LSEARCH & HSEARCH: non-incremental, do not profit from experience with previous tasks (unnecessary for asymptotic optimality): academic exercises illustrating practical irrelevance of O() notation.
Incremental search: Reduce constants by experience?
Adaptive LSEARCH (Schmidhuber et al. 1995-1997; compare Solomonoff’s ideas 1986-2002): after every solved task, increase probabilities of successful programs through learning rate.
Fixed way of adjusting P - does not learn it!
Not necessarily bias-optimal - may overfit!
Next: right way of doing it!
Back to J. Schmidhuber's OOPS page