Figure 4:

Speed and accuracy as a function of m, the number of synaptic states, and of a, the learning rate (α_n = α_r = α, γ = 0). (A) Time τ required to converge to an estimate of the baiting probability versus m. Different curves correspond to different values of α. τ(α,Μ) is approximately $\sqrt{m}/\alpha$. (B) Standard deviation of P_L versus m for different values of α. As m increases, the fluctuations decrease approximately as $1/\sqrt{m}$, and the accuracy of the estimate increases. The initial fractional baiting probability is $\frac{r_L}{r_L+r_R}=0.1$, and at time zero, it changes to $\frac{r_L}{r_L+r_R}=0.9$. τ is estimated as the time it takes to reach P_L = 0.5, and the standard deviation of P_L is estimated at equilibrium. The other parameters are T = 0.05 and r_L + r_R = 0.35.