Publications
2020
1.
Alaiz-Rodríguez, Rocío; Parnell, Andrew
A machine learning approach for lamb meat quality assessment using FTIR spectra Artículo de revista
En: Ieee Access, vol. 8, pp. 52385–52394, 2020, (Publisher: IEEE).
Resumen | Enlaces | BibTeX | Etiquetas: feature selection, food authenticity, neural networks, spectroscopy
@article{alaiz-rodriguez_machine_2020,
title = {A machine learning approach for lamb meat quality assessment using FTIR spectra},
author = {Rocío Alaiz-Rodríguez and Andrew Parnell},
url = {https://ieeexplore.ieee.org/abstract/document/9000861},
year = {2020},
date = {2020-01-01},
urldate = {2020-01-01},
journal = {Ieee Access},
volume = {8},
pages = {52385–52394},
abstract = {The food industry requires automatic methods to establish authenticity of food products. In this work, we address the problem of the certification of suckling lamb meat with respect to the rearing system. We evaluate the performance of neural network classifiers as well as different dimensionality reduction techniques, with the aim of categorizing lamb fat by means of spectroscopy and analyzing the features with more discrimination power. Assessing the stability of feature ranking algorithms also becomes particularly important. We assess six feature selection techniques (χ2, Information Gain, Gain Ratio, Relief and two embedded techniques based on the decision rule 1R and SVM (Support Vector Machine). Additionally, we compare them with common approaches in the chemometrics field like the Partial Least Square (PLS) model and Principal Component Analysis (PCA) regression. Experimental results with a fat sample dataset collected from carcasses of suckling lambs show that performing feature selection contributes to classification performance increasing accuracy from 89.70% with the full feature set to 91.80% and 93.89% with the SVM approach and PCA, respectively. Moreover, the neural classifiers yield a significant increase in the accuracy with respect to the PLS model (85.60% accuracy). It is noteworthy that unlike PCA or PLS, the feature selection techniques that select relevant wavelengths allow the user to identify the regions in the spectrum with the most discriminant power, which makes the understanding of this process easier for veterinary experts. The robustness of the feature selection methods is assessed via a visual approach.},
note = {Publisher: IEEE},
keywords = {feature selection, food authenticity, neural networks, spectroscopy},
pubstate = {published},
tppubtype = {article}
}
The food industry requires automatic methods to establish authenticity of food products. In this work, we address the problem of the certification of suckling lamb meat with respect to the rearing system. We evaluate the performance of neural network classifiers as well as different dimensionality reduction techniques, with the aim of categorizing lamb fat by means of spectroscopy and analyzing the features with more discrimination power. Assessing the stability of feature ranking algorithms also becomes particularly important. We assess six feature selection techniques (χ2, Information Gain, Gain Ratio, Relief and two embedded techniques based on the decision rule 1R and SVM (Support Vector Machine). Additionally, we compare them with common approaches in the chemometrics field like the Partial Least Square (PLS) model and Principal Component Analysis (PCA) regression. Experimental results with a fat sample dataset collected from carcasses of suckling lambs show that performing feature selection contributes to classification performance increasing accuracy from 89.70% with the full feature set to 91.80% and 93.89% with the SVM approach and PCA, respectively. Moreover, the neural classifiers yield a significant increase in the accuracy with respect to the PLS model (85.60% accuracy). It is noteworthy that unlike PCA or PLS, the feature selection techniques that select relevant wavelengths allow the user to identify the regions in the spectrum with the most discriminant power, which makes the understanding of this process easier for veterinary experts. The robustness of the feature selection methods is assessed via a visual approach.
2.
Castejón-Limas, Manuel; Alaiz-Moretón, Héctor; Fernández-Robles, Laura; Alfonso-Cendón, Javier; Fernández-Llamas, Camino; Sánchez-González, Lidia; Pérez, Hilde
Non-removal strategy for outliers in predictive models: The PAELLA algorithm case Artículo de revista
En: Logic Journal of the IGPL, vol. 28, no 4, pp. 418–429, 2020, (Publisher: Oxford University Press).
