Toward an Optimization Solver Based on Neighborhood Search
Frédéric Gardi, Thierry Benoist, Julien Darlay, Bertrand Estellon, Romain Megel · 2014
This chapter proposes a general-purpose optimization solver founded on a radically different architecture. Instead of embedding local search ingredients into tree search, it proposes to use neighborhood search as a global search strategy. Nevertheless, neighborhood search does not mean restricting the search to small neighborhoods and abandoning tree search. Very large-scale neighborhoods, namely neighborhoods of exponential size, can be explored through tree search or specific algorithms. Local search is a technique which is actually used in continuous optimization, under another name: direct search. Finally, the chapter outlines the target architecture of LocalSolver over the long term.