Fig. 7

Examples of robustness for networks in Fig. 4 of main text with \(v = 128\), \(k = 64\), \(n = 8128\). Adjacency matrices of noisy cliques (in red) have 1219 (top) or 1625 (bottom) bits corrupted out of 8128 (\(p=0.15 / 0.2\)) from the original 64-clique (in green). Images show the result of dynamics applied to these noisy patterns using networks with all-to-all MPF parameters after L-BFGS training on \(50\text{,}000\) 64-cliques (\({\approx}2\mathrm{e}{-}31\%\) of all 64-cliques), Large deviation parameters \((x, y, z) = (0.0091, 0, 1)\), or MPF Theory parameters \((x, y, z) = (0.0107, 0, 1)\) from Eq. (7) in the main text