Machine learning: Artificial Neural Networks

Home / Machine learning: Artificial Neural Networks

Artificial neural networks (ANN) (Schalkoff, 1997; Yegnanarayana, 2009) is a biologically inspired machine learning method that allows recognizing the pattern of a given data set. This technique is broadly employed in civil and petroleum engineerings (Flood and Kartam, 1994; Yeh, 1998; Mohaghegh, 2000) to deal with the hazardous of the materials' properties and to predict new data from historically collected data. Basically, known data are used to design and traine an ANN network that is then used to predict new data sets. Mathematically, an ANN relates input and output data by following equation systems: Y=b + ω * f(B + W * X), where X and Y are the vectors of input and output data, respectively; f an activation function; ω and W the weight matrix of the network; b and B the bias vectors. By mean of training a network, the number of hidden layers, the activation function, the weight and the bias are calibrated to minimize the mean square error (MSE) between the computed output and the measured data. One the network is trained, it can be used to predict the output Y of new data set.

Let consider a network with a single hidden layer (one hidden layer is enough to deal with a large range of problems) and a single output value for each sample. In this exemple, the sigmoid activation function, f(x)=1/(1+exp(-λx)), is choosen and the stepest method (Fletcher and Powell, 1963) is employed to speedup the convergence of the MSE to zero during the training procedure.

References
Schalkoff, R. J. (1997). Artificial neural networks (Vol. 1). New York: McGraw-Hill.
Yegnanarayana, B. (2009). Artificial neural networks. PHI Learning Pvt. Ltd..
Flood, I., & Kartam, N. (1994). Neural networks in civil engineering. I: Principles and understanding. Journal of computing in civil engineering, 8(2), 131-148.
Yeh, I. C. (1998). Modeling of strength of high-performance concrete using artificial neural networks. Cement and Concrete research, 28(12), 1797-1808.
Mohaghegh, S. (2000). Virtual-intelligence applications in petroleum engineering: Part 1—Artificial neural networks. Journal of Petroleum Technology, 52(09), 64-73.
Fletcher, R., & Powell, M. J. (1963). A rapidly convergent descent method for minimization. The computer journal, 6(2), 163-168.

Use the default data or copy/past a training data set in the text area below (X1, X2, ... are the input arguments; Y the measured output. Each row corresponds to a sample. Four samples with two input arguments are considered in the default example but you can train a data set with more/less samples and arguments)



Input the parameters, then click Compute button to train the network.

Number of hidden nodes

Learning rate

Momentum

Activation parameter λ

(Calibrate these parameters to speed up the training process and increase the accuracy of the results.)









Trained weights and bias: