A Deep Reinforcement Learning-Based Scheme for Solving Multiple Knapsack Problems

A knapsack problem is to select a set of items that maximizes the total profit selected while keeping weight no less than capacity knapsack. As generalized form with multiple knapsacks, multi-knapsack (MKP) disjointed for each To solve MKP, we propose deep reinforcement learning (DRL) based approach, which takes as input available capacities profits and weights items, normalized unselected determines next item be mapped largest capacity. expedite process, adopt Asynchronous Advantage Actor-Critic (A3C) policy model. The experimental results indicate proposed method outperforms random greedy methods achieves comparable performance an optimal in terms ratio sum, particularly when have non-linear relationship such quadratic forms.