Departamento de Ciencias de la Computación

Permanent URI for this community

https://hdl.handle.net/20.500.12590/17959

Browse

Now showing 1 - 20 of 86

A categorization of simultaneous localization and mapping knowledge for mobile robots
(Universidad Católica San Pablo, 2020) Cornejo Lupa, Maria Alejandra; Ticona Herrera, Regina Paola
Autonomous robots are playing important roles in academic, technologi-cal, and scientific activities. Thus, their behavior is getting more complex. The main tasks of autonomous robots include mapping an environment and localize themselves. These tasks comprise the Simultaneous Localization and Mapping (SLAM) problem. Representation of the SLAM knowledge (e.g., robot charac-teristics, environment information, mapping and location information), with a standard and well-defined model, provides the base to develop efficient and interoperable solutions. However, as far as we know, there is not a common classification of such knowledge. Many existing works based on Semantic Web, have formulated ontologies to model information related to only some SLAM aspects, without a standard arrangement. In this work, we propose a category-zation of the knowledge managed in SLAM, based on existing ontologies and SLAM principles. We also classify recent and popular ontologies according to our proposed categories and highlight the lessons to learn from existing solu- tions. Showing the neccesity to develop a complete SLAM ontology in mobile robots.
A deep learning approach for sentiment analysis in Spanish Tweets
(Springer Verlag, 2018) Vizcarra Aguilar, Gerson; Mauricio, Antoni; Mauricio, Leonidas
Sentiment Analysis at Document Level is a well-known problem in Natural Language Processing (NLP), being considered as a reference in NLP, over which new architectures and models are tested in order to compare metrics that are also referents in other issues. This problem has been solved in good enough terms for English language, but its metrics are still quite low in other languages. In addition, architectures which are successful in a language do not necessarily works in another. In the case of Spanish, data quantity and quality become a problem during data preparation and architecture design, due to the few labeled data available including not-textual elements (like emoticons or expressions). This work presents an approach to solve the sentiment analysis problem in Spanish tweets and compares it with the state of art. To do so, a preprocessing algorithm is performed based on interpretation of colloquial expressions and emoticons, and trivial words elimination. Processed sentences turn into matrices using the 3 most successful methods of word embeddings (GloVe, FastText and Word2Vec), then the 3 matrices merge into a 3-channels matrix which is used to feed our CNN-based model. The proposed architecture uses parallel convolution layers as k-grams, by this way the value of each word and their contexts are weighted, to predict the sentiment polarity among 4 possible classes. After several tests, the optimal tuple which improves the accuracy were <1, 2>. Finally, our model presents %61.58 and %71.14 of accuracy in InterTASS and General Corpus respectively. © Springer Nature Switzerland AG 2018.
A graph-based approach for transcribing ancient documents
(Springer Verlag, 2012) Meza Lovón, Graciela Lecireth
Over the last years, the interest in preserving digitally ancient documents has increased resulting in databases with a huge amount of image data. Most of these documents are not transcribed and thus querying operations are limited to basic searching. We propose a novel approach for transcribing historical documents and present results of our initial experiments. Our method divides a text-line image into frames and constructs a graph using the framed image. Then Dijkstra algorithm is applied to find the line transcription. Experiments show a character accuracy of 79.3%. © Springer-Verlag Berlin Heidelberg 2012.
A new boosting design of Support Vector Machine classifiers
(Elsevier, 2015) Mayhua López, Efraín; Gómez Verdejo, Vanessa; Figueiras Vidal, Aníbal
Boosting algorithms pay attention to the particular structure of the training data when learning, by means of iteratively emphasizing the importance of the training samples according to their difficulty for being correctly classified. If common kernel Support Vector Machines (SVMs) are used as basic learners to construct a Real AdaBoost ensemble, the resulting ensemble can be easily compacted into a monolithic architecture by simply combining the weights that correspond to the same kernels when they appear in different learners, avoiding to increase the operation computational effort for the above potential advantage. This way, the performance advantage that boosting provides can be obtained for monolithic SVMs, i.e., without paying in classification computational