TY - GEN

T1 - A State Aggregation Approach for Solving Knapsack Problem with Deep Reinforcement Learning

AU - Refaei Afshar, Reza

AU - Zhang, Yingqian

AU - Firat, Murat

AU - Kaymak, Uzay

PY - 2020

Y1 - 2020

N2 - This paper proposes a Deep Reinforcement Learning (DRL) approach for solving knapsack problem. The proposed method consists of a state aggregation step based on tabular reinforcement learning to extract features and construct states. The state aggregation policy is applied to each problem instance of the knapsack problem, which is used with Advantage Actor Critic (A2C) algorithm to train a policy through which the items are sequentially selected at each time step. The method is a constructive solution approach and the process of selecting items is repeated until the final solution is obtained. The experiments show that our approach provides close to optimal solutions for all tested instances, outperforms the greedy algorithm, and is able to handle larger instances and more flexible than an existing DRL approach. In addition, the results demonstrate that the proposed model with the state aggregation strategy not only gives better solutions but also learns in less timesteps, than the one without state aggregation.

AB - This paper proposes a Deep Reinforcement Learning (DRL) approach for solving knapsack problem. The proposed method consists of a state aggregation step based on tabular reinforcement learning to extract features and construct states. The state aggregation policy is applied to each problem instance of the knapsack problem, which is used with Advantage Actor Critic (A2C) algorithm to train a policy through which the items are sequentially selected at each time step. The method is a constructive solution approach and the process of selecting items is repeated until the final solution is obtained. The experiments show that our approach provides close to optimal solutions for all tested instances, outperforms the greedy algorithm, and is able to handle larger instances and more flexible than an existing DRL approach. In addition, the results demonstrate that the proposed model with the state aggregation strategy not only gives better solutions but also learns in less timesteps, than the one without state aggregation.

UR - http://www.acml-conf.org/2020/video/paper/refaei-afshar20a

M3 - Conference contribution

T3 - Proceedings of Machine Learning Research

SP - 81

EP - 96

BT - Proceedings of The 12th Asian Conference on Machine Learning (ACML2020)

T2 - 12th Asian Conference on Machine Learning (virtual)

Y2 - 18 November 2020 through 20 November 2020

ER -