Examinando por Materia "Deep learning"
Mostrando 1 - 15 de 15
Resultados por página
Opciones de ordenación
Ítem Aplicación de técnicas de aprendizaje automático para la proyección de la tasa de cambio entre COP y USD(Universidad EAFIT, 2022) Granada Carvajal, Lorena; Pérez Ramírez, Fredy OcarisÍtem Automatic detection of building typology using deep learning methods on street level images(PERGAMON-ELSEVIER SCIENCE LTD, 2020-03-20) Duque, J.; Gonzalez, D.; Rueda Plata, Diego; Acevedo, A.; Ramos, R.; Betancourt, A.; García, S.; Mecánica AplicadaAn exposure model is a key component for assessing potential human and economic losses from natural disasters. An exposure model consists of a spatially disaggregated description of the infrastructure and population of a region under study. Depending on the size of the settlement area, developing such models can be a costly and time-consuming task. In this paper we use a manually annotated dataset consisting of approximately 10,000 photos acquired at street level in the urban area of Medellín to explore the potential for using a convolutional neural network (CNN) to automatically detect building materials and types of lateral-load resisting systems, which are attributes that define a building's structural typology (which is a key issue in exposure models for seismic risk assessment). The results of the developed model achieved a precision of 93% and a recall of 95% when identifying nonductile buildings, which are the buildings most likely to be damaged in an earthquake. Identifying fine-grained material typology is more difficult, because many visual clues are physically hidden, but our model matches expert level performances, achieving a recall of 85% and accuracy scores ranging from 60% to 82% on the three most common building typologies, which account for 91% of the total building population in Medellín. Overall, this study shows that a CNN can make a substantial contribution to developing cost-effective exposure models. © 2020 Elsevier LtdÍtem Automatic detection of building typology using deep learning methods on street level images(PERGAMON-ELSEVIER SCIENCE LTD, 2020-03-20) Duque, J.; Gonzalez, D.; Rueda Plata, Diego; Acevedo, A.; Ramos, R.; Betancourt, A.; García, S.; Universidad EAFIT. Departamento de Economía y Finanzas; Research in Spatial Economics (RISE)An exposure model is a key component for assessing potential human and economic losses from natural disasters. An exposure model consists of a spatially disaggregated description of the infrastructure and population of a region under study. Depending on the size of the settlement area, developing such models can be a costly and time-consuming task. In this paper we use a manually annotated dataset consisting of approximately 10,000 photos acquired at street level in the urban area of Medellín to explore the potential for using a convolutional neural network (CNN) to automatically detect building materials and types of lateral-load resisting systems, which are attributes that define a building's structural typology (which is a key issue in exposure models for seismic risk assessment). The results of the developed model achieved a precision of 93% and a recall of 95% when identifying nonductile buildings, which are the buildings most likely to be damaged in an earthquake. Identifying fine-grained material typology is more difficult, because many visual clues are physically hidden, but our model matches expert level performances, achieving a recall of 85% and accuracy scores ranging from 60% to 82% on the three most common building typologies, which account for 91% of the total building population in Medellín. Overall, this study shows that a CNN can make a substantial contribution to developing cost-effective exposure models. © 2020 Elsevier LtdÍtem Automatic detection of building typology using deep learning methods on street level images(PERGAMON-ELSEVIER SCIENCE LTD, 2020-03-20) Duque, J.; Gonzalez, D.; Rueda Plata, Diego; Acevedo, A.; Ramos, R.; Betancourt, A.; García, S.; Duque, J.; Gonzalez, D.; Rueda Plata, Diego; Acevedo, A.; Ramos, R.; Betancourt, A.; García, S.; Universidad EAFIT. Departamento de Ingeniería de Producción; Materiales de IngenieríaAn exposure model is a key component for assessing potential human and economic losses from natural disasters. An exposure model consists of a spatially disaggregated description of the infrastructure and population of a region under study. Depending on the size of the settlement area, developing such models can be a costly and time-consuming task. In this paper we use a manually annotated dataset consisting of approximately 10,000 photos acquired at street level in the urban area of Medellín to explore the potential for using a convolutional neural network (CNN) to automatically detect building materials and types of lateral-load resisting systems, which are attributes that define a building's structural