Modular neural networks with Hebbian learning rule

Goltsev, Alexander; Гриценко, В. И.

doi:10.1016/j.neucom.2008.11.011

Cited by 16 publications

(7 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…[10, 20-23, 31, 32, 48, 51]). These models often use supervised neural network classifiers, such as [31-34, 37, 46], and may use associative (assembly) neural networks [17,18,[20][21][22][23].…”

Section: ])mentioning

confidence: 99%

Extraction of Homogeneous Fine-Grained Texture Segments in Visual Images

Goltsev¹,

Гриценко²,

Húsek³

2017

NNW

Self Cite

View full text Add to dashboard Cite

This paper describes the first attempt of hardware implementation of Multistream Compression (MSC) algorithm. The algorithm is transformed to series of Finite State Machines with Datapath using Register-Transfer methodology. Those state machines are then implemented in VHDL to selected FPGA platform. The algorithm utilizes a special tree data structure, called MSC tree. For storage purpose of the MSC tree a Left Tree Representation is introduced. Due to parallelism, the algorithm uses multiple port access to SDRAM memory.

show abstract

Section: ])mentioning

confidence: 99%

Extraction of Homogeneous Fine-Grained Texture Segments in Visual Images

Goltsev¹,

Гриценко²,

Húsek³

2017

NNW

Self Cite

View full text Add to dashboard Cite

show abstract

“…The compared classifiers are support vector classifiers with radial basis functions (SVC-rbf) [1], support vector machine (SVM) [2,21], neural networks [7,9,33] and neuro-fuzzy classifier [30]. The results of this comparison are shown in Table 5 for MNIST dataset wherein the classifiers were compared according to the number of features (NF) and the percentage recognition rates (RR).…”

Section: Experimental Studiesmentioning

confidence: 99%

“…Among popular handwritten digit databases, the MNIST database has been widely used in recent years as a benchmark for testing new feature extraction and selection methods or for evaluating new classifiers [1][2][3][4][5][6][7][8][9][10]. In the literature, the best accuracies have been achieved for the MNIST database, such as 99.41 % for the robust visionbased features and classification schemes [3], 99.58 % for SVM with gradient features [1], 99.81 % for the convolutional neural networks and elastic distortion [4].…”

Section: Introductionmentioning

confidence: 99%

“…In many studies, more than 100 features have been used to evaluate handwritten digit accuracies for the MNIST database [5][6][7][8]. In certain circumstances, the number of features per character can exceed the number of pixels in the character image [3,9,10]. This increases the processing time of the character recognition.…”

Section: Introductionmentioning

confidence: 99%

“…The other main issue of handwritten digit recognition is to choose a proper classifier. Statistical and machine learning methods have been widely used for handwritten digit classification in various applications [1][2][3][4][5][6][7][8][9][10]. The linear and quadratic discriminant functions [17], the k-nearest neighbor (k-NN) rules [18] and Bayesian networks [19] are some examples to the statistical classifiers.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Use of wavelet-based two-dimensional scaling moments and structural features in cascade neuro-fuzzy classifiers for handwritten digit recognition

Çetişli

Edizkan

2014

Neural Comput & Applic

View full text Add to dashboard Cite

In this paper, a novel handwritten digit recognition system is proposed. The system consist of feature extraction, feature selection and classification stages. The features of digits are extracted by using the moment-based and structural-based methods. For the moment-based method, wavelet-based two-dimensional scaling moments (2-DSMs), which have uniquely different angular divisions of polar form, are considered. The structural-based features including profiles, intersections of horizontal and vertical straight lines, concavity, number and location of holes are used. In the feature selection stage, Fisher's linear discriminant analysis is used to obtain the discriminative features. The feature selection is performed to improve not only the processing time but also recognition rates. In the classification stage, the digits are classified by neuro-fuzzy classifiers (NFCs). A three-stage cascade NFCs with rejection strategy is used in the system to improve the misclassification rate for the handwritten digit recognition task. The experiments are performed on the MNIST and USPS handwritten digit databases. The high correct classification rates of 98.72 % for MNIST and 97.21 % for USPS are attained by using only one hundred robust hybrid features and cascade NFCs. The experiments showed that the proposed system yields better results among those systems that use only moment-based features.

show abstract

Identifying Learning Standards of Artificial Neural Networks: A Study

Desisa

Janbhasha

Meru

2021

ICT With Intelligent Applications

View full text Add to dashboard Cite

Artificial neural networks (ANNs) are now omnipresent, and there is continuous growth in excellent research to create new ones. ANN is capable of adaptive, dynamic learning. In ANN learning, we need to change the weights to enhance the input/output behavior. Therefore, with the help of which the weights can be adjusted, a method is required. These strategies are called learning laws, which are simply formulas or algorithms. Rules of learning are algorithms or mathematical logic that guides modifications is the weight of the network links. By employing the disparity between both the expected output and the real outcome to update its weights during training, they incorporate an error reduction mechanism. Learning laws improve the effectiveness of the artificial neural network and extend this rule to the network. Usually, the learning rule is repeatedly applied to the same set of training inputs over a large number of cycles, with error steadily decreasing over epochs. They are fine-tuned as the weights are. The present research strives, however, to assess the objective of the artificial neural network and the learning principles. In this, we analyzed ten different regulations studied by leading researchers that include rules based on supervised learning (perceptron, memory-based, delta, error correction, correlation, out star, supervised Hebbian) and unsupervised rules based on learning (competitive, competitive Hebbian, Hebbian); this defines how to adjust the weight of the nodes of a network.

show abstract

Modular neural networks with Hebbian learning rule

Cited by 16 publications

References 17 publications

Extraction of Homogeneous Fine-Grained Texture Segments in Visual Images

Extraction of Homogeneous Fine-Grained Texture Segments in Visual Images

Use of wavelet-based two-dimensional scaling moments and structural features in cascade neuro-fuzzy classifiers for handwritten digit recognition

Identifying Learning Standards of Artificial Neural Networks: A Study

Contact Info

Product

Resources

About