![]() |
ИСТИНА |
Войти в систему Регистрация |
Интеллектуальная Система Тематического Исследования НАукометрических данных |
||
Recent experimental findings suggest that plasticity mechanisms of long-term depression involve selective elimination of weak synapses and retaining of strong synapses[1], implying a reorganization of network topology rather than just adjustment of connection weights. Here we present a simple model that explores network topology changes resulting from two different plasticity rules. Networks of excitable neurons were modeled with deterministic excitable nodes involving discrete susceptible, exited and refractory states (SER model), where susceptible nodes nodes became excited by excited linked neighbors. Neurons were connected into random networks with varying numbers of nodes (up to 279) and different density of links (varying from 25% to 75% of fully connected). Simulations were randomly initialized with 10% of excited nodes and equal assignment of susceptible and refractory nodes. After fixed intervals, a Hebbian or Spike time-dependent (STD) like plasticity rule was applied, based on the simultaneous or sequential coactivation matrix of the network, respectively. Specifically, the probability of pruning links was proportional to the normalized (simultaneous or sequential) coactivation of linked nodes. After application of the pruning rules, the same number of pruned connections could be returned to the network in random uniform or local fashion. Simulations with the Hebbian like plasticity rule shaped initially random networks into modular networks, with an associated limited range of sustained network activity. By contrast, simulations with the STD like plasticity rule resulted in hub networks, with a clear segregation on in the degree distribution between averagely connected nodes and a small number of highly connected hub nodes. The latter networks displayed a wide dynamic range. Thus, despite the relative simplicity of the dynamic model, the two plasticity rules produced networks with distinctive topological features of modules and hubs that are widely found in biological neural networks.