Approximation algorithms under uncertainty