Figure 4
From: Dendritic processing of spontaneous neuronal sequences for single-trial learning
Robust single-trial learning of place fields on a one-dimensional track. (a) Our CA3 network model consists of 500 EC neurons projecting to the distal compartments of 300 two-compartment CA3 neurons, which have inhibitory feedback to both distal dendritic and somatic compartments. DG input activates neuron 1 to neuron 10 of CA3 in a probabilistic manner. (b) DG-evoked preexisting activity patterns in CA3 were simulated without EC input. The animal was immobile from 0 to 5 sec and ran from 5 to 10 sec. (c) The behavioral paradigm and activities of EC neurons in the present simulations. Position-dependent sensory features are encoded by 300 EC neurons, whereas other 200 EC neurons (neuron ID 300 to 500) show position-independent distractor activity. (d,e) Activities of the two-compartment network model (d) and single-compartment network model (e) for animal’s movements shown in the top panels. The single-compartment network model was simulated with η = 0.5. (f) Time evolution during learning is shown for the dynamical variables of the two-compartment neuron. The examples were from CA3 neuron #100. (g) Average information per spike was calculated in various conditions. Three simulation trials were performed in each condition with different initial conditions. The strength of recurrent connections was measured relative to the connection strength used in c and d. In simulating familiar tracks, we used the initial weights of EC-to-CA3 synapses optimized to generate place-dependent firing. In the simulations of unfamiliar tracks, these initial weights were randomly shuffled.
