F-CutLearn: A feasibility-guaranteed cut learning method for network design problems

doi:10.1016/j.tre.2026.104856

F-CutLearn: A feasibility-guaranteed cut learning method for network design problems

2026-04-15

0

PRE

AI

DOI：10.1016/j.tre.2026.104856

Original

Original request for help

Save

Abstract

En 中文

• Propose F-CutLearn, a novel learning-optimization integration framework for improving solution performance of NDP variants. • Develop a provably feasible cut generation mechanism with tunable aggressiveness and automatic detection of infeasible cuts. • The F-CutLearn method can provably yield globally optimal solutions under certain conditions. • Design a tailored GCN predictor to estimate variable selection probabilities, effectively supporting variable fixing and cut generation. • Verify the computational performance, generalizability, and effectiveness of key components through extensive experimental evaluation.

Keywords:

F-CutLearn

cut generation

network design problems

graph neural networks

variable selection

F-CutLearn: A feasibility-guaranteed cut learning method for network design problems

Abstract

Journal

Organization