Efficient supervised learning with reduced training exemplars

Abstract

In this article, we propose a new supervised learning approach for pattern classification applications involving large or imbalanced data sets. In this approach, a clustering technique is employed to reduce the original training set into a smaller set of representative training exemplars, represented by weighted cluster centers and their target outputs. Based on the proposed learning approach, two training algorithms are derived for feedforward neural networks. These algorithms are implemented and tested on two pattern classification applications -skin detection and image classification. Experimental results show that with the proposed learning approach, it is possible to design networks in a fraction of time taken by the standard learning approach, without compromising the generalization ability and overall classification performance. Abstract-In this article, we propose a new supervised learning approach for pattern classification applications involving large or imbalanced data sets. In this approach, a clustering technique is employed to reduce the original training set into a smaller set of representative training exemplars, represented by weighted cluster centers and their target outputs. Based on the proposed learning approach, two training algorithms are derived for feed-forward neural networks. These algorithms are implemented and tested on two pattern classification applications -skin detection and image classification. Experimental results show that with the proposed learning approach, it is possible to design networks in a fraction of time taken by the standard learning approach, without compromising the generalization ability and overall classification performance. Disciplines Computer Sciences | Physical Sciences and Mathematics

CaPrM: Carbonation prediction model for reinforced concrete using machine learning methods

“…The measured carbonation depth is assigned as output neuron. The applied learning algorithm is Levenberg-Marquardt [58]. It is the fastest backpropagation procedure that updates weight and bias values in the negative gradient direction.…”

Predicting mutation score using source code and test suite metrics

“…Random sampling appears to perform well for reducing the training set size while retaining prediction performance [PJO99,WNC05]. There exist more complex approaches to this problem, one uses centroids of weighted clusters which essentially groups similar items in the training set and treating them as one item [NBP08]. With our approach with respect to data availability, we are interested in minimizing the data required to make accuracy predictions.…”

Modeling Net Ecosystem Carbon Dioxide Exchange Using Temporal Neural Networks after Wavelet Denoising

“…Unlike online learning, the performance and applicability of batch learning generally are limited by how much training data a given system memory can handle. However, online learning cannot fully optimize a cost function defined for training examples, whereas batch learning can completely optimize a cost function defined for training examples (Nguyen, Bouzerdoum, and Phung ).…”