This paper introduces Neural Combinatorial Optimization, a framework that uses reinforcement learning and neural networks to solve combinatorial optimization problems, focusing on the Traveling Salesman Problem (TSP). The authors train a recurrent neural network to predict city permutations based on their coordinates, using negative tour length as the reward signal. They compare learning parameters on training graphs against individual test graphs, achieving close to optimal results on 2D Euclidean graphs with up to 100 nodes. The method is also applied to the Knapsack problem, achieving optimal solutions for instances with up to 200 items. The paper discusses the architecture of the pointer network, the training process using policy gradients, and various search strategies. Experiments show that Neural Combinatorial Optimization outperforms supervised learning and heuristic methods, demonstrating its effectiveness in solving complex combinatorial optimization problems.This paper introduces Neural Combinatorial Optimization, a framework that uses reinforcement learning and neural networks to solve combinatorial optimization problems, focusing on the Traveling Salesman Problem (TSP). The authors train a recurrent neural network to predict city permutations based on their coordinates, using negative tour length as the reward signal. They compare learning parameters on training graphs against individual test graphs, achieving close to optimal results on 2D Euclidean graphs with up to 100 nodes. The method is also applied to the Knapsack problem, achieving optimal solutions for instances with up to 200 items. The paper discusses the architecture of the pointer network, the training process using policy gradients, and various search strategies. Experiments show that Neural Combinatorial Optimization outperforms supervised learning and heuristic methods, demonstrating its effectiveness in solving complex combinatorial optimization problems.