Asymptotic behaviour in learning from stochastic examples: one-step RSB calculation of the learning curve
Tatsuya Uezu, Yoshiyuki Kabashima · Journal of Physics A Mathematical and General · 1996
The asymptotic learning curve of a N-dimensional stochastic learning model is calculated in the statistical mechanical framework. For the N = 1 case, Kabashima and Shinomoto (1992) gave a power law with exponent . However, this is inconsistent with the RS solution with the exponent obtained by Györgyi and Tishby (1990). We show that the one-step RSB solution is consistent with Kabashima and Shinomoto's result up to a logarithmic correction.