Operant conditioning experiment with LIF neurons.

<p><b>A:</b> Mean weight into the reinforced neuron () over time for LIF neurons receiving 10,000 excitatory inputs where the weights are updated using the classical (dashed) and dopamine (solid) RSTDP models. Horizontal lines represent analytical predictions for classical RSTDP (dashed), dopamine RSTDP where (dotted), and dopamine RSTDP where the correct value of is assumed (solid). <b>B:</b> Same as A with 8,000 excitatory inputs and 2,000 inhibitory inputs (inhibitory synaptic strengths were fixed at 0.01). <b>C:</b> The mean firing rates of the reinforced (blue), surround (red), and control (green) neurons for the last 30 minutes of the simulations (shaded areas in A and B) with classical (C) and dopamine (D) RSTDP in A (E (log)), B (E+I (log)), and as in B but with additive weight dependence (E+I (add)), as described by <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0087123#pone.0087123.e328" target="_blank">Equation (19</a>). Horizontal lines represent analytical predictions as in A and B. <b>D:</b> The average reward signal after the reinforced neuron's spikes (spike triggered reward) for neurons with different ratios between the excitatory and inhibitory input currents. The different ratios shown increase from no inhibitory inputs (lightest blue) up to the strongest inhibitory inputs (darkest blue), and correspond to the points in <a href="http://www.plosone.org/article/info:doi/10.1371/journal.pone.0087123#pone-0087123-g003" target="_blank">Figure 3D</a>. The first line corresponds to the E case in A and C while the sixth line corresponds to the E+I case in B and C. The inset shows the relationship between the net area of the auto-covariance (AC) function and the peak of the spike triggered reward (STR) curve normalized by the peak of the reward kernel (red dashed line).