Algorithm 1.

Growing When Required

1:	Start with a set A consisting of two map nodes, n1 and n2, at random positions.
2:	Initialize an empty set of connections C = ∅.
3:	At each iteration, generate an input sample ξ according to the input distribution P(ξ).
4:	For each node i calculate the distance from the input ||ξ − wi||.
5:	Select the best matching node and the second-best matching node such that: s = arg minn ∈ A||ξ − wn||, t = arg minn ∈ A/{s} ||ξ − wn||.
6:	Create a connection C = C ∪ {(s, t)} if it does not exist and set its age to 0.
7:	Calculate the activity of the best matching unit: a = exp(−||ξ − ws||).
8:	If a < activity threshold aT and firing counter < firing threshold fT then: Add a new node between s and t: A = A ∪ {(r)} Create the weight vector: wr = 0.5 · (ws + ξ) Create edges and remove old edge: C = C ∪ {(r, s), (r, t)} and C = C/{(s, t)}.
9:	Else, i.e., no new node is added, adapt the positions of the winning node and its neighbours i: Δws = ϵb · hs · (ξ − ws) Δwi = ϵn · hi · (ξ − wi) where 0 < ϵn < ϵb < 1 and hs is the value of the firing counter for node s.
10:	Increment the age of all edges connected to s: age(s,i) = age(s,i) + 1.
11:	Reduce the firing counters according to Equation (2): $h_s(t)=h_0-\frac{S(t)}{{\alpha }_b}·(1-exp(-{\alpha }_{b}t/{\tau }_b))$ $h_i(t)=h_0-\frac{S(t)}{{\alpha }_n}·(1-exp(-{\alpha }_{n}t/{\tau }_n))$.
12:	Remove all edges with ages larger than amax and remove nodes without edges.
13:	If the stop criterion is not met, go to step 3.