Abstract—Most methods used to compare text documentsare based on the space vector model; however, this modeldoes not capture the relations between words, which isconsidered necessary to make better comparisons. In thisresearch, we propose a method based on the creation ofgraphs to get semantic relations between words and weadapt algorithms of the theory of non-rigid 3D modelanalysis.
Llerena J. López C. "Non-rigid 3D Shape Classification based on Convolutional Neural Networks" , Computational Intelligence (LA-CCI), 2017 IEEE Latin American Conference, 2017
Abstract—Over the years, the scientific interest towards 3D models analysis has become more popular. Problems such as classification, retrieval and matching are studied with the idea to offer robust solutions. This paper introduces a 3D object classification method for non-rigid shapes, based on the detection of key points, the use of spectral descriptors and deep learning techniques. We adopt an approach of converting the models into a “spectral image”. By extracting interest points and calculating three types of spectral descriptors (HKS, WKS and GISIF), we generate a three-channel input to a convolutional neural network. This CNN is trained to automatically learn features such as topology of 3D models. The results are evaluated and analyzed using the Non-Rigid Classification Benchmark SHREC 2011. Our proposal shows promising results in classification tasks compared to other meth- ods, and also it is robust under several types of transformations.
Arbieto C. , Huillca J., Ocsa A. , Coronado R., Lopez C, "Time Series Mining using Approximate Nearest Neighbor Search on GPUs". , 2017
Abstract— Motif discovery and outlier detection in time series databases is of prime impor- tance and a major challenge in many areas, so exists a necessity of real-time algorithms. For this challenges, single CPU architectures impose limits on performance to deliver this kinds of solutions. Resorting to GPU programming is one approach to overcome these performance limitations. In this work, we discuss various time series motif discovery and outlier detection algorithms in order to select only those with highly parallelizable algorithm structure. More important, we present a comparative study among paralleliz- able exact and approximate algorithm to find motifs and outliers in very large time series databases. We use a NVIDIA CUDA framework to create a massively parallel implemen- tation of the analyzed techniques. Experimental studies allow us to run experiments on large real and synthetic datasets thanks to the highly CUDA parallel implementation. Our performance evaluation on GeForce GTX 1080 GPU, result in average runtime speedups of 10 (15) using the MK (Random Projection) and Local Outlier Factor algorithms. The runtime speedups of CUDA is better than to those of Parallel MK and Parallel Random Projection running on 8 CPU cores of high-performance workstation.
Ocsa A., Mango J., Coronado R., Quispe O., Arbieto C., López C, "Approximate nearest neighbors by deep hashing on large-scale search: Comparison of representations and retrieval performance Autoencoders", Computational Intelligence (LA-CCI), 2017 IEEE Latin American Conference, 2017
Abstract—The growing volume of data and its increasing complexity require even more efficient and faster information retrieval techniques. Approximate nearest neighbor search algorithms based on hashing were proposed to query high-dimensional datasets due to its fast retrieval speed and low storage cost. Recent studies, promote the use of Convolutional Neural Network (CNN) with hashing techniques to improve the search accuracy. However, there are challenges to solve in order to find a practical and efficient solution to index CNN features, such as the need for heavy training process to achieve accurate query results and the critical dependency on data-parameters. In this work we execute exhaustive experiments in order to compare recent methods that are able to produces a better representation of the data space with a less computational cost for a better accuracy by computing the best data-parameters values for optimal sub-space projection exploring the correlations among CNN features attributes using fractal theory. We give an overview of these different techniques and present our comparative experimental for data representation and retrieval performance.
Ocsa A., Huillca J., Lopez C., "On semantic solutions for efficient approximate similarity search on large-scale dataset.". Springer International Publishing, 2017
Abstract— Approximate similarity search algorithms based on hashing were proposed to query high-dimensional datasets due to its fast retrieval speed and low storage cost. Recent studies, promote the use of Convolutional Neural Network (CNN) with hashing techniques to improve the search accuracy. However, there are challenges to solve in order to find a practical and efficient solution to index CNN features, such as the need for heavy training process to achieve accurate query results and the critical dependency on data-parameters. Aiming to overcome these issues, we propose a new method for scalable similarity search, i.e., Deep frActal based Hashing (DAsH), by computing the best data-parameters values for optimal sub-space projection exploring the correlations among CNN features attributes using fractal theory. Moreover, inspired by recent advances in CNNs, we use not only activations of lower layers which are more general-purpose but also previous knowledge of the semantic data on the latest CNN layer to improve the search accuracy. Thus, our method produces a better representation of the data space with a less computational cost for a better accuracy. This significant gain in speed and accuracy allows us to evaluate the framework on a large, realistic, and challenging set of datasets.
