Comparison of our RSTDP model and classical RSTDP models.

<p><b>A:</b> Amounts of LTP (green) and LTD (red) vs. reward level, with our RSTDP model (solid) and with classical RSTDP with and without unsupervised learning (dashed and dot-dashed, respectively) at the equilibrium synaptic weight. For classical RSTDP without unsupervised learning the reward signal has been shifted such that there is no LTP and LTD at the base reward level, (vertical, black, dashed line) instead of at zero reward, . <b>B:</b> An increase (decrease) in firing rate is predicted to occur in the hatched (unhatched) regions for LTP:LTD ratios at the base reward level (:) of 2:1 (red), 1:2 (green), and 1:1 (blue). On the lines that divide these regions no increase or decrease is predicted. The points marked as C and D correspond to a base level ratio of 1:1 and represent the classical and dopamine parameter sets used in this paper.