Combinatorial optimization is a common theme in computer science which underlies a considerable variety of problems. While in general such problems are NP-Hard, from a practical point of view, locally optimal solutions can be quite useful. However, combinatorial problems require special solution strategies, and generic approaches like gradient methods for the continuous setting are hard to come by. We present a new stochastic optimization algorithm that can be applied in a generic fashion to a variety of combinatorial problems, and which only relies on function evaluations. In contrast to standard optimization algorithms, the new algorithm is highly parallelizable, and can rely on multiple evaluations simultaneously. The new algorithm is provided as part of a new Julia package that capitalizes on Julia’s parallel computation infrastructure in a manner that is completely transparent to the end user. Together with Julia’s excellent JIT compilation, this leads to a highly effective distributed optimizer for combinatorial problems that requires end users to provide nothing more than the function they seek to optimize.
