Repositorio Institucional Universidad EAFIT :: Examinando por Materia "Redes neuronales"

Examinando por Materia "Redes neuronales"

Mostrando 1 - 20 de 22

Aplicación de métodos de visión artificial para la detección de huevos en una planta de producción avícola
(Universidad EAFIT, 2024) Cardona Ortiz, César Augusto; Ortiz Arias, Santiago
Comparación entre el método tradicional y algunos basados en inteligencia artificial para el estudio del riesgo crediticio en instituciones financieras colombianas
(Universidad EAFIT, 2018) Arango Correa, Diana Marcela; Colmenares Colmenares, Laura Juliana; Rave Contreras, Isabel Cristina; Torres Guerra, Idier Albeiro
Artificial intelligence models are an open problem for application in various fields of science and search of variable relationships especially when the distribution of events doesn’t depend on a linear function; through this work we want to compare the traditional method most used for credit behavior monitoring with advanced models of artificial intelligence -- The guides that exist in Colombia for management of credit risk are given by the Financial Superintendence of Colombia, international standards such as Basel II, Basel III and Solvency are based on the logistic regression and the discriminant analysis, models used by financial institutions in Colombia to measure credit behavior, thus we carried out an investigation to explore the utility of new models -- This paper addresses one of the traditional methods used in financial institutions, that is, logistic regression, and compares it with alternative methods such as neural networks and random forests -- From the literature review and using a database provided by a banking entity, the dependent variables and the response variable are selected, the logistic regression models, random forests and neural networks are calibrated in the Microsoft Azure Machine Learning application and they are compared to each other with indicators of precision and accuracy such as ROC (from receiver operating characteristic) curve and confusion matrix, obtaining for the models of artificial intelligence, results as good as the traditional one; so they can be used by the financial sector as alternate and / or complementary methods in the analysis of credit risk
Desarrollo de librería para manejo de redes neuronales en Java para tecnofactor
(Universidad EAFIT, 2019) Arboleda Echeverry, Juan Camilo; Rincón Bermúdez, Rafael David
The aim of the present project is to design and develop a working library that enables the creation and adaptation of Neural Networks, defined in a way that is simple to use by Java developers. The developed library will be used to design and obtain a neural network capable of recognizing handwritten digits, from the MNIST database.
Desarrollo de un algoritmo de aprendizaje por refuerzo profundo para resolver el despacho hidrotérmico colombiano considerando escenarios hidrológicos y de demanda bajo incertidumbre
(Universidad EAFIT, 2022) Ramírez Arango, Alejandro; Aguilar Castro, José Lisandro
Economic dispatch is a widely analyzed optimization problem in the electricity sector, which seeks to make the best use of available resources to meet demand at minimum cost. This problem has a great complexity in its solution due to the uncertainty of multiple parameters, being of special interest the hydrological uncertainty for the Colombian case due to its high dependence on hydroelectric plants. In this paper, we view economic dispatch as a multistage decision making problem and propose a Reinforcement Learning to solve the Colombian economic dispatch problem considering hydrological scenarios, due to its ability to handle uncertainty and sequential decisions. The policy performance of our algorithm is compared with classic deterministic method. The main advantage of our method is it can learn from a robust policy to deal the inflow and load demand scenarios.
Estimación de precio de oferta para una planta hidroeléctrica de baja regulación en la bolsa de energía
(Universidad EAFIT, 2021) Mosquera Galvis, Liceth Cristina; Quintero Montoya,Olga Lucia; Celsia
Evaluación de una red neuronal para la solución de ecuaciones diferenciales
(Universidad EAFIT, 2023) Machado-Loaiza, José Manuel; Guarín-Zapata, Nicolás
Hurto a personas en la ciudad de Medellín : análisis predictivo de la cantidad de casos en diferentes zonas de la ciudad a partir de modelos de machine learning implementando técnicas de MLOps
(Universidad EAFIT, 2023) Arboleda Colorado, Jeferson Stiven; Martínez Vargas, Juan David
Robbery of individuals in Medellín is an issue demanding immediate attention. This prompted the study of the phenomenon within an analytics project, spanning data collection, database construction, modeling, and production deployment. It's worth noting that MLOps methodology was employed utilizing AWS services. Visual tools related to the phenomenon were integrated, facilitating analysis.
Identificación de patrones socioeconómicos en Medellín a partir de imágenes satelitales
(Universidad EAFIT, 2024) Ceballos Betancur, Mariana; Martínez Vargas, Juan David; Torres Madronero, María Constanza
Método basado en aprendizaje profundo para la mitigación de artefactos espejo y el Aliasing producido por el submuestreo lateral en el dominio de Fourier en la tomografía de coherencia óptica
(Universidad EAFIT, 2025) Pulgarín Suárez, Diego Alexander; Restrepo Gómez, René; Trujillo Anaya, Carlos Alejandro; Cadavid Muñoz, Juan José
