Figure 8.

Performance with threshold readout. As in Figure 5 the running average of the reward $\overline{R}$ is plotted averaged over 20 runs. The lower plots show the difference to the baseline trace in Figure 5A. (A) Performance traces for continuous and 8 bit weights. In gray reward R for every trial in a single run with continuous weights is shown. (B) Performance traces for 4 bit resolution with deterministic and probabilistic updates.