effort because many learners are needed. However, SVMs are both stable and strong, and their use for boosting requires to unstabilize and to weaken them. Yet previous attempts in this direction show a moderate success. In this paper, we propose a combination of a new and appropriately designed subsampling process and an SVM algorithm which permits sparsity control to solve the difficulties in boosting SVMs for obtaining improved performance designs. Experimental results support the effectiveness of the approach, not only in performance, but also in compactness of the resulting classifiers, as well as that combining both design ideas is needed to arrive to these advantageous designs. © 2014 Elsevier B.V.All rights reserved.
A New Improvement of Human Bodies Detection
(Institute of Electrical and Electronics Engineers Inc., 2016) Cervantes Jilaja, Claudia; Tejada Begazo, Maria; Patiño Escarcina, Raquel Esperanza; Barrios Aranibar, Dennis
The HOG method is applied in the detection of human bodies in an vertical position. However, when human bodies are in other positions, HOG method has several fails, another disadvantage of this method is the occlusion of human bodies by objects. This research presents a new improvement to HOG method with erosion morphological operator and cascade classifier (vector points describing the face of a person). The experiments show that the improved HOG method add erosion operator and classifier of faces (MFHD, Morphological Face HOG Detection) in 79.02% with respect the HOG method whose percentage is 64.46%. © 2015 IEEE.
A New Method for Static Video Summarization Using Local Descriptors and Video Temporal Segmentation
(Scopus, 2013) Cayllahua Cahuina, Edward; Camara Chavez, Guillermo
The continuous creation of digital video has caused an exponential growth of digital video content. To increase the usability of such large volume of videos, a lot of research has been made. Video summarization has been proposed to rapidly browse large video collections. To summarize any type of video, researchers have relied on visual features contained in frames. In order to extract these features, different techniques have used local or global descriptors. In this paper, we propose a method for static video summarization that can produce meaningful and informative video summaries. We perform an evaluation using over 100 videos in order to achieve a stronger position about the performance of local descriptors in semantic video summarization. Our experimental results show, with a confidence level of 99%, that our proposed method using local descriptors and temporal video segmentation produces better summaries than state of the art methods. We also demonstrate the importance of a more elaborate method for temporal video segmentation, improving the generation of summaries, achieving 10% improvement in accuracy. We also acknowledge a marginal importance of color information when using local descriptors to produce video summaries. © 2013 IEEE.
A new parallel training algorithm for optimum-path forest-based learning
(Springer Verlag, 2017) Culquicondor, Aldo; Castelo Fernández, Cesar; Paulo Papa, Joao
In this work, we present a new parallel-driven approach to speed up Optimum-Path Forest (OPF) training phase. In addition, we show how to make OPF up to five times faster for training using a simple parallel-friendly data structure, which can achieve the same accuracy results to the ones obtained by traditional OPF. To the best of our knowledge, we have not observed any work that attempted at parallelizing OPF to date, which turns out to be the main contribution of this paper. The experiments are carried out in four public datasets, showing the proposed approach maintains the trade-off between efficiency and effectiveness. © Springer International Publishing AG 2017.
A Semi-Automated Approach for Recognizing Moving Targets Using a Global Vision System
(Institute of Electrical and Electronics Engineers Inc., 2016) Ripas Mamani, Roger; Cervantes Jilaja, Claudia; Rosas Cuevas, Yessica; Patiño Escarcina, Raquel Esperanza; Barrios Aranibar, Dennis
Global vision system works with processes of sorting, recognition and identification through some external characteristics as: color, shape and size depending of specific targets. In this paper we propose a semi-automated approach to recognize the targets in moving, where first is performed the image calibration with respect to the lighting and then proceeds to recognize a variety of colors and sizes, through several channels of different color spaces in the processing of video sequences to recognize moving targets, using the proposed algorithm called Color Segmentation (Algorithm 1) to identify a variety of light and dark colors. After semi-automated process is performed the sorting or recognizing of the moving target, where is obtained the position (x, y) of central point and the size of the area (pixels) of the segmentation region. Tests were conducted in: the location of robots in a soccer robot environment (with 94.36% of accuracy) and chestnuts selection process (with 91.80% of accuracy), if the image needs to recognize more than five detections then it proceeds to add parallelism, i.e. add a thread for each segmented color, thus improving processing time. © 2016 IEEE.
