Incremental search: Update probability distribution by experience?
Heuristic ideas: Adaptive LS (Schmidhuber, 1995-1997; compare Solomonoff’s related ideas 1986-2001): increase probabilities of successful programs through learning rate. Not necessarily optimal.
Next: non-heuristic approach; bias updates computed by the self-improving system itself.
Back to J. Schmidhuber's OOPS page