Figure 4. The neural network implementing the memory transfer model.

a A schematic representation of the neural network architecture. Here we show stage 1 and 2, but the other memory stages are wired in the same way. Neurons are represented by triangles and synaptic connections by circles. The axons are red, purple and orange, and the dendritic trees are black. Each neuron connects to all the neurons in the same stage (recurrent connections, orange) and to the corresponding neuron in the downstream stage (feed-forward connections, purple). The recurrent connections are plastic whereas the feed-forward connections are fixed. b,c memory encoding: a pattern of activity is imposed in the first stage only and synapses are updated stochastically according to a Hebbian rule. Specifically, if both the pre and the post-synaptic neurons are simultaneously active (b, the activity is schematically represented by trains of spikes), the synapse is potentiated. If the pre-synaptic neuron is active and the post-synaptic neuron is inactive, the synapse is depressed (c).