typology (which is a key issue in exposure models for seismic risk assessment). The results of the developed model achieved a precision of 93% and a recall of 95% when identifying nonductile buildings, which are the buildings most likely to be damaged in an earthquake. Identifying fine-grained material typology is more difficult, because many visual clues are physically hidden, but our model matches expert level performances, achieving a recall of 85% and accuracy scores ranging from 60% to 82% on the three most common building typologies, which account for 91% of the total building population in Medellín. Overall, this study shows that a CNN can make a substantial contribution to developing cost-effective exposure models. © 2020 Elsevier LtdÍtem Descubriendo la distribución espacial de comercio en ciudades con economías informales(Universidad EAFIT, 2020) Saldarriaga Marín, Juan Camilo; Duque Cardona, Juan CarlosAn economic census aims to record economic activity in a city by collecting geo-referenced survey data. Although its benefits are significant, the costs are very high and, for this reason, it is very rare that the economic census information is up to date and complete. In this work we propose a new methodology to detect and georeference the visible commercial activity in a city or region in an efficient way, generating automated reports of visible commercial activity in a region of interest. This methodology tries to estimate the spatial distribution that allows having an economic census but only for visible commerce. We contrast the results of our methodology with official information from the Chamber of Commerce to estimate the spatial distribution of informal visible commerce or unregistered commerce in the municipality of Envigado.Ítem Graffiti and government in smart cities : a deep learning approach applied to Medellín city, Colombia(Universidad EAFIT, 2021) Rozo Alzate, Javier Arturo; Vallejo Correa, Paola Andrea; Tabares Betancur, Marta Silvia; Tabares Betancur, Marta SilviaÍtem A Low-Cost Raspberry Pi-based System for Facial Recognition(Universidad EAFIT, 2021-12-01) Miranda Orostegui, Cristian; Navarro Luna, Alejandro; Manjarés García, Alejandro; Fajardo Ariza, Carlos Augusto; Universidad Industrial de Santader; Universidad Industrial de Santander; Instituto Nacional de Astrofísica, Óptica y Electrónica; Universidad Industrial de SantanderDeep learning has become increasingly popular and widely applied to computer vision systems. Over the years, researchers have developed various deep learning architectures to solve different kinds of problems. However, these networks are power-hungry and require high-performance computing (i.e., GPU, TPU, etc.) to run appropriately. Moving computation to the cloud may result in traffic, latency, and privacy issues. Edge computing can solve these challenges by moving the computing closer to the edge where the data is generated. One major challenge is to fit the high resource demands of deep learning in less powerful edge computing devices. In this research, we present an implementation of an embedded facial recognition system on a low cost Raspberry Pi, which is based on the FaceNet architecture. For this implementation it was required the development of a library in C++, which allows the deployment of the inference of the Neural Network Architecture. The system had an accuracy and precision of 77.38% and 81.25%, respectively. The time of execution of the program is 11 seconds and it consumes 46 [kB] of RAM. The resulting system could be utilized as a stand-alone access control system. The implemented model and library are released at https://github.com/cristianMiranda-Oro/FaceNet_EmbeddedSystemÍtem A Method for Detecting Coffee Leaf Rust through Wireless Sensor Networks, Remote Sensing, and Deep Learning: Case Study of the Caturra Variety in Colombia(Universitatea Politehnica Bucuresti, 2020-01-19) Velasquez, D.; Sánchez, A.; Sarmiento Garavito, Sebastián; Toro, M.; Maiza Galparsoro, Mikel; Sierra Araujo, Basilio; Velasquez, D.; Sánchez, A.; Sarmiento Garavito, Sebastián; Toro, M.; Maiza Galparsoro, Mikel; Sierra Araujo, Basilio; Universidad EAFIT. Departamento de Ingeniería de Sistemas; I+D+I en Tecnologías de la Información y las ComunicacionesAgricultural activity has always been threatened by the presence of pests and diseases that prevent the proper development of crops and negatively affect the economy of farmers. One of these pests is Coffee Leaf Rust (CLR), which is a fungal epidemic disease that affects coffee trees and causes massive defoliation. As an example, this disease has been affecting coffee trees in Colombia (the third largest producer of coffee worldwide) since the 1980s, leading to devastating losses between 70% and 80% of the harvest. Failure to detect pathogens at an early stage can result in infestations that cause massive destruction of plantations and significantly damage the commercial