Abnormal event detection in video using motion and appearance information
(Springer Verlag, 2018) Menejes Palomino, Neptalí; Cámara Chávez, Guillermo
This paper presents an approach for the detection and localization of abnormal events in pedestrian areas. The goal is to design a model to detect abnormal events in video sequences using motion and appearance information. Motion information is represented through the use of the velocity and acceleration of optical flow and the appearance information is represented by texture and optical flow gradient. Unlike literature methods, our proposed approach provides a general solution to detect both global and local abnormal events. Furthermore, in the detection stage, we propose a classification by local regions. Experimental results on UMN and UCSD datasets confirm that the detection accuracy of our method is comparable to state-of-the-art methods. © Springer International Publishing AG, part of Springer Nature 2018.
ACM/IEEE-CS computer science curricula 2013: Implementing the final report
(Association for Computing Machinery, 2014) Sahami, Mehran; Roach, Steve; Cuadros Vargas, Ernesto; Hawthorne, Elizabeth; Kumar, Amruth; LeBlanc, Richard; Reed, David; Seker, Remzi
For over 40 years, the ACM and IEEE-Computer Society have sponsored international curricular guidelines for undergraduate programs in computing. The rapid evolution and expansion of the computing field and the growing number of topics in computer science have made regular revision of curricular recommendations necessary. Thus, the Computing Curricula volumes are updated on an approximately 10-year cycle, with the aim of keeping curricula modern and relevant. The latest volume in the series, Computer Science Curricula 2013 (CS2013), is due for release in the Fall of 2013. This panel seeks to inform the SIGCSE community about the final version of the report, provide insight on interpreting the CS2013 guidelines, and give guidance regarding how the guidelines may be implemented at different institutions.
Actionable emotion detection in context-aware systems
(Universidad Católica San Pablo, 2018) Suni Lopez, Franci; Condori Fernandez, Nelly
Ensuring the quality of user experience is very important for increasing the acceptance likelihood of software applications, which can be affected by several contextual factors that can continuously change over time (e.g., emotional status of end-user). Due to these changes in the context, software continually needs to be (self-) adaptive for delivering software services that can satisfy user needs continuously. So far, online explicit user feedback has become one of the most used information sources for evaluating users’ satisfaction and discovering new requirements of a given software application. However, most of these online reviews are not authenticated, and they may not always be reliable. In order to complement this explicit feedback derived from user reviews, this research proposes an approach that exploits both physiological and contextual data to be used as main inputs for detecting actionable emotions. These actionable emotions, detected during the user interaction with context-aware software applications, can be used as implicit feedback for improving the adaptability of the software and quality of the user experience. The evaluation involved in total 23 subjects in three rounds of experiments. The results of this research support the idea that emotional data expressed by users when interacting with service-based applications can be used as implicit feedback.
AL-DDoS attack detection optimized with genetic algorithms
(Springer Verlag, 2018) Quequezana Buendia, Jan Camilo; Santisteban Pablo, Julio Omar
Application Layer DDoS (AL-DDoS) is a major danger for Internet information services, because these attacks are easily performed and implemented by attackers and are difficult to detect and stop using traditional firewalls. Managing to saturate physically and computationally the information services offered on the network. Directly harming legitimate users, to deal with this type of attacks in the network layer previous approaches propose to use a configurable statistical model and observed that when being optimized in various configuration parameters Using Genetic Algorithms was able to optimize the effectiveness to detect Network Layer DDoS (NL-DDoS), however this method is not enough to stop DDoS at the level of application because this level presents different characteristics, that is why we propose a new method Configurable and optimized for different scenarios of Attacks that effectively detect AL-DDoS. © Springer Nature Switzerland AG 2018.
