A constructive method for multivariate function approximation by multilayer perceptrons

Geva, Shlomo; Sitte, Joaquin

doi:10.1109/72.143376

Cited by 79 publications

(27 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The MLP is the most common architecture employed for ANN [26] and can implement arbitrary mappings between input and output [27]- [30]. It uses back propagation as learning algorithm wherein the synaptic strengths are systematically modified so that network approximates the desired response more closely.…”

Section: Construction Of Ann Modelsmentioning

confidence: 99%

Modelling and Design of Inverter Threshold Quantization based Current Comparator using Artificial Neural Networks

Bhatia¹,

Pandey²,

Bhattacharyya³

2016

IJECE

View full text Add to dashboard Cite

Performance of a MOS based circuit is highly influenced by the transistor dimensions chosen for that circuit. Thus, proper dimensioning of the transistors plays a key role in determining its overall performance. While choosing the dimension is critical, it is equally difficult, primarily due to complex mathematical formulations that come into play when moving into the submicron level. The drain current is the most affected parameter which in turn affects all other parameters. Thus, there is a constant quest to come up with techniques and procedure to simplify the dimensioning process while still keeping the parameters under check. This study presents one such novel technique to estimate the transistor dimensions for a current comparator structure, using the artificial neural networks approach. The approach uses Multilayer perceptrons as the artificial neural network architectures. The technique involves a two step process. In the first step, training and test data are obtained by doing SPICE simulations of modelled circuit using 0.18μm TSMC CMOS technology parameters. In the second step, this training and test data is applied to the developed neural network architecture using MATLAB R2007b.

show abstract

Section: Construction Of Ann Modelsmentioning

confidence: 99%

Modelling and Design of Inverter Threshold Quantization based Current Comparator using Artificial Neural Networks

Bhatia¹,

Pandey²,

Bhattacharyya³

2016

IJECE

View full text Add to dashboard Cite

show abstract

“…Furthermore, empirical (Murtagh 1991) and theoretical (Geva & Sitte 1992) considerations suggest that the optimal structure for the approximation of a continuous function is through the use of one hidden layer with 2N + 1 nodes (N being the number of input nodes). It was decided that these guidelines would be followed when setting up the ANNs for our classification problem for as long as it remained computationally feasible.…”

Section: Number Of Hidden Layers and Nodesmentioning

confidence: 99%

Machine learning classification of SDSS transient survey images

Buisson

Sivanandam

Bassett

et al. 2015

Mon. Not. R. Astron. Soc.

View full text Add to dashboard Cite

We show that multiple machine learning algorithms can match human performance in classifying transient imaging data from the SDSS supernova survey into real objects and artefacts. This is a first step in any transient science pipeline and is currently still done by humans, but future surveys such as LSST will necessitate fully machine-enabled solutions. Using features trained from eigenimage analysis (PCA) of single-epoch g, r, and i-difference images we can reach a completeness (recall) of 96%, while only incorrectly classifying at most 18% of artefacts as real objects, corresponding to a precision (purity) of 84%. In general random forests performed best, followed by the k-nearest neighbour and the SkyNet artificial neural net algorithms, compared to other methods such as naive Bayes and kernel SVM. Our results show that PCA-based machine learning can match human success levels and can naturally be extended by including multiple epochs of data, transient colours and host galaxy information which should allow for significant further improvements, especially at low signal to noise.

show abstract

“…Often the unknown target function ( ) is inherently complex and cannot be closely approximated by a network comprising of a single hidden layer of neurons implementing simple transfer functions (e.g., sigmoid). To overcome this difficulty, some constructive algorithms use different transfer functions (e.g., the Gaussian [21]) while others such as the projection pursuit regression [18] use a summation of several nonlinear transfer functions. Alternatively, algorithms such as the cascade correlation family construct multilayer networks wherein the structural interconnections among the hidden neurons allow the network to approximate complex functions using relatively simple neuron transfer functions like the sigmoid [13], [39], [49].…”

Section: Constructive Algorithms For Pattern Classificationmentioning

confidence: 99%

Constructive neural-network learning algorithms for pattern classification

Parekh

Yang

Honavar

2000

IEEE Trans. Neural Netw.

208

108

View full text Add to dashboard Cite

Abstract-Constructive learning algorithms offer an attractive approach for the incremental construction of near-minimal neural-network architectures for pattern classification. They help overcome the need for ad hoc and often inappropriate choices of network topology in algorithms that search for suitable weights in a priori fixed network architectures. Several such algorithms are proposed in the literature and shown to converge to zero classification errors (under certain assumptions) on tasks that involve learning a binary to binary mapping (i.e., classification problems involving binary-valued input attributes and two output categories). We present two constructive learning algorithms MPyramid-real and MTiling-real that extend the pyramid and tiling algorithms, respectively, for learning real to M-ary mappings (i.e., classification problems involving real-valued input attributes and multiple output classes). We prove the convergence of these algorithms and empirically demonstrate their applicability to practical pattern classification problems. Additionally, we show how the incorporation of a local pruning step can eliminate several redundant neurons from MTiling-real networks.

show abstract

A constructive method for multivariate function approximation by multilayer perceptrons

Cited by 79 publications

References 9 publications

Modelling and Design of Inverter Threshold Quantization based Current Comparator using Artificial Neural Networks

Modelling and Design of Inverter Threshold Quantization based Current Comparator using Artificial Neural Networks

Machine learning classification of SDSS transient survey images

Constructive neural-network learning algorithms for pattern classification

Contact Info

Product

Resources

About