Estimating the Uncertainty of a Multilayer Perceptron Using the Monte Carlo Method

2012 ◽  
Vol 628 ◽  
pp. 324-329
Author(s):  
F. García Fernández ◽  
L. García Esteban ◽  
P. de Palacios ◽  
A. García-Iruela ◽  
R. Cabedo Gallén

Artificial neural networks have become a powerful modeling tool. However, although they obtain an output with very good accuracy, they provide no information about the uncertainty of the network or its coverage intervals. This study describes the application of the Monte Carlo method to obtain the output uncertainty and coverage intervals of a particular type of artificial neural network: the multilayer perceptron.

2018 ◽  
Vol 7 (2) ◽  
pp. 1
Author(s):  
Paulo Marcelo Tasinaffo ◽  
Gildárcio Sousa Gonçalves ◽  
Adilson Marques da Cunha ◽  
Luiz Alberto Vieira Dias

This paper proposes to develop a model-based Monte Carlo method for computationally determining the best mean squared error of training for an artificial neural network with feedforward architecture. It is applied for a particular non-linear classification problem of input/output patterns in a computational environment with abundant data. The Monte Carlo method allows computationally checking that balanced data are much better than non-balanced ones for an artificial neural network to learn by means of supervised learning. The major contribution of this investigation is that, the proposed model can be tested by analogy, considering also the fraud detection problem in credit cards, where the amount of training patterns used are high.


2018 ◽  
pp. 77-83

Modelización del ensayo de resistencia a compresión del concreto de alta resistencia mediante una red neuronal artificial. Obtención de la incertidumbre del resultado Modeling the resistance to compression of high performance concrete test by means of an artificial neural network. Obtaining the uncertainty of the results Francisco García Fernández1, Ana Torre Carrillo2, Isabel Moromi Nakata2, Pedro Espinoza Haro3 y Luis Acuña Pinaud3 1 Dpto. Sistemas y Recursos Naturales. Universidad Politécnica de Madrid. Ciudad Universitaria S/N, 28040 Madrid, España 2 Facultad de Ingeniería Civil, Universidad Nacional de Ingeniería. Av. Túpac Amaru, 210. Lima 25, Perú 3 Facultad de Ingeniería Industrial y de Sistemas. Universidad Nacional de Ingeniería. Av. Túpac Amaru, 210. Lima 25, Perú DOI: https://doi.org/10.33017/RevECIPeru2015.0012/  Resumen En los últimos años las ANN han tenido un gran desarrollo en el control de procesos industriales debido principalmente a su capacidad de modelizar relaciones complejas, que los sistemas tradicionales no han sido capaces de hacer, y predecir satisfactoriamente si las características de un producto se adecuan o no a las especificaciones correspondientes. Estas estructuras han sido ampliamente utilizadas en la caracterización de otros materiales como cemento, hormigón, algunos metales o la madera. El perceptrón multicapa, una de las redes neuronales artificiales más populares, se ha convertido en los últimos tiempos en una potente herramienta de modelización en numerosos campos que van desde las finanzas, a la ingeniería o la medicina. Esta herramienta consigue mejorar sustancialmente cualquier modelo previo propuesto para modelizar cualquier sistema independientemente de la naturaleza de éste, con la ventaja añadida de que no necesitan ninguna suposición previa sobre la estructura de los datos Sin embargo, la red sólo proporciona el valor de la salida sin ninguna información acerca de su precisión. La obtención de la incertidumbre de salida es importante, no sólo porque proporciona un intervalo de confianza sobre el valor de salida, sino porque da una idea de la calidad del método de medida. Esta incertidumbre procede de dos fuentes, por un lado el ruido inherente a los valores de entrada y por otro la simplificación del fenómeno que todo modelo matemático supone. En este trabajo se va a desarrollar una nueva metodología para obtener tanto la incertidumbre como los intervalos de confianza de la salida de un modelo específico de red neuronal, el perceptrón multicapa, basándose en el método de simulación de Montecarlo especificado en Suplemento 1 de la GUM para posteriormente aplicarlo a la modelizacion del ensayo de resistencia a compression del concreto. Descriptores: Concreto de alta resistencia, red neuronal artificial, resistencia a compresión, incertidumbre, Método de Monte Carlo Abstract Major advances have been made with the use of ANNs in recent years in industrial process control, mainly because they are capable of modeling complex relations, unlike conventional systems, and can adequately predict whether or not the characteristics of a product are in line with specifications. They have been widely used to characterize other materials such as cement, concrete, certain metals or wood. The multilayer perceptron, one of the most popular artificial neural networks, has become a powerful modeling tool in numerous fields, ranging from finances to engineering and medicine. This tool is capable of considerably improving on all previous models proposed for modeling any system, regardless of its nature, with the added advantage that no prior assumption on the structure of the data is necessary. However, the network provides only the output value, with no information about its accuracy. Obtaining the output uncertainty is important, not only because it provides a coverage interval for the output value, but also because it indicates the quality of the measuring method. This uncertainty comes from two sources: firstly, the inherent uncertainty in the input data, and secondly, the simplification of the phenomenon involved in any mathematical model. This study develops a new methodology for obtaining both the output uncertainty and coverage intervals of a specific neural network model - the multilayer perceptron - based on the Monte Carlo simulation method indicated in Supplement 1 to the Guide to the Expression of Uncertainty in Measurement (GUM), in order to use it when modelling the test of resistance to compression of concrete. Keywords: High performance concrete, artificial neural network, resistance to compression, uncertainty, Monte Carlo method.