An approach to real-coded quantum inspired evolutionary algorithm using particles filter
(Institute of Electrical and Electronics Engineers Inc., 2016) Chire Saire, Josimar Edinson; Túpac Valdivia, Yván Jesús
This work proposes, implements and evaluates the FP-QIEA-R model as a new quantum inspired evolutionary algorithm based on the concept of quantum superposition that allows the optimization process to be carried on with a smaller number of evaluations. This model is based on a QIEA-R, but instead of just using quantum individuals based on uniform probability density functions, where the update consists on change the width and mean of each pdf; this proposal uses a combined mechanism inspired in particle filter and multilinear regression, re-sampling and relative frequency with the QIEA-R to estimate the probability density functions in a better way. To evaluate this proposal, some experiments under benchmark functions are presented. The obtained statistics from the outcomes show the improved performance of this proposal optimizing numerical problems. © 2015 IEEE.
An Architecture for Computational Control of an Industrial Machine for Classifying Chestnuts
(Institute of Electrical and Electronics Engineers Inc., 2016) Álvarez Valera, Hernán; Bolivar Vilca, Edwin; Cervantes Jilaja, Claudia; Cuadros Zegarra, Emil; Barrios Aranibar, Dennis; Patiño Escarcina, Raquel Esperanza
Nowadays, the automation of industrial machines increase the productivity and efficiency of the mass production business. These machines are mainly composed of expensive electrical and mechanical modules to achieve companies production goals. However, many of these do not have information systems capable of providing the user relevant production data. In this work, the authors present an architecture for industrial automation machines of the chestnuts selection process, obtaining some features such as efficiency, effectiveness, high free configurability and low cost of maintenance and construction. This architecture is composed by three different modules: Mechanical components module, responsible of the physical parts management which interact directly with the products. Electrical components module, responsible for transferring data between the computational and mechanical layer through sensors and programs made. Finally, computational layer, responsible for two main tasks: process the necessary selection algorithms, sending the results to the electronic layer and run an information system, used to manage basic machine control operations, and generate the production data through the time. © 2015 IEEE.
An automatic emotion recognition system that uses the human body posture
(Universidad Católica San Pablo, 2021) Heredia Parillo, Juanpablo Andrew; Ticona Herrera, Regina Paola
Non-verbal communication is very present in our lives, but it can be interpreted in different ways according to many factors. With nonverbal gestures people can express explicit and implicit messages, which makes them important to understand. Computer vision methods for recognising body gestures and machine learning classification methods offer an opportunity to understand what people express with their bodies. This research work focuses on the emotions expressed by body gestures, particularly the posture. Thus, an automatic emotion recognition system from images is proposed, which uses a graph convolutional neural network to perform the classification. Generally, deep learning approach needs many training samples, but these are difficult to obtain for posture emotion recognition, thus, the proposed model trains under a meta-learning algorithm based on the “agnostic model”, which allows training with few examples. Only the meta-learning algorithm was tested, which demonstrated the adaptability and expands the applicability of the graph convolutional neural networks.
An enhanced triplet CNN based on body parts for person re-identificacion
(IEEE Computer Society, 2018) Durand Espinoza, Jonathan; Camara Chavez, Guillermo; Hinojosa Torres, Geraldine
Person re-identificacion consists of reidentificating person through a set of images that is taken by different camera views. Despite recent advances in this field, this problem still remains a challenge due to partial occlusions, changes in illumination, variation in human body poses. In this paper, we present an enhanced Triplet CNN based on body-parts for person re-identification (AETCNN). We design a new model able to learn local body-part features and integrate them to produce the final feature representation of each input person. In addition, to avoid over-fitting due to the small size of the dataset, we propose an improvement in triplet assignment to speed up the convergence and improve performance. Experiments show that our approach achieves very promising results in (CUHK01) dataset and we advance state of the art, improving most of the results of the state of the art with a simpler architecture, achieving 76.50% in rank 1. © 2017 IEEE.
