Fastest way of making data
FAST algorithm: Execute one instruction of n-th program every 2n steps on average; separate program outputs by blanks
Suppose fastest program for x computes first n bits of x within f(n) steps. Then FAST will take only O(f(n)) steps.
Costs asymptotically minimal
Levinís optimal universal search (1973): seek solution y to problem g(y)=x. Run FAST and check for each output whether it is a solution.
Back to J. Schmidhuber's Speed Prior page