Optical Coherence Tomography (OCT) is an imaging technique that has revolutionized medical diagnosis by providing high-resolution, three-dimensional images of biological tissues non-invasively. Its initial implementation, Time-Domain OCT (TD-OCT), relied on mechanical scanning to acquire each depth profile, which limited its speed and made it susceptible to motion artifacts from the patient, compromising the quality of in-vivo images. To overcome these limitations, Fourier-Domain OCT (FD-OCT) was developed, which captures the entire information of a depth profile simultaneously. This advancement dramatically increased speed and sensitivity but also introduced new challenges. One of the most significant is the mirror artifact, a conjugate copy of the image that appears due to the real nature of the interferometric signal and its subsequent Fourier Transform. Another key problem is aliasing, which arises from the trade-off between acquisition speed and compliance with the Nyquist sampling theorem. In practice, lateral sampling density is reduced to minimize scan time and motion artifacts, but this undersampling causes high spatial frequencies of the tissue to be incorrectly represented as lower frequencies, degrading image fidelity. The primary goal of the research is to address two specific problems: lateral undersampling, which causes aliasing, and the mirror artifact. Solving the former would allow for reduced acquisition time without sacrificing tomogram quality, while eliminating the latter would optimize the use of the image's depth range. Traditionally, solutions for these artifacts have depended on hardware modifications to the OCT system. These approaches, while effective, are costly, increase equipment complexity, and are not easily applicable to existing systems in clinical settings. In response to these limitations, this work proposes an innovative solution based on deep learning that operates exclusively in the digital post-processing stage of the images. The main advantage of this method is that it requires no physical intervention in the optical system, allowing it to be implemented on data acquired with any conventional OCT equipment. To carry out this task, different Generative Adversarial Network (GAN) architectures were explored, with the Pix2Pix model being identified as the most effective for simultaneously correcting both aliasing artifacts from undersampling and the mirror artifact. The effectiveness of this methodology was rigorously validated through qualitative and quantitative analyses. This research not only demonstrates the potential of GANs to solve complex problems in OCT imaging but also lays a solid foundation for their future application in clinical practice. By improving image quality and acquisition efficiency without the need for new hardware, this post-processing approach has the potential to significantly optimize diagnosis, where accuracy and speed are essential.
Modelo de clasificación de insolvencia en Entidades Prestadoras de Salud (EPS) mediante redes neuronales
(Universidad EAFIT, 2025) Muñoz Calderón, Cindy Lorena; Silva Castro, Jaime Andrés; Támara Ayús, Armando Lenin
Health Provider Entities (EPS) are the institutions responsible for managing the resources of Colombia’s healthcare system, thereby ensuring the provision of basic health services. This research aims to develop an insolvency classification model for EPS using the statistical technique of neural networks. The database comprises 51 EPS registered in the EMIS University database, providing financial variables to the model, complemented by non-financial variables (location and age). Four classification models are built and compared using Altman’s Z-Score model as a reference. Among them, the decision tree model stands out for its accuracy in predicting financial insolvency. Additionally, the constructed models identify the variables with the greatest impact on default risk, including ROE, net profit/loss, and current ratio, which significantly influence insolvency risk.
Modelo de medición del riesgo de liquidez para un fondo de empleados a partir del modelo de la Superintendencia Financiera de Colombia
(Universidad EAFIT, 2019) Valencia Sánchez, Edwin; Villanueva, Eduart Humberto
This research project aims to address the current need for employee funds to implement a methodology for liquidity risk measurement that is optimal, reliable and generates quality information for decision-making. The methodology currently applied does not allow the identification of liquidity problems with sufficient anticipation. The employee funds have similar characteristics to the entities monitored by the Financial Superintendence of Colombia; therefore, a model of liquidity risk measurement will be designed from the model implemented by this regulatory entity. The model will be proposed incorporating qualitative and transactional variables, which through the use of neural networks allow determining the probability of withdrawal of resources from the partners and thus prevent the materialization of risk. In this way, the model complies with the regulations for the entities monitored by the Superintendence of the Solidarity Economy.
Movement in video classification using structured data : Workout videos applicationMovement in video classification using structured data : Workout videos application
(Universidad EAFIT, 2023) Múnera Muñoz, Jonathan Damián; Tabares Betancur, Marta Silvia
Nowadays, 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.
Optimización de lotes de fabricación en una industria cosmética para maximizar el GMROI : un enfoque integrado de algoritmos de aprendizaje automático y ARIMA
(Universidad EAFIT, 2025) Idárraga Ojeda, Leidy Viviana; Almonacid Hurtado, Paula María
Optimización de portafolios de inversión mediante pronósticos de volatilidad de «commodities» y acciones, utilizando modelos GARCH y «deep learning»
(Universidad EAFIT, 2024) Villa Cardona, Jairo Alonso; Cruz Castañeda, Vivian
Planteamiento e identificación de las variables para la adaptación del modelo de evaluación de crédito Factoring con redes neuronales (Warren) para Redcapital Colombia
(Universidad EAFIT, 2024) Jaramillo Hurtado, Andrés; Sánchez Ribero, Gustavo Alberto
Predicción de deserción estudiantil mediante técnicas de aprendizaje de máquina
(Universidad EAFIT, 2020) Lopera Vargas, Andrés; Ramírez Echeverri, Sergio Augusto
Predicción del cargue de rutas de distribución mediante aprendizaje de máquina
(Universidad EAFIT, 2023) Ramírez Aguilar, Santiago; Téllez Falla, Diego Fernando; Marentes Cubillos, Luis Andrés
Predicción del precio de transacción sobre el tipo de cambio XAU-USD (oro) para el mercado de contado del commodittie a corto plazo
(Universidad EAFIT, 2023) Cardona Restrepo, Jorge Esteban; Castilla Rueda, Rafael Andrés; Almonacid Hurtado, Paula María
Red asociativa usando LSTM para la predicción diaria de apertura y cierres de índices bursátiles
(Universidad EAFIT, 2022) Mejía Uribe, Simón Pedro José; Laniado, Henry; Almonacid Hurtado, Paula María
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, Tomás
This 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.

Examinando por Materia "Redes neuronales"

Resultados por página

Opciones de ordenación