Designing algorithms by sampling Original Research Article

This paper describes a method for empirical algorithm design, called database learning, using which an algorithm is constructed based on the solutions produced by an oracle on instances from a given input domain. We present experimental results of applying the strategy to designing heuristics for the problem of constructing a maximum independent set of a given graph. The heuristic designed by database learning is compared with the best general-purpose heuristic for the problem.