Fig. 7: Recurrent network of spiking neurons develops an asymmetric connectivity pattern after learning period with co-dependent synaptic plasticity.
a, Network of 1,000 excitatory and 250 inhibitory neurons. Connections between excitatory neurons and from inhibitory to excitatory neurons are plastic (indicated by *). b, Histogram of the membrane potential during the learning period. Spatial: instantaneous (at a given timestep of the simulation), taking into account all excitatory neurons. Temporal: a single excitatory neuron over 300 s. Both: all excitatory neurons over 300 s. c, Examples of membrane potential dynamics during periods of depolarization (left) and hyperpolarization (right). d, NMDA (red) and GABAA (blue) currents as a function of membrane potential (as in Fig. 6b), highlighting the possible excitatory weight change during periods of hyperpolarization (green bar) and depolarization (yellow bar). e, Sum of excitatory weight changes per second as a function of the membrane potential of the presynaptic (top) and the postsynaptic (bottom) neuron of the connection. Left and right show examples of two distinct neurons of the network. Dots show the amount of change in consecutive 1-s bins given the average membrane potential during that bin. f–h, Mean excitatory input (f) and output (g) connection and inhibitory input connection (h) received per excitatory neuron before (gray) and after (pink) learning. Neurons are ordered from strongest to weakest mean excitatory input connection after a learning period of 10 h. i, Pearson correlation between mean excitatory input and output connections (red) and between mean excitatory and inhibitory input connections (blue) as a function of the balance term used in the co-dependent inhibitory plasticity model. j,k, Firing rate distribution (j) and as a function of the ratio between input excitatory and inhibitory synapses (k), before (gray) and after (pink) the learning period. Mem. pot., membrane potential.
