ReviBranch trains a reinforcement learning branching policy for MILP solvers on revived historical trajectories with densified rewards, reporting 4.0% fewer search nodes and 2.2% fewer LP iterations on large instances.
M.; Louveaux, Q.; and Wehenkel, L
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
ReviBranch: Deep Reinforcement Learning for Branch-and-Bound with Revived Trajectories
ReviBranch trains a reinforcement learning branching policy for MILP solvers on revived historical trajectories with densified rewards, reporting 4.0% fewer search nodes and 2.2% fewer LP iterations on large instances.