Fast Learning in Networks of Locally-Tuned Processing Units

Moody, John; Darken, Christian J.

doi:10.1162/neco.1989.1.2.281

Cited by 3,752 publications

(1,459 citation statements)

References 2 publications

Supporting

Mentioning

1,421

Contrasting

Unclassified

Order By: Relevance

“…The aim of this paper is to report the application of these DRBFN and IRBFN methods in solving DEs. In contrast to the approach t a k en by other authors as reviewed above, in the present methods the width of the ith neuron (centre) a (i) is determined according to the following simple relation (Moody and Darken, 1989) a (i) = d (i) (2) where is a factor, > 0, and d (i) is the distance from the ith centre to the nearest centre. Relation (2) indicates that it is reasonable to assign a larger width where the centres are widely separated from each other and a smaller width where the centres are closer.…”

mentioning

confidence: 99%

Numerical solution of differential equations using multiquadric radial basis function networks

2001

View full text Add to dashboard Cite

Abstract. This paper presents mesh-free procedures for solving linear di erential equations (ODEs and elliptic PDEs) based on Multiquadric (MQ) Radial Basis Function Networks (RBFNs). Based on our study of approximation of function and its derivatives using RBFNs that was reported in an earlier paper (Mai-Duy and Tran-Cong, 1999), new RBFN approximation procedures are developed in this paper for solving DEs, which can also be classi ed into two t ypes: a direct (DRBFN) and an indirect (IRBFN) RBFN procedure. In the present procedures the width of the RBFs is the only adjustable parameter according to a (i) = d (i) , where d (i) is the distance from the ith centre to the nearest centre. The IRBFN method is more accuarte than the DRBFN one and experience so far shows that can be chosen in the range 7 10 for the former. Di erent c o m binations of RBF centres and collocation points (uniformly and randomly distributed) are tested on both regularly and irregularly shaped domains. The results for a 1D Poisson's equation show that the DRBFN and the IRBFN procedures achieve a norm of error of at least O(1:0e ; 4) and O(1:0e ; 8), respectively, with a centre density of 50. Similarly, the results for a 2D Poisson's equation show t h a t t h e DRBFN and the IRBFN procedures achieve a norm of error of at least O(1:0e;3) and O(1:0e ; 6), respectively, with a centre density o f 1 2 12.

show abstract

mentioning

confidence: 99%

Numerical solution of differential equations using multiquadric radial basis function networks

2001

View full text Add to dashboard Cite

show abstract

“…The way in which the network is used for data modeling is different when approximating time-series and in pattern recognition. In pattern classification applications the most used radial activated function is the Gaussian [9], [10]. The Gaussian's centers influence the performance of the RBF network.…”

Section: A Classifiers 1) Feed-forward Neural Networkmentioning

confidence: 99%

“…To avoid these situations, they suggested the use of a clustering algorithm to position the centers. The RBF used in this study is combined with the K-means clustering algorithm [9] for initialization of class centers.…”

Section: A Classifiers 1) Feed-forward Neural Networkmentioning

confidence: 99%

A Voting-Based Combination System for Protein Cellular Localization Sites Prediction

Bouziane¹,

Messabih²,

Chouarfa³

2013

IJCTE

View full text Add to dashboard Cite

 Abstract-During recent years, machine learning techniques have been attracting significant attentions in molecular biology and genomic era. They have become increasingly important to solve real-world problems such as elucidating protein function. An important step in the search for knowledge of protein function is to predict its cellular localization sites. Many computational methods that try to solve this problem have been developed over the years but the imbalanced distribution of proteins in cellular locations enormously influences the behavior of these methods. Hence, the performance and efficiency of the existing prediction methods still need to be improved. A computational method for efficiently predicting protein cellular localization is highly required. In this paper, we explore the use of four supervised machine learning algorithms in predicting the cellular localization sites of proteins from the primary sequence information. Our experiments were performed using Naï ve Bayesian, k-Nearest Neighbor and feed-forward Neural Network classifiers. The experts were evaluated with and without cross-validation on E.coli and Yeast benchmarks and combined using majority voting rule for improving classification accuracy on each dataset. The experimental results show that the proposed combination system significantly outperforms the best individual classifier.Index Terms-Protein localization, naï ve Bayesian classifier, k-nearest neighbor classifier, neural network classifier, combination of classifiers, E.coli, yeast.

show abstract

“…It has also been used for determining initial center values for the subsequent supervised training of radial basis function networks (Moody & Darken, 1989). The K-means algorithm finds K vectors m i i 1; …; K given N data points {x n }.…”

Section: K-means Algorithmmentioning

confidence: 99%

Unsupervised visual learning of three-dimensional objects using a modular network architecture

1999

View full text Add to dashboard Cite

This paper presents a modular network architecture that learns to cluster multiple views of multiple three-dimensional (3D) objects. The proposed network model is based on a mixture of non-linear autoencoders, which compete to encode multiple views of each 3D object. The main advantage of using a mixture of autoencoders is that it can capture multiple non-linear sub-spaces, rather than multiple centers for describing complex shapes of the view distributions. The unsupervised training algorithm is formulated within a maximum-likelihood estimation framework. The performance of the modular network model is evaluated through experiments using synthetic 3D wire-frame objects and gray-level images of real 3D objects. It is shown that the performance of the modular network model is superior to the performance of the conventional clustering algorithms, such as the K-means algorithm and the Gaussian mixture model. ᭧

show abstract

Fast Learning in Networks of Locally-Tuned Processing Units

Cited by 3,752 publications

References 2 publications

Numerical solution of differential equations using multiquadric radial basis function networks

Numerical solution of differential equations using multiquadric radial basis function networks

A Voting-Based Combination System for Protein Cellular Localization Sites Prediction

Unsupervised visual learning of three-dimensional objects using a modular network architecture

Contact Info

Product

Resources

About