Resumen | Enlaces | BibTeX | Etiquetas: neural networks, Outlier Detection, PAELLA, Robust Regression
@article{castejon-limas_non-removal_2020,
title = {Non-removal strategy for outliers in predictive models: The PAELLA algorithm case},
author = {Manuel Castejón-Limas and Héctor Alaiz-Moretón and Laura Fernández-Robles and Javier Alfonso-Cendón and Camino Fernández-Llamas and Lidia Sánchez-González and Hilde Pérez},
url = {https://academic.oup.com/jigpal/article-abstract/28/4/418/5670471?login=true},
year = {2020},
date = {2020-01-01},
urldate = {2020-01-01},
journal = {Logic Journal of the IGPL},
volume = {28},
number = {4},
pages = {418–429},
abstract = {his paper proposes an innovative use of the PAELLA algorithm, originally designed for outlier detection, in robust regression. It leverages the occurrence vector from the algorithm to strengthen the influence of reliable samples and reduce the impact of outliers. Several experiments were conducted to assess the use of this vector, comparing different approaches and showing that using weighted neural networks outperforms traditional methods.},
note = {Publisher: Oxford University Press},
keywords = {neural networks, Outlier Detection, PAELLA, Robust Regression},
pubstate = {published},
tppubtype = {article}
}
his paper proposes an innovative use of the PAELLA algorithm, originally designed for outlier detection, in robust regression. It leverages the occurrence vector from the algorithm to strengthen the influence of reliable samples and reduce the impact of outliers. Several experiments were conducted to assess the use of this vector, comparing different approaches and showing that using weighted neural networks outperforms traditional methods.
2012
3.
Alegre, Enrique; González-Castro, Víctor; Alaiz-Rodríguez, Rocío; García-Ordás, María Teresa
Texture and moments-based classification of the acrosome integrity of boar spermatozoa images Artículo de revista
En: Computer Methods and Programs in Biomedicine, vol. 108, no 2, pp. 873–881, 2012, (Publisher: Elsevier).
Resumen | Enlaces | BibTeX | Etiquetas: acrosome integrity, boar semen, Discrete Wavelet Transform, Invariant Moments, k-Nearest Neigbours, neural networks, texture descriptors
@article{alegre_texture_2012,
title = {Texture and moments-based classification of the acrosome integrity of boar spermatozoa images},
author = {Enrique Alegre and Víctor González-Castro and Rocío Alaiz-Rodríguez and María Teresa García-Ordás},
url = {https://www.sciencedirect.com/science/article/pii/S0169260712000314},
year = {2012},
date = {2012-01-01},
journal = {Computer Methods and Programs in Biomedicine},
volume = {108},
number = {2},
pages = {873–881},
abstract = {This paper addresses the automated assessment of sperm quality in the veterinary field by using image analysis to categorize boar spermatozoa acrosomes as intact or damaged. The acrosomes are characterized using texture features derived from first-order statistics, co-occurrence matrices, and Discrete Wavelet Transform coefficients. The study compares texture-based descriptors with moment-based ones and finds that texture descriptors outperform moment-based descriptors, achieving a classification accuracy of 94.93% using Multilayer Perceptron and k-Nearest Neighbors classifiers, offering a promising approach for veterinarians.},
note = {Publisher: Elsevier},
keywords = {acrosome integrity, boar semen, Discrete Wavelet Transform, Invariant Moments, k-Nearest Neigbours, neural networks, texture descriptors},
pubstate = {published},
tppubtype = {article}
}
This paper addresses the automated assessment of sperm quality in the veterinary field by using image analysis to categorize boar spermatozoa acrosomes as intact or damaged. The acrosomes are characterized using texture features derived from first-order statistics, co-occurrence matrices, and Discrete Wavelet Transform coefficients. The study compares texture-based descriptors with moment-based ones and finds that texture descriptors outperform moment-based descriptors, achieving a classification accuracy of 94.93% using Multilayer Perceptron and k-Nearest Neighbors classifiers, offering a promising approach for veterinarians.
