Figure 2.

Performance of pure memory networks in terms of retrieval quality (left) and criticality (right) as a function of the load parameter α. Synaptic weights are fixed and defined by {W_ij}. There is neither homeostatic learning nor activity dependent synapses adaptation. (A) Average overlap between initially stored patterns and corresponding retrieved patterns. Averages are taken over 10 trials for each α and error bars indicate one standard deviation from the mean. (B) Mean squared deviation of obtained avalanche size distributions from the best-fit power law. The blue dashed line denotes the threshold Δγ^max = 0.005, below which avalanche size distributions can be considered critical. The inset shows an example avalanche size distribution P(L) obtained in the range of critical load parameters, along with the slope of the best-fit power law (red dashed line).