“…For fully connected models, we have and then For models with large connectivity but that are sub-linear in the number of neurons, one can consider with , whereas for finite-dimensional models we have that . From preliminary analysis (we extracted data from Figure 1f of Tkacik et al, 2009 [ 63 ] with G3data and performed a Gaussian fit to find the variances, Figure 4 ), we confirm that the couplings are approximately Gaussian, but the scaling exponent appears to be and not as for the Sherrington–Kirkpatrick model [ 56 , 57 , 58 , 60 , 87 ]. This would be interesting, since the system would admit a thermodynamic limit and still have sufficient connectivity to manage the data with a mean-field theory [ 56 , 57 , 58 , 60 , 87 ].…”