Algorithm 1.

General steps of the optimization strategy for the case of homeostatic regulation (see text for details).

Homeostatic learning of synaptic weights (Equation 3) based on TL avalanches. Recording of Aava avalanches and their sizes L. Calculate mean squared deviation Δγ of size distribution P(L) from best-fit power law; if Δγ < Δγmax continue, otherwise restart at step 1. Check for convergence using Equation 12 with Δ = Δconv and averaging over np (np ≫ 1) perturbations per pattern; if retrieved states show large enough overlap with stored patterns, network has converged; if not, continue with step 5. Hebbian learning of synaptic weights using Equation 7; after each learning step check retrieval quality according to Equation 11 with Δ = Δhebb; if retrieval quality good enough start at step 1.