Upper bounds on the number of hidden neurons in feedforward networks with arbitrary bounded nonlinear activation functions

Huang, Guang-Bin; Babri, Haroon A.

doi:10.1109/72.655045

Cited by 457 publications

(53 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In our proposed ensemble, the inputs (to the ensemble) are the output decisions of four known classifiers; for miRNA prediction; and the output is the corresponding ground truth decision. Therefore, our objective is to learn/calculate the best network weights that map the decisions of the adopted classifiers into a single fused output decision [35,36]. …”

Section: Methodsmentioning

confidence: 99%

Ensemble-based classification approach for micro-RNA mining applied on diverse metagenomic sequences

2014

View full text Add to dashboard Cite

BackgroundMicroRNAs (miRNAs) are endogenous ∼22 nt RNAs that are identified in many species as powerful regulators of gene expressions. Experimental identification of miRNAs is still slow since miRNAs are difficult to isolate by cloning due to their low expression, low stability, tissue specificity and the high cost of the cloning procedure. Thus, computational identification of miRNAs from genomic sequences provide a valuable complement to cloning. Different approaches for identification of miRNAs have been proposed based on homology, thermodynamic parameters, and cross-species comparisons.ResultsThe present paper focuses on the integration of miRNA classifiers in a meta-classifier and the identification of miRNAs from metagenomic sequences collected from different environments. An ensemble of classifiers is proposed for miRNA hairpin prediction based on four well-known classifiers (Triplet SVM, Mipred, Virgo and EumiR), with non-identical features, and which have been trained on different data. Their decisions are combined using a single hidden layer neural network to increase the accuracy of the predictions. Our ensemble classifier achieved 89.3% accuracy, 82.2% f–measure, 74% sensitivity, 97% specificity, 92.5% precision and 88.2% negative predictive value when tested on real miRNA and pseudo sequence data. The area under the receiver operating characteristic curve of our classifier is 0.9 which represents a high performance index.The proposed classifier yields a significant performance improvement relative to Triplet-SVM, Virgo and EumiR and a minor refinement over MiPred.The developed ensemble classifier is used for miRNA prediction in mine drainage, groundwater and marine metagenomic sequences downloaded from the NCBI sequence reed archive. By consulting the miRBase repository, 179 miRNAs have been identified as highly probable miRNAs. Our new approach could thus be used for mining metagenomic sequences and finding new and homologous miRNAs.ConclusionsThe paper investigates a computational tool for miRNA prediction in genomic or metagenomic data. It has been applied on three metagenomic samples from different environments (mine drainage, groundwater and marine metagenomic sequences). The prediction results provide a set of extremely potential miRNA hairpins for cloning prediction methods. Among the ensemble prediction obtained results there are pre-miRNA candidates that have been validated using miRbase while they have not been recognized by some of the base classifiers.

show abstract

Section: Methodsmentioning

confidence: 99%

Ensemble-based classification approach for micro-RNA mining applied on diverse metagenomic sequences

2014

View full text Add to dashboard Cite

show abstract

“…As introduced in [19], H is called the hidden layer output matrix of the neural network; the i-th column of H is the i-th hidden node output with respect to inputs x 1 , x 2 , ..., x N . The gradient-based algorithm can be used to train the value of β, ω 1 , ..., ωN , b 1 , ..., bN [20].…”

Section: Extreme Learning Machinementioning

confidence: 99%

Approximate Uncertain Program

2019

View full text Add to dashboard Cite

Chance constrained program where one seeks to minimize an objective over decisions which satisfy randomly disturbed constraints with a given probability is computationally intractable. This paper proposes an approximate approach to address chance constrained program. Firstly, a single layer neural-network is used to approximate the function from decision domain to violation probability domain. The algorithm for updating parameters in single layer neural-network adopts sequential extreme learning machine. Based on the neural violation probability approximate model, a randomized algorithm is then proposed to approach the optimizer in the probabilistic feasible domain of decision. In the randomized algorithm, samples are extracted from decision domain uniformly at first. Then, violation probabilities of all samples are calculated according to neural violation probability approximate model. The ones with violation probability higher than the required level are discarded. The minimizer in the remained feasible decision samples is used to update sampling policy. The policy converges to the optimal feasible decision. Numerical simulations are implemented to validate the proposed method for non-convex problems comparing with scenario approach and parallel randomized algorithm. The result shows that proposed method have improved performance.

show abstract

“…An extension of the above theorem is proposed by Leshno et al [2] which states that a multilayer feedforward network with additive nodes and locally bounded piecewise continuous activation function can approximate any continuous function if and only if the network's activation function is non-polynomial. In addition, the universal approximation capability is generalized for randomized feedforward networks in some research works [3][4][5] and is investigated for a network trained with N distinct samples [6].…”

Section: Introductionmentioning

confidence: 99%

A New Sparse Learning Machine

Nayyeri

Maskooki

Monsefi

2016

Neural Process Lett

View full text Add to dashboard Cite

Many algorithms have been proposed so far for pruning and sparse approximation of feedforward neural networks with random weights in order to obtain compact networks which are fast and robust on various datasets. One drawback of the randomization process is that the resulted weight vectors might be highly correlated. It has been shown that ensemble classifiers' error depends on the amount of error correlation between them. Thus, decrease in correlation between output vectors must lead to generation of more efficient hidden nodes. In this research a new learning algorithm called New Sparse Learning Machine (NSLM) for single-hidden layer feedforward networks is proposed for regression and classification. In the first phase, the algorithm creates hidden layer with small correlation among nodes by orthogonalizing the columns of the output matrix. Then in the second phase, using L 1 -norm minimization problem, NSLM makes the components of the solution vector become zero as many as possible. The resulted network has higher degree of sparsity while the accuracy is maintained or improved. Therefore, the method leads to a new network with a better generalization performance. Numerical comparisons on several classification and regression datasets confirm the expected improvement in comparison to the basic network.

show abstract

Upper bounds on the number of hidden neurons in feedforward networks with arbitrary bounded nonlinear activation functions

Cited by 457 publications

References 20 publications

Ensemble-based classification approach for micro-RNA mining applied on diverse metagenomic sequences

Ensemble-based classification approach for micro-RNA mining applied on diverse metagenomic sequences

Approximate Uncertain Program

A New Sparse Learning Machine

Contact Info

Product

Resources

About