A First Guess is Rarely the Final Answer: Learning to Search in the Traveling Salesperson Problem
Researchers have developed NICO-TSP, a novel neural improvement framework designed to enhance solutions for the Traveling Salesperson Problem (TSP). Unlike previous methods that focus on generating a single solution, NICO-TSP learns a policy to iteratively refine existing tours through local modifications. This approach demonstrates superior performance and efficiency compared to existing learned and heuristic search baselines, particularly in step-efficiency and generalization to larger problem instances. AI
IMPACT Introduces a new learned search method for combinatorial optimization problems like TSP, potentially improving efficiency and generalization.