Alfonte R., López C., Llerena Jan, Cuadros Ana María, "Characterization of Climatological Time Series using Autoencoders" , Computational Intelligence (LA-CCI), 2017 IEEE Latin American Conference, 2017
Abstract—Common problems in climatological time series data are high dimensionality, correlation between the sequential values and noise due to calibration of meteorological stations influencing dramatically in the quality of clustering, classification, climate pattern finding and data processing. One way to deal with this problem is through feature extraction technique. In order to extract features from large climatological time series data, we propose a feature extraction method based on autoencoder neural network (AUTOE). As a first step, time series is standardized. Then, different architectures of autoencoder is applied on it to reduce dimensionality. Finally, k-means clustering algorithm are used to evaluate them through quality measures. As a result, autoencoder performs well and is competitive with other feature extraction techniques over Synthetic Control Chart Time Series. Index Terms—Dimensionality reduction, autoencoder, time series.
C. López, L. Arnaldo, L. Fuentes, "Efficient approach for interest points detection in non-rigid shapes", XLI Latin American Computing Conference, pp. 62-69, 2015.
Abstract—Due to the increasing amount of data and the reduction of costs in 3D data acquisition devices, there has been a growing interest, in developing efficient and robust feature extraction algorithms for 3D shapes, invariants to isometric, topological and noise changes, among others. One of the key tasks for feature extraction in 3D shapes is the interest points detection; where interest points are salient structures, which can be used, instead of the whole object. In this research, we present a new approach to detect interest points in 3D shapes by analysing the triangles that compose the mesh which represent the shape, in different way to other algorithms more complex such as Harris 3D or HKS. Our results and experiments of repeatability, confirm that our algorithm is stable and robust, in addition, the computational complexity is O(n log n), where n represents the number of faces of the mesh. Video: Click aquí
J. Hurtado, M. Coaquira, C. López, "3D Mesh Interest Point Detection using GISIFs and Heat Diffusion", XLI Latin American Computing Conference, pp. 1-40, 2015.
Abstract—To facilitate processing of 3D objects is common to use high-level representations. The interest points are one of them. An interest point should possess a distinctive feature regarding its locality and should be stable in different instances of the object. This article proposes a descriptor based on symmetry (GISIFs) and heat diffusion (HKS). From this features, we select a set of representative points. The GISIFs referenced in this article has not been used to extract local features. We compare our results with the results of other techniques, which make up the state of the art in interest point detection. We use a benchmark that
evaluates the accuracy of the selected points with respect to an ideal set of interest points.
C. López, L. Fuentes, L. Arnaldo, "Parallelization of the Algorithm k-means applied in image Segmentation", International Journal of Computer Applications, Volume 88 - No. 17, February 2014.
Abstract—Algorithm k-means is useful for grouping operations; however, when is applied to large amounts of data, its computational cost is high. This research propose an optimization of k-means algorithmby using parallelization techniques and synchronization, which is applied to image segmentation. In the results obtained, the parallel k-means algorithm, improvement 50% to the algorithm sequential k-means.
C. López, L. Arnaldo, L. Fuentes, W. Ramos, "Agrupamiento por similitud de imágenes mediante Arbol de Expansión Mínima y Soft Heap", XLI Latin American Computing Conference, pp. 1-7, 2013.
Abstract—Due to the advancement of computing and the power of the new hardware, more economical, it is now feasible to have thousands of images which can be analyzed to allow classification for its shape and/or color. Furthermore, techniques and efficiency of the classification depends on the characteristics to be obtained of images in order to compare and classify them according to their similarity. Some images, such as model cars, planes and boats, can be discriminated by their shape. However, other images such as butterfly species where the shape is similar, the color plays an important role in the discrimination task. In this research we propose a novel approach to extract distinctive features of images by combining the HSV color model and wavelets filters. Furthermore, we investigate the best combination of features color and form. Experiments have shown improved performance by combining the HSV color model with Gabor wavelets.
L. Fuentes, L. Arnaldo, C. López "A novel approach for image feature extraction using HSV model color and filters wavelets.", XLI Latin American Computing Conference, pp. 1-7, 2013.
Abstract—En esta investigación, se propone un nuevo enfoque, para extraer características distintivas de imágenes, basado en el modelo de color HSV y filtros wavelets, con la finalidad de agrupar imágenes similares entre sí por ejemplo mariposas de la misma especie. Ademàs se investiga la mejor combinación de características de color y forma. Los experimentos han demostrado un mejor rendimiento en la combinación de color con el filtro de Gabor.
H. Chuctaya, C. Portugal, C. Beltran, J. Gutierrez, C. Lopez, Y. Tupac, "M-CBIR: A medical content-based image retrieval system using metric data-structures", 30th International Conference of the Chilean Computer Science Society, pp 135 - 141, 2011.
Abstract—This work is focused on the modelling and development of a CBIR (Content-based image retrieval) system applied to the recovery of digital medical images of a human body, denominated M-CBIR. This model is composed on two methodologies: features extraction techniques and metric data structures. When this set of techniques is applied to the search of different human body regions, it can retrieve the most relevant similar images to a query image. A real database of medical images composed of 772 medical studies was used to compare the robustness of the extraction techniques and evaluate the performance of the system, through four different extractors. The objective of this work will result in a digital atlas of human body for medical radiological center. Finally, analysis and conclusions are also discussed.