Given problem set R; search space C; bias = distribution P(q|r) on candidates q in C given r in R; procedure for creating and testing q on any r within time t(q,r).
Searcher is n-bias-optimal if for any maximal search time T it will solve any r in R with solution p in C such that t(p,r) ? P(p|r) T/n
LSEARCH near-bias-optimal: Time-shares total search time among all solution candidates in proportion to their probabilities; loses at most factor of 2 or so...
Back to J. Schmidhuber's OOPS page