Середовище моделювання нейронних мереж для розв’язання задачі кластеризації
The task of clustering is solving in various fields of application. In order to achieve a fast and sufficiently accurate clustering solution, it is possible to use special neural networks like Kohonen's self-organization card. This type of neural network is always improved both at the algorithm...
Saved in:
| Date: | 2019 |
|---|---|
| Main Author: | |
| Format: | Article |
| Language: | Ukrainian |
| Published: |
Kamianets-Podilskyi National Ivan Ohiienko University
2019
|
| Online Access: | http://mcm-tech.kpnu.edu.ua/article/view/184494 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Journal Title: | Mathematical and computer modelling. Series: Technical sciences |
Institution
Mathematical and computer modelling. Series: Technical sciences| Summary: | The task of clustering is solving in various fields of application. In order to achieve a fast and sufficiently accurate clustering solution, it is possible to use special neural networks like Kohonen's self-organization card. This type of neural network is always improved both at the algorithm level and at the software level. So, it is necessary to create special software tools that provide the opportunity of training in the same conditions and quickly computational experiments to solve the clustering problem. And also to carry out a comparative analysis of the received results.The second task of such software is to create examples of tasks in technical diagnostics, such as: search of abnormality, classification of signal with losses, and others.This paper include such existing software implementations of SOM (self-organization map) and their respective MLP (multilayer perceptron) to solve precisely the classification problem.All selected software implementations are freely available and spread under free license.SOM and MLP parameters have been defined that may be influenced by the researcher.Criteria for comparing SOM implementations have been selected. This paper presents the architecture of the modeling environment and functionality of its components.For the demonstration is taken the solution of classical problems of machine learning. It helps properly compare the results of computational experiments and to implement the effectiveness of software implementations on both basic and optimized algorithms. |
|---|