Training stochastic stabilized networks by dynamics neutral growth