Channel noise induced stochastic effect of Hodgkin-Huxley neurons in a real classification task.

Autor: Erkan Y; Department of Electrical & Electronics Engineering, Bartin University, Bartin, 74100, Turkiye., Erkan E; Computer Engineering, Bartin University, Bartin, 74100, Turkiye. Electronic address: eerkan@bartin.edu.tr.
Jazyk: angličtina
Zdroj: Journal of theoretical biology [J Theor Biol] 2024 Dec 16; Vol. 599, pp. 112028. Date of Electronic Publication: 2024 Dec 16.
DOI: 10.1016/j.jtbi.2024.112028
Abstrakt: Noise is generally considered to have negative effects on information processing performance. However, it has also been proven that adding random noise or a certain level of stochastic (random) variability to a nonlinear system can increase its performance or sensitivity to weak signals. Despite the studies on this concept, called stochastic resonance in computational neuroscience, this phenomenon is still among the topics that need detailed research, especially in machine learning. In this study, the effect of noise arising from the intrinsic dynamics of the neurons forming the network in a spiking neural network consisting of Hodgkin-Huxley neurons on the image classification success of the network is investigated. In the first part of this two-part study, a practical neural network model consisting of Hodgkin-Huxley neurons is proposed and the network is tested in a 4-class real classification task. It is observed that the network consisting of Hodgkin-Huxley neurons has a classification performance at least as successful as the artificial neural network. In the second part of the study, the neurons in the network are replaced with stochastic Hodgkin-Huxley neurons, which more realistically represent the biological neuron, and the classification performance of the network at different cell membrane sizes is examined. Findings reveal that a spiking network consisting of stochastic Hodgkin-Huxley neurons, in which intrinsic noise dynamics are incorporated into the system, shows maximum classification performance at an optimal intrinsic noise level. It is called this reflection observed in the classification performance of a spiking network, which is referred to as stochastic resonance in computational neuroscience, as stochastic classification resonance in this study. This study also highlights the importance of bridging the gap between biological neuroscience and artificial neural networks for a better understanding of neurological structure.
Competing Interests: Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
(Copyright © 2024 Elsevier Ltd. All rights reserved.)
Databáze: MEDLINE
Externí odkaz: