Using TreeLSTMs to Solve Combinatorial Optimization Papers
MetadataShow full item record
We study the problem of building and traversing a search tree for combinatorial optimization problems, which are generally exponential in time complexity and often characterize NP-Hard problems. A typical approach for such problems is to build a solution search tree, with subproblems involving a combinatorial subset of a potential solution or a relaxation of the optimization constraint. A known approach for these problems is the Branch and Bound paradigm, which uses estimated bounds on the optimal solution of branches to improve the tree search. We propose the use of a TreeLSTM network in place of the solution search tree to improve the performance of the Branch and Bound approach.