4.
Morala-Argüello, Patricia; Barreiro, Joaquín; Alegre, Enrique
A evaluation of surface roughness classes by computer vision using wavelet transform in the frequency domain Artículo de revista
En: The International Journal of Advanced Manufacturing Technology, vol. 59, pp. 213–220, 2012, (Publisher: Springer-Verlag).
Resumen | Enlaces | BibTeX | Etiquetas: neural networks, quality inspection, surface roughness, texture analysis, wavelet transform
@article{morala-arguello_evaluation_2012,
title = {A evaluation of surface roughness classes by computer vision using wavelet transform in the frequency domain},
author = {Patricia Morala-Argüello and Joaquín Barreiro and Enrique Alegre},
url = {https://link.springer.com/article/10.1007/s00170-011-3480-6},
year = {2012},
date = {2012-01-01},
journal = {The International Journal of Advanced Manufacturing Technology},
volume = {59},
pages = {213–220},
abstract = {This study proposes a multiresolution method for unmanned visual quality inspection and surface roughness discrimination in turning. Using wavelet transform, texture features were extracted from surface images, focusing on gray levels in vertical detail sub-images. A multilayer Perceptron neural network classified textures, achieving error rates between 2.59% and 4.17%.},
note = {Publisher: Springer-Verlag},
keywords = {neural networks, quality inspection, surface roughness, texture analysis, wavelet transform},
pubstate = {published},
tppubtype = {article}
}
This study proposes a multiresolution method for unmanned visual quality inspection and surface roughness discrimination in turning. Using wavelet transform, texture features were extracted from surface images, focusing on gray levels in vertical detail sub-images. A multilayer Perceptron neural network classified textures, achieving error rates between 2.59% and 4.17%.
2011
5.
Alaiz-Rodríguez, Rocío; Guerrero-Curieses, Alicia; Cid-Sueiro, Jesús
Class and subclass probability re-estimation to adapt a classifier in the presence of concept drift Artículo de revista
En: Neurocomputing, vol. 74, no 16, pp. 2614–2623, 2011, (Publisher: Elsevier).
Resumen | Enlaces | BibTeX | Etiquetas: Changing Operational Conditions, Concept Drift, Imprecise Class Distribution, Imprecise Data Distribution, neural networks, Posterior Probability Estimation, Supervised Classification
@article{alaiz-rodriguez_class_2011,
title = {Class and subclass probability re-estimation to adapt a classifier in the presence of concept drift},
author = {Rocío Alaiz-Rodríguez and Alicia Guerrero-Curieses and Jesús Cid-Sueiro},
url = {https://www.sciencedirect.com/science/article/pii/S0925231211002359},
year = {2011},
date = {2011-01-01},
journal = {Neurocomputing},
volume = {74},
number = {16},
pages = {2614–2623},
abstract = {This work proposes an algorithm to adapt classifiers when training and test data have different distributions. Unlike methods that adjust only class priors, it models changes at the subclass level without retraining. Experiments show better adaptation to new conditions.},
note = {Publisher: Elsevier},
keywords = {Changing Operational Conditions, Concept Drift, Imprecise Class Distribution, Imprecise Data Distribution, neural networks, Posterior Probability Estimation, Supervised Classification},
pubstate = {published},
tppubtype = {article}
}
This work proposes an algorithm to adapt classifiers when training and test data have different distributions. Unlike methods that adjust only class priors, it models changes at the subclass level without retraining. Experiments show better adaptation to new conditions.
2008
6.
