Fig. 3

Distribution of network parameters learned by minimizing probability flow (MPF) sharpens around three critical values. (a) Portion of network weights W after minimizing (5) given 100 (bottom), 1000 (middle), or 10,000 (top) random 40-cliques X (of about 10²³ in total) on \(v = 80\) vertices. These networks represent the marked points in Fig. 2. (b) Histograms of weight and threshold parameters for networks in (a) (histogram of thresholds θ in inset). Network parameters are scaled so that thresholds have mean 1 (this does not affect the dynamics). Groups of similar network weights and thresholds are labeled with corresponding parameter x, y, z