value of the products. The most common way to detect this disease is by walking through the crop and performing a human visual inspection. As a result of this problem, different research studies have proven that technological methods can help to identify these pathogens. Our contribution is an experiment that includes a CLR development stage diagnostic model in the Coffea arabica, Caturra variety,scale crop through the technological integration of remote sensing (through drone capable multispectral cameras), wireless sensor networks (multisensor approach), and Deep Learning (DL) techniques. Our diagnostic model achieved an F1-score of 0.775. The analysis of the results revealed a p-value of 0.231, which indicated that the difference between the disease diagnosis made employing a visual inspection and through the proposed technological integration was not statistically significant. The above shows that both methods were significantly similar to diagnose the disease. © 2020 by the authors.Ítem A Method for Detecting Coffee Leaf Rust through Wireless Sensor Networks, Remote Sensing, and Deep Learning: Case Study of the Caturra Variety in Colombia(Universitatea Politehnica Bucuresti, 2020-01-19) Velasquez, D.; Sánchez, A.; Sarmiento Garavito, Sebastián; Toro, M.; Maiza Galparsoro, Mikel; Sierra Araujo, Basilio; Universidad EAFIT. Departamento de Ingeniería Mecánica; Estudios en Mantenimiento (GEMI)Agricultural activity has always been threatened by the presence of pests and diseases that prevent the proper development of crops and negatively affect the economy of farmers. One of these pests is Coffee Leaf Rust (CLR), which is a fungal epidemic disease that affects coffee trees and causes massive defoliation. As an example, this disease has been affecting coffee trees in Colombia (the third largest producer of coffee worldwide) since the 1980s, leading to devastating losses between 70% and 80% of the harvest. Failure to detect pathogens at an early stage can result in infestations that cause massive destruction of plantations and significantly damage the commercial value of the products. The most common way to detect this disease is by walking through the crop and performing a human visual inspection. As a result of this problem, different research studies have proven that technological methods can help to identify these pathogens. Our contribution is an experiment that includes a CLR development stage diagnostic model in the Coffea arabica, Caturra variety,scale crop through the technological integration of remote sensing (through drone capable multispectral cameras), wireless sensor networks (multisensor approach), and Deep Learning (DL) techniques. Our diagnostic model achieved an F1-score of 0.775. The analysis of the results revealed a p-value of 0.231, which indicated that the difference between the disease diagnosis made employing a visual inspection and through the proposed technological integration was not statistically significant. The above shows that both methods were significantly similar to diagnose the disease. © 2020 by the authors.Ítem Modelo de aprendizaje profundo reforzado aplicado al trading de Bitcoin(Universidad EAFIT, 2022) Obando Morales, Sebastián; Jaramillo Posada, Juan RodrigoThe stock market is affected by many types of factors, such as market sentiment, going upwards (bulls) or downwards (bears), the behavior of the economy, or unexpected political events. By For this reason, it is not possible to predict its behavior, which means that it is not possible to decide when to enter or when to exit with certainty. An approach such as deep reinforcement learning, which can emulate the experience of a negotiator (trader) who does not necessarily predict prices, but, market entry and exit times, would be a viable option. The present work sought to implement a reinforced deep learning approach to stock trading (bitcoins, stocks, and commodities), which has shown positive results in the literature with returns positive on investment. The bot, the result of this work, obtained a return of 5%. These positive results open the door to trying new approaches that include new combinations in the way of interpreting indicators to find winning strategies that increase profitability.Ítem Modelo de retorno de equilibrio en la deuda americana(Universidad EAFIT, 2024) Montañez Díaz, Adrián Felipe; Angulo Forero, Santiago Felipe; Velásquez Durán, Isabella; Gallo Restrepo, Juan Carlos; Durango Gutiérrez, María Patricia; Díaz, WalterNowadays, many institutional investors use classic models, such as the Nelson-Siegel-Svensson model, which, through certain parameters, allow the estimation of the yield curve to make informed investment decisions. Despite the general approval of these models, the economic dynamics are changing, which is why, in this work, the nodes of the US yield curve were modeled through linear regression models, machine learning and deep learning, given a defined future scenario. With the result of the projection, the curve was generated by 4 methods, of which the cubic spline resulted to be the best fit.