Automated hard instance search via black-box optimization

Abstract

In this work, we developed a protocol to generate hard instances using black-box opti- mization and a library for using the protocol. We applied these tools to several problems: SAT, Traveling Salesman, Hamiltonian Cycle and Flow-shop Scheduling. Using it, we found hard instances to each of these problems. It worked particularly well on Flow-shop Scheduling, showing an exponential scale of hardness with size and producing instances harder than any benchmark we had.