Algorithm for iterated DDP
initialise the bias matrix Q
select the seed pairs using Q
place the selected pairs in I
R := {0} Set high level matrix to zero
for each pair (ai,bj) in I do
accumulate the low level results to R
update Q using old Q and R
(s,P) := DPQ(A,B) High level DP
select new pairs (ai,bj) based on Q
until termination criterion is satisfied