Análisis comparativo de técnicas de aprendizaje automático para detectar fraude en tarjetas de crédito
(Universidad Católica San Pablo, 2019) Tong Chabes, Luis; Ochoa Luna, Jose Eduardo
Este estudio resalta la importancia de llevar un control para detectar fraudes en tarjetas de crédito para prevenir diferentes riesgos hacia nuestros bienes. Las técnicas de Aprendizaje Automático han demostrado ser la solución para el aprendizaje supervisado. Este trabajo identifica técnicas como Máquinas de Vectores de Soporte, Clasificador Bayesiano Ingenuo, Bosques Aleatorios, Red Neuronal y Extreme Gradiente Boost como las mejores técnicas según los trabajos relacionados. Este trabajo se enfocó en realizar todo el proceso que aborda un proyecto como este, es decir ingeniería de características, preparar los datos, lidiar con el desbalance de datos, entre otros. Se usó como herramienta de evaluación de rendimiento la validación cruzada k-fold para encontrar la mejor parametrización de cada una de estas técnicas, que son evaluadas con métricas de desempeño como exactitud y puntaje f1. Y finalmente hacer una comparación de estos resultados agregando pruebas estadísticas como t de estudiante para obtener la técnica ganadora.
Analyzing the effect of hyperparameters in a automobile classifier based on convolutional neural networks
(IEEE Computer Society, 2017) Laura Riveros, Elian; Galdos Chávez, José; Gutiérrez Cáceres, Juan
In the recent years the convolutional neural network is used successfully in applications of image classification, due to its deep and hierarchical architecture. The hyper parameters of the convolutional neural networks are of great influence to obtain good results in binary classification without the need of a large number of layers. The activation function, the weights initialization and the sub sampling function are the three main hyper parameters. In the present work 27 models of convolutional neural network are trained and tested with automobile images taken from a surveillance camera. The illumination intensity of the test images are different from the training images, because they were taken from scenes of day, evening and night. We also demonstrate the influence of the mean of the images and the size of the filter kernel. The convolutional neural network model with the best result reached 95.6% of accuracy. The results of experiments show that neural networks predict successfully automobile images with varied illumination intensities overcome the techniques Haar Cascade and the Support Vector Machine. © 2016 IEEE.
Assertion role in a hybrid link prediction approach through probabilistic ontology
(CEUR-WS, 2013) Armada, Marcius; Revoredo, Kate; Luna, J.; Cozman, F.
Link prediction in a network is mostly based on information about the neighborhood topology of the nodes. Recently, the interest for hybrid link prediction approaches that combine topology information with information about the network individuals, has grown. However, considering the whole set of individuals may not be necessary and sometimes not even suitable. Therefore, mechanisms to automatically discover the relevant set of individuals are demanding. In this paper, we encompass this problem by proposing an algorithm that combines structure and semantic metrics to find the set of relevant individuals. We empirically evaluate this proposal analyzing the assertion role of these individuals when predicting a link through a probabilistic ontology.
Automatic interpretation of map visualizations with color-encoded scalar values from bitmap images
(Universidad Católica San Pablo, 2018) Mayhua Quispe, Angela Gabriela; Poco Medina, Jorge Luis
Map visualizations are used in diverse domains to show geographic data (e.g., climate research, oceanography, business analyses, etc.). These visualizations can be found in news articles, scientific papers, and on the Web. However, many map visualizations are available only as bitmap images, hindering machine interpretation of the visualized data for indexing and reuse. In this work, we propose a pipeline to recover the visual encodings from bitmap images of geographic maps with color-encoded scalar values. We evaluate our results using map images from scientific documents, achieving high accuracy along each step of the pipeline. In addition, we present iGeoMap, our web-based system that uses the extracted visual encoding to enable user-interaction over bitmap images of map visualizations.

Browse

Browsing Departamento de Ciencias de la Computación by Description "Trabajo de investigación"

Results Per Page

Sort Options