Adaptive large neighborhood search for mixed integer programming

Abstract Large Neighborhood Search (LNS) heuristics are among the most powerful but also expensive for mixed integer programs (MIP). Ideally, a solver adaptively concentrates its limited computational budget by learning which LNS work best MIP problem at hand. To this end, introduces Adaptive (ALNS) MIP, primal heuristic that acts as framework eight popular such Local Branching and Relaxation Induced (RINS). We distinguish available their individual search spaces, we call auxiliary problems . The decision should be executed is guided selection strategies multi armed bandit , related optimization during suitable actions have to chosen maximize reward function. In paper, propose an LNS-specific function learn between based on successful calls failures. A second, algorithmic enhancement generic variable fixing prioritization, ALNS employs adjust subproblem complexity needed. This particularly useful some do not fix variables themselves. proposed has been implemented within SCIP. An extensive study conducted compare different our large set of publicly instances from MIPLIB Coral benchmark sets. results simulation used calibrate parameters strategies. second experiment shows benefits