Figure 6.

(A) Time evolution of the sum of the neuron-wise information terms ∑_iI [x^t_i; x^{t − 1}] during the course of learning according to the objective function with the lower bound A₁, Equation (9) (the red curve), and the modified objective function obtained by replacing A₁ with A′₁, Equation (10) (the blue curve). (B) The results in Figure 3B are reproduced after learning with the modified objective function and the same learning parameters as in Figure 3B. Log-log plots of size of bursts s against the occurrence of bursts of that size p(s) before (brown) and after learning (red for p_max = 0.4 and N = 50, green for p_max = 0.2 and N = 50, blue for p_max = 0.2 and N = 100, and magenta for p_max = 0.2 and N = 200). The size distributions after learning fall on the straight line of slope around −3/2 (the black line).