Long term memory storage capacity of multiconnected neural networks

摘要: Quantitative expressions of long-term memory storage capacities complex neural network are derived. The networks made neurons connected by synapses any order, the axono-axonal type considered Kandel et al. for example. effect link deletion possibly related to aging, is also considered. central result this study that, within framework Hebb's laws, number stored bits proportional synapses. proportionality factor however, decreases when order involved synaptic contact increases. This tends favor architectures with low-order connectivities. It finally shown that can be optimized a partition into neuron clusters size comparable observed cortical microcolumns.