Alaiz-Rodríguez, Rocío; Alegre, Enrique; González-Castro, Víctor; Sánchez-González, Lidia
Quantifying the proportion of damaged sperm cells based on image analysis and neural networks Artículo de revista
En: Proceedings of SMO, vol. 8, pp. 383–388, 2008.
Resumen | Enlaces | BibTeX | Etiquetas: Class Distribution Estimation, neural networks, Quantification Methods, Semen Image Analysis
@article{alaiz-rodriguez_quantifying_2008,
title = {Quantifying the proportion of damaged sperm cells based on image analysis and neural networks},
author = {Rocío Alaiz-Rodríguez and Enrique Alegre and Víctor González-Castro and Lidia Sánchez-González},
url = {https://www.researchgate.net/profile/Victor-Gonzalez-Castro/publication/234818790_Quantifying_the_proportion_of_damaged_sperm_cells_based_on_image_analysis_and_neural_networks/links/0912f5130e6b0a202c000000/Quantifying-the-proportion-of-damaged-sperm-cells-based-on-image-analysis-and-neural-networks.pdf},
year = {2008},
date = {2008-01-01},
journal = {Proceedings of SMO},
volume = {8},
pages = {383–388},
abstract = {This paper addresses the challenge of quantifying the proportion of damaged and intact sperm cells in a sample using computer vision techniques and supervised learning. A novel approach based on Posterior Probability (PP) estimates is introduced to improve classifier accuracy despite changes in class distributions. The PP-based quantification outperforms traditional methods, such as Adjusted Count, Median Sweep, and the naive counting approach, in terms of Mean Absolute Error, Kullback Leibler divergence, and Mean Relative Error. This approach ensures consistent accuracy regardless of class distribution variations.},
keywords = {Class Distribution Estimation, neural networks, Quantification Methods, Semen Image Analysis},
pubstate = {published},
tppubtype = {article}
}
This paper addresses the challenge of quantifying the proportion of damaged and intact sperm cells in a sample using computer vision techniques and supervised learning. A novel approach based on Posterior Probability (PP) estimates is introduced to improve classifier accuracy despite changes in class distributions. The PP-based quantification outperforms traditional methods, such as Adjusted Count, Median Sweep, and the naive counting approach, in terms of Mean Absolute Error, Kullback Leibler divergence, and Mean Relative Error. This approach ensures consistent accuracy regardless of class distribution variations.
7.
Barreiro, Joaquín; Alaiz-Rodríguez, Rocío; Alegre, Enrique; Ablanedo, D
Surface finish control in machining processes using textural descriptors based on moments Miscelánea
2008.
Resumen | Enlaces | BibTeX | Etiquetas: Computer vision, Haralick Descriptors, neural networks, Surface Finish Control, surface roughness
@misc{barreiro_surface_2008,
title = {Surface finish control in machining processes using textural descriptors based on moments},
author = {Joaquín Barreiro and Rocío Alaiz-Rodríguez and Enrique Alegre and D Ablanedo},
url = {https://link.springer.com/chapter/10.1007/978-3-642-12712-0_21},
year = {2008},
date = {2008-01-01},
publisher = {na},
abstract = {This paper introduces a computer vision method for controlling the surface finish of steel parts by classifying them into acceptable and defective categories based on surface roughness. The study uses 143 images of AISI 303 stainless steel, described with three techniques: texture local filters, Haralick descriptors, and wavelet transform features. The classification is done with K-NN and neural networks. The best result, with a 4.0% error rate, was achieved using texture descriptors with K-NN. The optimal configuration with a neural network, using Haralick descriptors, resulted in a 0.0% error rate.},
keywords = {Computer vision, Haralick Descriptors, neural networks, Surface Finish Control, surface roughness},
pubstate = {published},
tppubtype = {misc}
}
This paper introduces a computer vision method for controlling the surface finish of steel parts by classifying them into acceptable and defective categories based on surface roughness. The study uses 143 images of AISI 303 stainless steel, described with three techniques: texture local filters, Haralick descriptors, and wavelet transform features. The classification is done with K-NN and neural networks. The best result, with a 4.0% error rate, was achieved using texture descriptors with K-NN. The optimal configuration with a neural network, using Haralick descriptors, resulted in a 0.0% error rate.
