Analysis of network structure constructed by self-organizing recurrent network model

摘要

Self-organizing recurrent network (SORN) model is a neural network model that can reproduce statistical and dynamical properties of synaptic connection strengths in the cerebral cortex. In the SORN model, fundamental characteristics of excitatory synaptic connections can be explained as a consequence of self-organization by integrating different forms of plasticity. However, the network structure constructed by the SORN model has not yet been clarified. In this report, we investigated the fraction of connections through spike-timing-dependent plasticity learning. As a result, we revealed that the SORN model has structural sensitive dependence on initial network structures. In addition, we showed that the effect of the structural sensitive dependence on initial network structures is induced by inhibitory spike-timing-dependent plasticity. These results imply that a self-organizing recurrent network with the STDP learning can realize various synaptic connectivities without tuning parameters.