Rigorous Results and Critical Capacity for a Short-Term Memory Model
1996, v.2, №4, 539-554
We consider a palimpsest neural network with the property of short term memory. We will prove the self-averaging (s.a.) property of the free-energy and under the assumption of the s.a. property of the Edwards-Anderson parameter $q$, we will derive the saddle-point equations in the case of replica-symmetry but without using the replica trick. We will numerically find the capacity of the model with marginalist learning using a new approach, and we will get the same result obtained by MГ©zard et al. [M. MГ©zard, J.P. Nadal and G. Toulouse. Solvable models of working memories. J. Physique, 1986, v. 47, 1457-1462] for zero temperature.
Keywords: critical capacity,self-averaging,Edwards-Anderson parameter,saddle-point equation,short-term memory