Attractor neural networks with double well synapses
Fig 8
Robustness of the double well model in the sparse coding case.
Orange squares represent the storage capacity for the model with the learning rule defined in Eq (6) in the flat potential case (r1 = 10−3). Blue squares represent the storage capacity for the model with the learning rule defined in Eq (7) in the deep potential case (r1 = 0.5). Each point is averaged over 10 independent realizations, and the error bar represents the standard deviation. Other parameters are given as: r2 = 1, c = 1, N = 1, 500, f = 4log(N)/N.