2007
8.
Alaiz-Rodríguez, Rocío; Guerrero-Curieses, Alicia; Cid-Sueiro, Jesús
Minimax regret classifier for imprecise class distributions Artículo de revista
En: Journal of Machine Learning Research, vol. 8, no 4, pp. 103–130, 2007.
Resumen | Enlaces | BibTeX | Etiquetas: Classification, Imprecise Class Distribution, Minimax Deviation, Minimax Regret, neural networks
@article{alaiz-rodriguez_minimax_2007,
title = {Minimax regret classifier for imprecise class distributions},
author = {Rocío Alaiz-Rodríguez and Alicia Guerrero-Curieses and Jesús Cid-Sueiro},
url = {https://www.jmlr.org/papers/volume8/alaiz-rodriguez07a/alaiz-rodriguez07a.pdf},
year = {2007},
date = {2007-01-01},
journal = {Journal of Machine Learning Research},
volume = {8},
number = {4},
pages = {103–130},
abstract = {This paper addresses the challenge of designing a classifier when the stationarity assumption—i.e., the agreement between training and test conditions—does not hold in real-world applications. In these cases, misclassification costs and data generation processes may differ between training and testing. The paper proposes a minimax regret (minimax deviation) approach to classifier design, aiming to minimize the maximum deviation from the optimal risk classifier's performance. Unlike traditional minimax methods, which can lead to severe performance degradation, this approach offers a more robust classification without significant loss of accuracy. The paper presents a neural-based minimax regret classifier for multi-class decision problems and demonstrates its robustness through experimental results.},
keywords = {Classification, Imprecise Class Distribution, Minimax Deviation, Minimax Regret, neural networks},
pubstate = {published},
tppubtype = {article}
}
This paper addresses the challenge of designing a classifier when the stationarity assumption—i.e., the agreement between training and test conditions—does not hold in real-world applications. In these cases, misclassification costs and data generation processes may differ between training and testing. The paper proposes a minimax regret (minimax deviation) approach to classifier design, aiming to minimize the maximum deviation from the optimal risk classifier's performance. Unlike traditional minimax methods, which can lead to severe performance degradation, this approach offers a more robust classification without significant loss of accuracy. The paper presents a neural-based minimax regret classifier for multi-class decision problems and demonstrates its robustness through experimental results.
2005
9.
Alaiz-Rodríguez, Rocío; Guerrero-Curieses, Alicia; Cid-Sueiro, Jesús
Minimax classifiers based on neural networks Artículo de revista
En: Pattern Recognition, vol. 38, no 1, pp. 29–39, 2005, (Publisher: Pergamon).
Resumen | Enlaces | BibTeX | Etiquetas: Minimax Decision Rules, neural networks, Pattern Classification, Uncertainty in Priors
@article{alaiz-rodriguez_minimax_2005,
title = {Minimax classifiers based on neural networks},
author = {Rocío Alaiz-Rodríguez and Alicia Guerrero-Curieses and Jesús Cid-Sueiro},
url = {https://www.sciencedirect.com/science/article/pii/S0031320304002365},
year = {2005},
date = {2005-01-01},
journal = {Pattern Recognition},
volume = {38},
number = {1},
pages = {29–39},
abstract = {This paper addresses the challenge of designing classifiers when prior probabilities are unknown or not representative of the underlying data distribution. Traditional methods assume stationary class priors, leading to suboptimal results when there is a mismatch between training priors and real-world priors. To mitigate this issue, a minimax approach is proposed. The paper presents two algorithms for a neural-based minimax classifier: a learning rate scaling algorithm and a gradient-based algorithm. Experimental results demonstrate that both algorithms successfully find the minimax solution. Additionally, the paper highlights the differences between common approaches and the minimax classifier in dealing with prior uncertainty.},
note = {Publisher: Pergamon},
keywords = {Minimax Decision Rules, neural networks, Pattern Classification, Uncertainty in Priors},
pubstate = {published},
tppubtype = {article}
}
This paper addresses the challenge of designing classifiers when prior probabilities are unknown or not representative of the underlying data distribution. Traditional methods assume stationary class priors, leading to suboptimal results when there is a mismatch between training priors and real-world priors. To mitigate this issue, a minimax approach is proposed. The paper presents two algorithms for a neural-based minimax classifier: a learning rate scaling algorithm and a gradient-based algorithm. Experimental results demonstrate that both algorithms successfully find the minimax solution. Additionally, the paper highlights the differences between common approaches and the minimax classifier in dealing with prior uncertainty.