Ítem Movement in video classification using structured data : Workout videos application(Universidad EAFIT, 2023) Múnera Muñoz, Jonathan Damián; Tabares Betancur, Marta SilviaNowadays, several video movement classification methodologies are based on reading and processing each frame using image classification algorithms. However, it is rare to find approaches using angle distribution over time. This paper proposes video movement classification based on the exercise states calculated from each frame's angles. Different video classification approaches and their respective variables and models were analyzed to achieve this, using unstructured data: images. Besides, structure data as angles from critical joints Armpits, legs, elbows, hips, and torso inclination were calculated directly from workout videos, allowing the implementation of classification models such as the KNN and Decision Trees. The result shows these techniques can achieve similar accuracy, close to 95\%, concerning Neural Networks algorithms, the primary model used in the previously mentioned approaches. Finally, it was possible to conclude that using structured data for movement classification models allows for lower performance costs and computing resources than using unstructured data without compromising the quality of the model.Ítem Portfolio Optimization Using Predictive Auxiliary Classifier Generative Adversarial Networks : Application to the Colombian stock market(Universidad EAFIT, 2024) Arango López, Federico; Castellanos Ríos, SantiagoÍtem Predicting Stock prices in Latin America using Associative Deep Neural Networks(Universidad EAFIT, 2023) Gallego Rojas, Juan Fernando; Almonacid Hurtado, Paula MaríaThe stock market is a critical sector of the global economy, and predicting stock prices is of great interest to investors and companies. However, the movements of the market are volatile, non-linear, and complicated. This topic has attracted the attention of researchers, who have proposed formal models that demonstrate accurate predictions can be made with appropriate variables and techniques. Deep learning algorithms are often used for this purpose due to their superior accuracy in time series-based and complex pattern analysis. This paper proposes to predict the opening, closing, highest, and lowest stock prices of select Latin American market indexes using associative deep neural networks that can simultaneously predict related values based on the Long Short Term Memory (LSTM) technique, known for its high accuracy in this area. As well as using classic econometric methods for the analysis of time series such as ARIMA models. The proposed model achieved a good performance in terms of prediction, which in turn allows finding interesting trading opportunities for investors. The results of the models were measured using the average RMSE of the predicted prices metric and compared with those obtained using a naive model.Ítem Stock Market Forecasting : Comparing Machine Learning and Deep Learning with Risk-Return Model Selection and Evaluation in a Walk-forward Approach(Universidad EAFIT, 2024) Castro Marín, Carlos Andrés; Olarte Hernández, Tomás; Olarte Hernández, TomásThis study compares the effectiveness of machine learning and deep learning algorithms in forecasting stock market direction using daily market data of Apple Inc. stock. We aim to determine if these algorithms can identify repeatable patterns across time using price and volume history and assess which are most capable. To ensure robustness, we employ a walk-forward validation approach to maintain the temporal dimension of the data and simulate real trading conditions. This method allows us to test models across different market conditions and measure their predictive power. We prioritize model selection and evaluation based on financial return and risk metrics, focusing on profitability rather than traditional machine learning performance metrics, which often do not correlate with financial outcomes. Our findings show that traditional machine learning algorithms, specifically Random Forest, outperform deep learning models under the selected asset and conditions tested. Machine learning models exceed the stock benchmark regarding Sharpe ratio, while deep learning models struggle to manage risk effectively, leading to poorer performance. This discrepancy is likely due to the complex solution space deep learning algorithms navigate to optimize and the amount of data required by these models. However, we hypothesize that the latter could improve its performance if tested with different architectures and hyperparameters, including newly developed transformer attention-based architectures and models such as TimeGPT and others, shown in the related work section.