Universal Approximation using Incremental Constructive Feedforward Networks with Random Hidden Nodes

By G. Huang, Lei Chen, Chee Kheong Siew. IEEE Transactions on Neural Networks, 2006

2,569 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/db5aa767a0e8ceb09e7202f708e15a37bbc7ca01

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

The s2D method: simultaneous sequence-based prediction of the statistical populations of ordered and disordered regions in proteins. (2015, background): “Since the hidden weights are held fixed, the output weights can be analytically determined, yielding a learning speed that can be thousands of times faster than the one of conventional approaches for training feedforward neural networks [65].”
A structure optimisation algorithm for feedforward neural network construction (2013, methodology): “The ability to approximate functions makes the FNN applicable to the application areas such as modelling and identification in system control [5,6].”
A Fast and Accurate Online Sequential Learning Algorithm for Feedforward Networks (2006, methodology): “OS-ELM originates from the batch learning extreme learning machine (ELM) [20]–[22], [27], [30] developed for SLFNs with additive and RBF nodes.”
Sentic patterns: Dependency-based rules for concept-level sentiment analysis (2014, background): “Note that (5) can be further generalized to a wider class of functions [36] but for the subsequent analysis this aspect is not relevant.”
Minimum Variance Extreme Learning Machine for human action recognition (2014, methodology): “Despite the fact that the determination of the network hidden layer output is a result of randomly assigned weights, it has been shown that SLFN networks trained by using the ELM algorithm have the properties of global approximators [3].”
Foundations and Advances in Deep Learning (2014, background): “Extreme Learning Machines Another related model is an extreme learning machine (ELM) proposed recently by Huang et al. (2006b). The ELM is, in essence, an MLP with a single intermediate layer of nonlinear hidden units.”
Voting based extreme learning machine (2012, background): “It has been further shown [15–18] that many types of computational hidden nodes which may not be neuron alike nodes can be used in ELM as long as they are piecewise nonlinear, such as radial basis function (RBF) hidden nodes [16], fully complex nodes [18], wavelets [3,4], etc.”
Extreme Learning Machine for Regression and Multiclass Classification (2012, methodology): “According to ELM learning theory, a widespread type of feature mappings h ( x ) can be used in ELM so that ELM can approximate any continuous target functions (refer to [14]–[16] for details).”
Error Minimized Extreme Learning Machine With Growth of Hidden Nodes and Incremental Learning (2009, background): “Fortunately, since it has been proved in theory [4], [7] that SLFNs with random hidden nodes have the universal approximation capability, the hidden nodes can be randomly generated independent of the training data.”
The Graph Neural Network Model (2009, background): “Thus, second-order learning algorithms [62], pruning [63], and growing learning algorithms [64]–[66] designed for static networks cannot be directly applied to GNNs.”
Domain Adaptation Extreme Learning Machines for Drift Compensation in E-Nose Systems (2015, background): “…biases can be randomly assigned if the activation function is infinitely differentiable, and also showed that single SLFN with randomly generated additive or RBF nodes with such activation functions can universally approximate any continuous function on any compact subspace of Euclidean space [4].”
Revisiting Recursive Least Squares for Training Deep Neural Networks (2021, background): “Whereas, in striking contrast, they are full of vigor and vitality in extreme learning machines, which are a special type of neural networks without training hidden neurons [35]–[39].”