Swarm Intelligence

the “flipped” value is kept, or else it is set back to what it was. The greedy

algorithm goes through the entire bitstring in this way, flipping, eval-

uating, and choosing, once for every position, 10 times if N = 10. This al-

gorithm will result in improvement unless the search happened to ran-

domly start on a really good corner. What the greedy algorithm doesn’t

do is to capitalize on patterns of bits that work together. The difficulty of

this can be gleaned from the discussion of NK landscapes; when K is

high, the effect of a site depends on the states of many other sites. The

greedy algorithm may be a reasonable, quick first-guess approach to opti ...