Growth strategy determines the memory and structural properties of brain networks (Q6079048)
From MaRDI portal
scientific article; zbMATH DE number 7743462
Language | Label | Description | Also known as |
---|---|---|---|
English | Growth strategy determines the memory and structural properties of brain networks |
scientific article; zbMATH DE number 7743462 |
Statements
Growth strategy determines the memory and structural properties of brain networks (English)
0 references
28 September 2023
0 references
There is an investigation of interaction between dynamics and topology, structure and function, in the domain of neural networks. After a short review of recent research strategies and results connected to this problem, one concentrates the investigation on a computational model of synaptic pruning with an initial transient period of high synaptic density. The synaptic pruning network model is largely described in the second section of the article. Actually one uses the mathematical framework from earlier articles of the authors: [\textit{S. Johnson} et al., Phys. Rev. Lett. 104, No. 10, Article ID 108702, 4 p. (2010; \url{doi:10.1103/PhysRevLett.104.108702})], and [\textit{A. P. Millán} et al., Nature Commun. 9, No. 1, Article ID 2236, 10 p. (2018; \url{doi:10.1038/s41467-018-04537-6})]. The neuronal model is the well-known auto associative Amari-Hopfield model. Each neuron i is modelled by a stochastic binary variable \(s_i\)(t, \(\tau\)) = {0,1} and synapses are characterized by synaptic weights \(\omega _ {i,j}\). The dynamics is based on Hebbian learning. The macroscopic activity of the network can be measured by its overlap with the memorized pattern. The overlap indicates the memory state regardless of the number of memorized patterns and their mean activity. The topology of the network changes in time following a specific Markov process. The initial conditions for the neural states are randomly selected. All model parameters and system sizes are specified. The time scale of the neuronal dynamics is measured by Monte Carlo steps. Simulation results, observations on evolution of the system, non-linear effects of the initial density on the emerging behavior of the system, optimal transient connectivity, memory stability are presented and discussed in the third section. A large discussion on implications for brain development, robustness of the results, memory storage capacity and conclusion is to be found in the fourth section. Reading the References to this article one can observe that this work is a part of a larger investigation performed by the group of authors on the interaction topology-dynamics in neural networks.
0 references
brain development
0 references
co-evolving neural network
0 references
associative memory
0 references
complex networks
0 references
temporal networks
0 references