2017 ◽  
Vol 864 ◽  
pp. 363-368 ◽  
Author(s):  
Long Qi ◽  
Zi Chang Shangguan

With the continuous development of social economy, China's port construction scale has become saturated. As an important part of the harbor, the breakwater plays a crucial role in the safety of the working environment in the harbor, and the reliability of the breakwater is an assessment of its safety. Build on the previous studies, this paper puts forward a semi-submersible breakwater using waste tires. The reliability of this breakwater is analyzed by Monte Carlo method of artificial neural network based on Matlab, and the results are compared to those of Direct sampling Monte Carlo method and Important sampling Monte Carlo method. The results show that the Monte Carlo method is able to analyze the reliability of overturning failure of semi-submersible breakwater of waste tire. Compared with the other two consequences, the result is more accurate. The Monte Carlo method of artificial neural network based on Matlab has the obvious advantage in being devoted to the problem of complex structure and variable. Safety of the breakwater meets the relevant requirements and can be applied to the actual engineering. It can be seen that waste tires has a high degree of reliability, daptability, and a wide range of applications.


2012 ◽  
Vol 1 (4) ◽  
pp. 89-105 ◽  
Author(s):  
Ebru Turanoglu ◽  
Ozlem Senvar ◽  
Cengiz Kahraman

Oil and energy markets have experienced dramatic changes over the past three decades. Due to these changes, it may be difficult to model and forecast the oil consumption with traditional methods such as regression. Artificial Neural Networks (ANNs) are the strong rival of regression and time series in forecasting. ANNs provide good accuracy along with more reliable and precise forecasting for policy makers, in this regard, ANNs can establish the foundation for oil consumption management by providing good model results. This paper tries to unfold the oil consumption forecasting in Turkey using ANN through some predetermined inputs, which is data for population, GDP, import and export of Turkey from 1965 to 2010, with the aim of finding the essential structure of the data to forecast future oil consumption in Turkey with less error.


Author(s):  
BI Marchenko ◽  
NK Plugotarenko ◽  
OA Semina

Introduction: Ensuring a further improvement of efficiency of the public health monitoring system requires integration of the modern health risk analysis methodology with a complex of adapted unified traditional and innovative analytical methods and data exchange with the environmental monitoring system. Objectives: The study aimed to test and assess the accuracy of predicting the incidence of malignant neoplasms using an artificial neural network. Materials and methods: The analyzed time series are presented by information from statistical reporting forms on malignant neoplasms in the city of Taganrog, Rostov Region. We applied a regression model and a forecasting modeling technique based on a feedforward artificial neural network of a multilayer perceptron type. An artificial neural network with 117 neurons in a hidden layer was created in the environment of the Matlab R2021a application package with a set of tools for the synthesis and analysis of neural networks Neural Network Toolbox using the Levenberg-Marquardt algorithm for its learning. Results: Approbation of two forecasting models was carried out on learning samples of different duration including 15 and 34 years. In a comparative assessment of the accuracy of forecasts for 2018 and 2019, absolute and relative errors were estimated. The accuracy of the neural network forecasting model was higher than that of the regression model both for the total of malignant neoplasms and for most cancer sites. The absolute errors of forecasts for 2018 when using regression and neural network models were 17.05 and 1.49 per 100,000 population, for 2019 – 39.07 and 4.42, respectively. The prediction accuracy dropped with a decrease in the time series duration and an increase in the distance from the boundaries of the learning sample. Conclusions: The feedforward artificial neural network of the multilayer perceptron type provides more accurate predictions using minimal input information compared to the regression model, which is its undoubted advantage.


2021 ◽  
Vol 9 (3) ◽  
pp. 351
Author(s):  
Sawendo Eko Wijana ◽  
I Gede Santi Astawa ◽  
AAIN Eka Karyawati

Abstract Classification is the process of differentiating a set of models into several data classes. There are many methods that can be used for the classification process, one of which is the Artificial Neural Network method. Neural networks are a computational method that mimics biological syafar networks. Artificial condition networks can be used to model complex relationships between input and output to recognize patterns in data [1]. In this study, testing was conducted to determine the effect of uncorrelated or low-correlation features in the data classification process and the effect of changing the number of units in the hidden layer on the classification results. The data used in this study were liver disease dataobtained from the Kaggle Dataset.Where in comparing the results of using feature selection, it is divided into 4 predetermined scenarios through the search for significance values ??with the SPSS correlation test.In the results of the implementation of the Multilayer Perceptron which aims to determine the effect of feature selection on the classification results, the results are that feature selection does not really affect the computation time obtained, and correlated data has more influence on the accuracy obtained when compared to uncorrelated data. In the results of the implementation of the Multilayer Perceptron which aims to determine the effect of changing the number of hidden layer units on the classification results, the results show that changes in the number of units in the hidden layer in Artificial Neural Networks have increased significantly in accuracy in several scenarios, but the computation time increases if the number of units in the hidden layer increases. Keywords: Classification, Artificial Neural Network, Liver Disease, Accuracy, Time.


2016 ◽  
Vol 23 (2) ◽  
pp. 281-294 ◽  
Author(s):  
Rodrigo Coral ◽  
Carlos A. Flesch ◽  
Cesar A. Penz ◽  
Mauro Roisenberg ◽  
Antonio L. S. Pacheco

Abstract When an artificial neural network is used to determine the value of a physical quantity its result is usually presented without an uncertainty. This is due to the difficulty in determining the uncertainties related to the neural model. However, the result of a measurement can be considered valid only with its respective measurement uncertainty. Therefore, this article proposes a method of obtaining reliable results by measuring systems that use artificial neural networks. For this, it considers the Monte Carlo Method (MCM) for propagation of uncertainty distributions during the training and use of the artificial neural networks.


Sensors ◽  
2020 ◽  
Vol 21 (1) ◽  
pp. 47
Author(s):  
Vasyl Teslyuk ◽  
Artem Kazarian ◽  
Natalia Kryvinska ◽  
Ivan Tsmots

In the process of the “smart” house systems work, there is a need to process fuzzy input data. The models based on the artificial neural networks are used to process fuzzy input data from the sensors. However, each artificial neural network has a certain advantage and, with a different accuracy, allows one to process different types of data and generate control signals. To solve this problem, a method of choosing the optimal type of artificial neural network has been proposed. It is based on solving an optimization problem, where the optimization criterion is an error of a certain type of artificial neural network determined to control the corresponding subsystem of a “smart” house. In the process of learning different types of artificial neural networks, the same historical input data are used. The research presents the dependencies between the types of neural networks, the number of inner layers of the artificial neural network, the number of neurons on each inner layer, the error of the settings parameters calculation of the relative expected results.


Sign in / Sign up

Export Citation Format

Share Document