Problem: Universal M(x) not recursive
M(x): Optimal but infeasible inductive inference
More feasible watered-down approaches:
Minimum description length (MDL, Wallace & Boulton 1968, Rissanen 1986)
Priors based on (polynomial) time bounds (Li & Vitanyi 1997)
But lack of obvious, plausible, a priori justification
Which prior reflects “correct” simplicity bias?
Next: “simple” means “quickly computable”!
Back to J. Schmidhuber's Speed Prior page