Fast learning algorithm for deep evolving GMDH-SVM neural network in data stream mining tasks

Bodyanskiy, Yevgeniy; Vynokurova, Olena; Pliss, Iryna; Setlak, Galina; Mulesa, Pavlo

doi:10.1109/dsmp.2016.7583555

Cited by 16 publications

(6 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Secnction valbeing proe learning d NLB apvalues and the stream uch a case, e data and on we dee whether anged sigre, called algorithm neural netes. a drift deion values Train the network on current element; 6 Compute loss function for a current element; 7 Update v i according to (9) or (10) 8 for every data element in T do 9 Update pods according to (11) 10 Compute loss function on a validation set; 11 Update CuSum according to (13); 12 if CuSum > λ C then 13 Reinitialize pod's values;…”

Section: Resultsmentioning

confidence: 99%

“…It should be also noted that several authors tried to merge the fields of deep learning and data stream mining. In [7] the authors combined the evolving deep neural network with the Least Squares Support Vector Machine. Deep neural networks were also successfully applied in semi-supervised learning task in the context of streamming data.…”

Section: Related Workmentioning

confidence: 99%

“…Input: S -data stream, M -batch size, λ 1 Collect a new batch B from the stream S; 2 for every data element in B do 3 Train the network on current element Calculate a predicted class; 4 if predicted class == expected class then 5 Update v i according to (8); 6 for every data element in T do 7 Update pods according to (11) 8 Return to line 1;…”

Section: Normalized Loss Based (Nlb) Approachmentioning

confidence: 99%

See 2 more Smart Citations

On Training Deep Neural Networks Using a Streaming Approach

Duda

Jaworski

Cader

et al. 2019

Journal of Artificial Intelligence and Soft Computing Research

View full text Add to dashboard Cite

In recent years, many deep learning methods, allowed for a significant improvement of systems based on artificial intelligence methods. Their effectiveness results from an ability to analyze large labeled datasets. The price for such high accuracy is the long training time, necessary to process such large amounts of data. On the other hand, along with the increase in the number of collected data, the field of data stream analysis was developed. It enables to process data immediately, with no need to store them. In this work, we decided to take advantage of the benefits of data streaming in order to accelerate the training of deep neural networks. The work includes an analysis of two approaches to network learning, presented on the background of traditional stochastic and batch-based methods.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Related Workmentioning

confidence: 99%

Section: Normalized Loss Based (Nlb) Approachmentioning

confidence: 99%

See 1 more Smart Citation

On Training Deep Neural Networks Using a Streaming Approach

Duda

Jaworski

Cader

et al. 2019

Journal of Artificial Intelligence and Soft Computing Research

View full text Add to dashboard Cite

show abstract

“…The search for corner points can be realized within the framework of different approaches, taking into account the size of the found areas and their orientation. An example is the Harris angle detector [9], which is based on the study of monotony of image areas. However, since the entire system is formed with a neuronbased solution to the problem, it is possible to use the Hopfield neuron network and the network of radial neurons to accelerate and simplify the search.…”

Section: Genetic Algorithmmentioning

confidence: 99%

Image Processing Models and Methods Research and Ways of Improving Marker Recognition Technologies in Added Reality Systems

Bolohova¹,

Рубан²

2019

Innovative Technologies and Scientific Solutions for Industries

View full text Add to dashboard Cite

The subject matter of article is method of image processing, which identify and describe the local features of images. The aim of the article is the determination of ways for interconnection of the methods for processing the image and technologies creation in the development of markers in the systems of additional reality. The following tasks are solved in the article: to analyze the existing methods and algorithms for finding objects in two-dimensional images to determine the basic marker recognition technology in the complementary reality systems. Analyzed genetic, neural network, statistical and fractal methods, as well as approaches to the algorithms implementation of in the software construction for systems of complementary reality. The next results were obtained: a review and a comparative analysis of the main known algorithms for detecting key points in the images were conducted. It was suggested in the development of marker recognition methods it is necessary to develop a procedure of preliminary image processing for the formation algorithms of the front image for the marker under different conditions of obtaining images. At segmentation stages, it is expedient to use genetic algorithms based on the best indicators of proper segmentation and low processing time, but it is necessary to develop functions that are appropriate for the format of the markers. Improve existing methods for processing segmentation results based on a criterion base describing a visual model representing a marker. Conclusions: as a result of the analysis, the following conclusion can be drawn. The fastest and the most accurate algorithm for putting key points is the genetic algorithm (average time of the algorithm is 5.23 seconds, the number of correct answers is 84.25). The longest working time is the neural network method  8.45 seconds, the accuracy of this algorithm is also the lowest-52. Another advantage of the algorithm of point matching is that if the object goes beyond the frame and then returns again, the program will again continue to track this object. This is supported by algorithms of machine learning. You can also notice that the SIFT calculation works much faster than fractal texture analysis. These results suggest that there are currently no methods for recognizing markers, allowing high accuracy of less than one unit to recognize in a short time. In our opinion, one of the promising directions is the use of Royan methods, namely the development of target functions for accurate and fast recognition of the image by markers.

show abstract

“…The application of commonly used regression methods [3] does not provide satisfactory results in solving this task. In the big data era, the problem is deepened by the need for accurate and quick operation of such methods [4,5].…”

Section: Introductionmentioning

confidence: 99%

Development of the Non-Iterative Supervised Learning Predictor Based on the Ito Decomposition and SGTM Neural-Like Structure for Managing Medical Insurance Costs

Tkachenko

Izonin

Vitynskyi

et al. 2018

Data

View full text Add to dashboard Cite

The paper describes a new non-iterative linear supervised learning predictor. It is based on the use of Ito decomposition and the neural-like structure of the successive geometric transformations model (SGTM). Ito decomposition (Kolmogorov–Gabor polynomial) is used to extend the inputs of the SGTM neural-like structure. This provides high approximation properties for solving various tasks. The search for the coefficients of this polynomial is carried out using the fast, non-iterative training algorithm of the SGTM linear neural-like structure. The developed method provides high speed and increased generalization properties. The simulation of the developed method’s work for solving the medical insurance costs prediction task showed a significant increase in accuracy compared with existing methods (common SGTM neural-like structure, multilayer perceptron, Support Vector Machine, adaptive boosting, linear regression). Given the above, the developed method can be used to process large amounts of data from a variety of industries (medicine, materials science, economics, etc.) to improve the accuracy and speed of their processing.

show abstract

Fast learning algorithm for deep evolving GMDH-SVM neural network in data stream mining tasks

Cited by 16 publications

References 23 publications

On Training Deep Neural Networks Using a Streaming Approach

On Training Deep Neural Networks Using a Streaming Approach

Image Processing Models and Methods Research and Ways of Improving Marker Recognition Technologies in Added Reality Systems

Development of the Non-Iterative Supervised Learning Predictor Based on the Ito Decomposition and SGTM Neural-Like Structure for Managing Medical Insurance Costs

Contact Info

Product

Resources

About