Stability of the one-step replica-symmetry-broken phase in neural networks
E A Dorotheyev · Journal of Physics A Mathematical and General · 1992
The stability of the phase with one-step replica symmetry breaking is studied in fully connected neural networks with modified pseudo-inverse interactions. The interaction matrix has an intermediate form between the Hebb learning rule and the pseudo-inverse one. At low temperature there is a region of parameters where the one-step replica-symmetry-broken solution exists. Fluctuations around this solution are analysed by a replica group representations approach and it is found that the solution is stable for all ranges of the parameters where it exists.