Spike Response Model of neurons in the prediction error layer.

Each postsynaptic neuron receives an input current I_i. This current is integrated, with membrane time constant τ, to obtain the input potential h_i. The actual membrane potential of the neuron u_i is the combination of both the input potential and a refractory function η, where η is a strong negative potential activated after a spike, forcing the neuron to stay silent for a while. The spike times are then randomly drawn with probability ϕ(u_i) generating the spike train of neuron i.