10.
Alegre, Enrique; Aláiz-Rodríguez, Rocío; Barreiro, Joaquín; Viñuela, M
Tool insert wear classification using statistical descriptors and neuronal networks Artículo de revista
En: Progress in Pattern Recognition, Image Analysis and Applications: 10th Iberoamerican Congress on Pattern Recognition, CIARP 2005, Havana, Cuba, November 15-18, 2005. Proceedings 10, pp. 786–793, 2005, (Publisher: Springer Berlin Heidelberg).
Resumen | Enlaces | BibTeX | Etiquetas: image processing, machine vision, Manufacturing, neural networks, Tool wear
@article{alegre_tool_2005,
title = {Tool insert wear classification using statistical descriptors and neuronal networks},
author = {Enrique Alegre and Rocío Aláiz-Rodríguez and Joaquín Barreiro and M Viñuela},
url = {https://link.springer.com/chapter/10.1007/11578079_82},
year = {2005},
date = {2005-01-01},
journal = {Progress in Pattern Recognition, Image Analysis and Applications: 10th Iberoamerican Congress on Pattern Recognition, CIARP 2005, Havana, Cuba, November 15-18, 2005. Proceedings 10},
pages = {786–793},
abstract = {This work proposes an automated method to determine tool insert wear levels using image analysis. Images of tungsten carbide inserts were acquired during machining of AISI SAE 1045 and 4140 steel bars. After pre-processing and wear area segmentation, statistical moment-based descriptors were extracted. Two classification experiments (binary and three-class) were conducted using Lp2, k-NN, and neural networks. Zernike and Legendre descriptors achieved the best results with a multilayer perceptron (MLP) neural network.},
note = {Publisher: Springer Berlin Heidelberg},
keywords = {image processing, machine vision, Manufacturing, neural networks, Tool wear},
pubstate = {published},
tppubtype = {article}
}
This work proposes an automated method to determine tool insert wear levels using image analysis. Images of tungsten carbide inserts were acquired during machining of AISI SAE 1045 and 4140 steel bars. After pre-processing and wear area segmentation, statistical moment-based descriptors were extracted. Two classification experiments (binary and three-class) were conducted using Lp2, k-NN, and neural networks. Zernike and Legendre descriptors achieved the best results with a multilayer perceptron (MLP) neural network.
2004
11.
Guerrero-Curieses, Alicia; Alaiz-Rodríguez, Rocío; Cid-Sueiro, Jesús
A fixed-point algorithm to minimax learning with neural networks Artículo de revista
En: IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews), vol. 34, no 4, pp. 383–392, 2004, (Publisher: IEEE).
