Is Transfer Learning Helpful for Neural Combinatorial Optimization Applied to Vehicle Routing Problems?

Recently, combinatorial optimization problems have aroused a great deal of interest in Machine Learning, leading to interesting advances Neural Combinatorial Optimization (NCO): the study data-driven solvers for NP-Hard based on neural networks. This paper studies benefit Transfer Learning NCO by evaluating how model training can be improved taking advantage knowledge learned while solving similar tasks. We focus, particular, two famous routing problems: Traveling Salesman Problem (TSP) and Vehicle Routing (VRP). The latter being generalization former, we effect applying from trained solve TSP learning Capacitated VRP (CVRP). present adaptations state-of-the-art implementing Learning. Our results extensive empirical experiments different settings show that may help speed up process more sample efficient.