Main result (informally)
E.g., n-th task for NP-hard optimization: find better approximation than best found so far
Or n-th teacher-given task: find program that computes k! for k=1,2,…,n (so n-th task is a task set)
Bias in form of probability distribution on programs
OOPS is a time-optimal way of solving one task after another, exploiting solutions to previous tasks when possible, never suffering from previous solutions
Asymptotically as fast as any other search algorithm
With near-optimal constants (normally hidden in O()-notation): essentially, we lose a factor of 8 relative to unknown theoretically optimal method
Back to J. Schmidhuber's OOPS page