SYMBOLIC ANALYSIS OF CLASSICAL NEURAL NETWORKS FOR DEEP LEARNING
Deep learning is usually based on matrix computing with a large number of hidden parameters that are not visible outside the computing module. A deep learning algorithm can be implemented in hardware or software as a non-linear system. It is common for researchers to visualize a computing module and...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Center for Quality, Faculty of Engineering, University of Kragujevac, Serbia
2025-03-01
|
| Series: | International Journal for Quality Research |
| Subjects: | |
| Online Access: | http://ijqr.net/journal/v19-n1/6.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1823858042750369792 |
|---|---|
| author | Vladimir Milićević Igor Franc Maja Lutovac Banduka Nemanja Zdravković Nikola Dimitrijević |
| author_facet | Vladimir Milićević Igor Franc Maja Lutovac Banduka Nemanja Zdravković Nikola Dimitrijević |
| author_sort | Vladimir Milićević |
| collection | DOAJ |
| description | Deep learning is usually based on matrix computing with a large number of hidden parameters that are not visible outside the computing module. A deep learning algorithm can be implemented in hardware or software as a non-linear system. It is common for researchers to visualize a computing module and monitor its hidden parameters. In this paper, we propose, as a proof of concept, to start the system design by drawing a single neuron. A more complex scheme of the neural network is obtained by using the copy, move, and paste commands for the simplest unit. The number of neurons and layers can be chosen arbitrarily. When the scheme is complete, implementation code is automatically executed using symbolic inputs, system parameters, and symbolic activation functions. This cannot be done manually because the system response is extremely complex. With the symbolic expression of outputs obtained from inputs and parameters, including pure symbolic activation functions, many other properties can be derived in closed form, such as classification with respect to a single system parameter, activation function, or inputs. This unique original method can help scientists and programmers design complex machine learning algorithms and understand how deep learning algorithms work. This paper presents several examples with new achievements. The proposed algorithm can be implemented in any programming language with symbolic computing. Although it was developed for a classical neural network, the same methodology can be used for any type of neural network. |
| format | Article |
| id | doaj-art-52610f94f21f46db95247f8bf8131493 |
| institution | Kabale University |
| issn | 1800-6450 1800-7473 |
| language | English |
| publishDate | 2025-03-01 |
| publisher | Center for Quality, Faculty of Engineering, University of Kragujevac, Serbia |
| record_format | Article |
| series | International Journal for Quality Research |
| spelling | doaj-art-52610f94f21f46db95247f8bf81314932025-02-11T14:37:39ZengCenter for Quality, Faculty of Engineering, University of Kragujevac, SerbiaInternational Journal for Quality Research1800-64501800-74732025-03-011918510010.24874/IJQR19.01-06SYMBOLIC ANALYSIS OF CLASSICAL NEURAL NETWORKS FOR DEEP LEARNINGVladimir Milićević 0https://orcid.org/0000-0002-5587-2717Igor Franc 1https://orcid.org/0009-0000-4609-1081Maja Lutovac Banduka 2https://orcid.org/0000-0003-4446-706XNemanja Zdravković 3https://orcid.org/0000-0002-2631-6308Nikola Dimitrijević 4https://orcid.org/0000-0002-6595-9277Faculty of Mechanical and Civil Engineering in Kraljevo, University of Kragujevac, Kraljevo, Serbia Faculty of Mechanical and Civil Engineering in Kraljevo, University of Kragujevac, Kraljevo, Serbia RT-RK LLC (former Department of RT-RK Institute, Computer Based Systems), Belgrade, Serbia Belgrade Metropolitan University, Belgrade, Serbia Belgrade Metropolitan University, Belgrade, Serbia Deep learning is usually based on matrix computing with a large number of hidden parameters that are not visible outside the computing module. A deep learning algorithm can be implemented in hardware or software as a non-linear system. It is common for researchers to visualize a computing module and monitor its hidden parameters. In this paper, we propose, as a proof of concept, to start the system design by drawing a single neuron. A more complex scheme of the neural network is obtained by using the copy, move, and paste commands for the simplest unit. The number of neurons and layers can be chosen arbitrarily. When the scheme is complete, implementation code is automatically executed using symbolic inputs, system parameters, and symbolic activation functions. This cannot be done manually because the system response is extremely complex. With the symbolic expression of outputs obtained from inputs and parameters, including pure symbolic activation functions, many other properties can be derived in closed form, such as classification with respect to a single system parameter, activation function, or inputs. This unique original method can help scientists and programmers design complex machine learning algorithms and understand how deep learning algorithms work. This paper presents several examples with new achievements. The proposed algorithm can be implemented in any programming language with symbolic computing. Although it was developed for a classical neural network, the same methodology can be used for any type of neural network.http://ijqr.net/journal/v19-n1/6.pdfartificial neural networksclosed-form expressionfeature extractionmachine learning |
| spellingShingle | Vladimir Milićević Igor Franc Maja Lutovac Banduka Nemanja Zdravković Nikola Dimitrijević SYMBOLIC ANALYSIS OF CLASSICAL NEURAL NETWORKS FOR DEEP LEARNING International Journal for Quality Research artificial neural networks closed-form expression feature extraction machine learning |
| title | SYMBOLIC ANALYSIS OF CLASSICAL NEURAL NETWORKS FOR DEEP LEARNING |
| title_full | SYMBOLIC ANALYSIS OF CLASSICAL NEURAL NETWORKS FOR DEEP LEARNING |
| title_fullStr | SYMBOLIC ANALYSIS OF CLASSICAL NEURAL NETWORKS FOR DEEP LEARNING |
| title_full_unstemmed | SYMBOLIC ANALYSIS OF CLASSICAL NEURAL NETWORKS FOR DEEP LEARNING |
| title_short | SYMBOLIC ANALYSIS OF CLASSICAL NEURAL NETWORKS FOR DEEP LEARNING |
| title_sort | symbolic analysis of classical neural networks for deep learning |
| topic | artificial neural networks closed-form expression feature extraction machine learning |
| url | http://ijqr.net/journal/v19-n1/6.pdf |
| work_keys_str_mv | AT vladimirmilicevic symbolicanalysisofclassicalneuralnetworksfordeeplearning AT igorfranc symbolicanalysisofclassicalneuralnetworksfordeeplearning AT majalutovacbanduka symbolicanalysisofclassicalneuralnetworksfordeeplearning AT nemanjazdravkovic symbolicanalysisofclassicalneuralnetworksfordeeplearning AT nikoladimitrijevic symbolicanalysisofclassicalneuralnetworksfordeeplearning |