Resumen | Enlaces | BibTeX | Etiquetas: loss functions, minimax learning strategy, neural networks, robust classifiers
@article{guerrero-curieses_fixed-point_2004,
title = {A fixed-point algorithm to minimax learning with neural networks},
author = {Alicia Guerrero-Curieses and Rocío Alaiz-Rodríguez and Jesús Cid-Sueiro},
url = {https://ieeexplore.ieee.org/abstract/document/1347290},
year = {2004},
date = {2004-01-01},
journal = {IEEE Transactions on Systems, Man, and Cybernetics, Part C (Applications and Reviews)},
volume = {34},
number = {4},
pages = {383–392},
abstract = {In some real applications, such as medical diagnosis or remote sensing, available training data do not often reflect the true a priori probabilities of the underlying data distribution. The classifier designed from these data may be suboptimal. Building classifiers that are robust against changes in prior probabilities is possible by applying a minimax learning strategy. In this paper, we propose a simple fixed-point algorithm that is able to train a neural minimax classifier [i.e., a classifier minimizing the worst (maximum) possible risk]. Moreover, we present a new parametric family of loss functions that is able to provide the most accurate estimates for the posterior class probabilities near the decision regions, and we also discuss the application of these functions together with a minimax learning strategy. The results of the experiments carried out on different real databases point out the ability of the proposed algorithm to find the minimax solution and produce a robust classifier when the real a priori probabilities differ from the estimated ones.},
note = {Publisher: IEEE},
keywords = {loss functions, minimax learning strategy, neural networks, robust classifiers},
pubstate = {published},
tppubtype = {article}
}
In some real applications, such as medical diagnosis or remote sensing, available training data do not often reflect the true a priori probabilities of the underlying data distribution. The classifier designed from these data may be suboptimal. Building classifiers that are robust against changes in prior probabilities is possible by applying a minimax learning strategy. In this paper, we propose a simple fixed-point algorithm that is able to train a neural minimax classifier [i.e., a classifier minimizing the worst (maximum) possible risk]. Moreover, we present a new parametric family of loss functions that is able to provide the most accurate estimates for the posterior class probabilities near the decision regions, and we also discuss the application of these functions together with a minimax learning strategy. The results of the experiments carried out on different real databases point out the ability of the proposed algorithm to find the minimax solution and produce a robust classifier when the real a priori probabilities differ from the estimated ones.
12.
Marcos-Provecho, Mª Concepción; Guzmán-Martínez, Roberto; Alaiz-Rodríguez, Rocío
Autoguiado de robots móviles mediante redes neuronales Artículo de revista
En: XXV Jornadas de Automática: Ciudad Real, 8, 9, y 10 de septiembre de 2004, pp. 56, 2004, (Publisher: JA Somolinos).
Resumen | BibTeX | Etiquetas: autonomous navigation, machine learning, mobile robotics, neural networks, redes neuronales, robótica móvil
@article{marcos-provecho_autoguiado_2004,
title = {Autoguiado de robots móviles mediante redes neuronales},
author = {Mª Concepción Marcos-Provecho and Roberto Guzmán-Martínez and Rocío Alaiz-Rodríguez},
year = {2004},
date = {2004-01-01},
journal = {XXV Jornadas de Automática: Ciudad Real, 8, 9, y 10 de septiembre de 2004},
pages = {56},
abstract = {Este trabajo implementa una estrategia de autoguiado para un robot móvil en entornos desconocidos mediante una red neuronal. Se desarrolló un entorno en Matlab para la generación de trayectorias, el entrenamiento y la simulación. Las pruebas en el microbot PICBOT3 demostraron la capacidad de la red para navegar sin colisionar con obstáculos.},
note = {Publisher: JA Somolinos},
keywords = {autonomous navigation, machine learning, mobile robotics, neural networks, redes neuronales, robótica móvil},
pubstate = {published},
tppubtype = {article}
}
Este trabajo implementa una estrategia de autoguiado para un robot móvil en entornos desconocidos mediante una red neuronal. Se desarrolló un entorno en Matlab para la generación de trayectorias, el entrenamiento y la simulación. Las pruebas en el microbot PICBOT3 demostraron la capacidad de la red para navegar sin colisionar con obstáculos.
2002
13.
