Evaluating the Efficiency of Algorithms for Neural Module Adaptation to Classification Problems in Dynamic Neural Network
VFAST Transactions on Software Engineering · 02.06.2026 · 14 (2) · 192-208
DOI: 10.21015/vtse.v14i2.2382 ↗Kurz gefasst
Fachartikel in der Zeitschrift VFAST Transactions on Software Engineering, online erschienen am 02.06.2026. Verfasst von Tymur Dovzhenko und 4 weiteren Autor:innen. Die Arbeit nennt 39 Literaturangaben.
Abstract (Originalsprache)
The study addresses the problem of selecting effective adaptation algorithms for dynamic modular neural networks used in classification tasks with variable topology. Existing studies provide limited comparative evidence on how adaptation algorithms behave under different architectural reconfiguration regimes. The aim of the study was to evaluate the relationship between network structural dynamics, training stability, and adaptive efficiency. Three algorithms, namely Backpropagation, Resilient Propagation (Rprop), and Genetic Algorithm, were tested within three architectures: DNN-Full, DNN-Stochastic, and DNN-Evolutionary. The experimental protocol included nine classification scenarios based on MNIST, NSL-KDD, and HAR datasets, using AEI, accuracy, stability, generalization, ΔW-index, adaptation time, and statistical testing. The results showed that the DNN-Evolutionary/Genetic Algorithm configuration achieved the highest adaptive efficiency, especially under distribution shifts, whereas DNN-Stochastic/Rprop provided the shortest adaptation time in less complex scenarios. The DNN-Stochastic architecture also demonstrated increased plasticity under conflicting input conditions. The novelty of the study lies in the integrated AEI-based comparison of algorithm-architecture interactions. The proposed framework can support future benchmarking of reconfigurable neural models under stability and resource constraints.
Zitieren
APA
Dovzhenko, T., Storchak, K., Senkov, O., Bilavka, V., & Vladarchyk, Y. (2026). Evaluating the Efficiency of Algorithms for Neural Module Adaptation to Classification Problems in Dynamic Neural Network. VFAST Transactions on Software Engineering, 14(2), 192-208. https://doi.org/10.21015/vtse.v14i2.2382
Vancouver
Dovzhenko T, Storchak K, Senkov O, Bilavka V, Vladarchyk Y. Evaluating the Efficiency of Algorithms for Neural Module Adaptation to Classification Problems in Dynamic Neural Network. VFAST Transactions on Software Engineering. 2026;14(2):192-208. doi:10.21015/vtse.v14i2.2382
BibTeX
@article{Dovzhenko2026,
author = {Dovzhenko, Tymur and Storchak, Kamila and Senkov, Oleh and Bilavka, Volodymyr and Vladarchyk, Yurii},
title = {Evaluating the Efficiency of Algorithms for Neural Module Adaptation to Classification Problems in Dynamic Neural Network},
journal = {VFAST Transactions on Software Engineering},
year = {2026},
volume = {14},
number = {2},
pages = {192--208},
doi = {10.21015/vtse.v14i2.2382}
}
EUVIMED