We study the stochastic behavior of a single self-exciting model neuro
n with additive noise, a system that has bistable stochastic dynamics.
We use Langevin and Fokker-Planck equations to obtain analytical expr
essions for the stationary distribution of activities and for the cros
sing rate between two stable states. We adjust the parameters in these
expressions to fit observed histograms of neural activity, thus obtai
ning what we call an ''effective single neuron'' for a given network.
We construct an effective single neuron from an activity histogram of
a representative hidden neuron in a recurrent learning network. We als
o compare our result with an effective single neuron previously obtain
ed analytically through the adiabatic elimination approximation.