Alaíz-Rodríguez, Rocío; Cid-Sueiro, Jesús
Minimax strategies for training classifiers under unknown priors Artículo de revista
En: Proceedings of the 12th IEEE Workshop on Neural Networks for Signal Processing, pp. 249–258, 2002, (Publisher: IEEE).
Resumen | Enlaces | BibTeX | Etiquetas: class priors, Minimax Classifier, neural networks, robust learning
@article{alaiz-rodriguez_minimax_2002,
title = {Minimax strategies for training classifiers under unknown priors},
author = {Rocío Alaíz-Rodríguez and Jesús Cid-Sueiro},
url = {https://ieeexplore.ieee.org/abstract/document/1030036},
year = {2002},
date = {2002-01-01},
journal = {Proceedings of the 12th IEEE Workshop on Neural Networks for Signal Processing},
pages = {249–258},
abstract = {This paper addresses the challenge of training supervised learning algorithms when the stationarity assumption does not hold, particularly when class prior probabilities in the real data do not match those in the training data. The authors propose a two-step learning algorithm to train a neural network for estimating a minimax classifier that is robust to changes in class priors. In the first step, posterior probabilities based on training data priors are estimated. In the second step, the class priors are adjusted to minimize a cost function that is asymptotically equivalent to the worst-case error rate. The proposed method is applied to a softmax-based neural network, and experimental results demonstrate its advantages over traditional approaches.},
note = {Publisher: IEEE},
keywords = {class priors, Minimax Classifier, neural networks, robust learning},
pubstate = {published},
tppubtype = {article}
}
This paper addresses the challenge of training supervised learning algorithms when the stationarity assumption does not hold, particularly when class prior probabilities in the real data do not match those in the training data. The authors propose a two-step learning algorithm to train a neural network for estimating a minimax classifier that is robust to changes in class priors. In the first step, posterior probabilities based on training data priors are estimated. In the second step, the class priors are adjusted to minimize a cost function that is asymptotically equivalent to the worst-case error rate. The proposed method is applied to a softmax-based neural network, and experimental results demonstrate its advantages over traditional approaches.
14.
Alaiz-Rodríguez, Rocío; Cid-Sueiro, Jesús
Neural minimax classifiers Artículo de revista
En: International Conference on Artificial Neural Networks, pp. 408–413, 2002, (Publisher: Springer Berlin Heidelberg Berlin, Heidelberg).
Resumen | Enlaces | BibTeX | Etiquetas: class priors, minimax strategy, neural networks, robust learning, softmax
@article{alaiz-rodriguez_neural_2002,
title = {Neural minimax classifiers},
author = {Rocío Alaiz-Rodríguez and Jesús Cid-Sueiro},
url = {https://link.springer.com/chapter/10.1007/3-540-46084-5_66},
year = {2002},
date = {2002-01-01},
journal = {International Conference on Artificial Neural Networks},
pages = {408–413},
abstract = {This paper presents a method to train neural networks using a minimax strategy, which aims to minimize the error probability under the worst-case scenario where class priors in the training data differ from those in the test set. The approach is demonstrated using a softmax-based neural network, but it is also applicable to other neural network structures. Experimental results indicate that the proposed method outperforms other approaches in handling situations where the stationarity assumption (i.e., consistency between training and test data distributions) is violated.},
note = {Publisher: Springer Berlin Heidelberg Berlin, Heidelberg},
keywords = {class priors, minimax strategy, neural networks, robust learning, softmax},
pubstate = {published},
tppubtype = {article}
}
This paper presents a method to train neural networks using a minimax strategy, which aims to minimize the error probability under the worst-case scenario where class priors in the training data differ from those in the test set. The approach is demonstrated using a softmax-based neural network, but it is also applicable to other neural network structures. Experimental results indicate that the proposed method outperforms other approaches in handling situations where the stationarity assumption (i.e., consistency between training and test data distributions) is violated.