Abstract
The value-based reinforcement learning algorithms train agents by storing previous experience rewards, however, this simply sampling at the same probability results in a slow learning rate. In reality, the importance of each sample is not exactly the same. The use of prioritized experience replay greatly improves the learning rate of reinforcement learning, but good experiences and more effective strategies may be ignored or missed. In order to overcome two shortcomings, a Deep Q-learning with phased experience replay (MixDQN) is put forward in this article, where the priority is used to improve the training rate in the early stage of training and the random sampling in the later stage to make good use of good experience. Experiments with three classic control problems are based on OpenAI Gym. The experimental results prove that the MixDQN can enable an agent’s learning more stably, quickly and efficiently.
Supported in part by the National Natural Science Foundation of China under Grant 61572074 and in part by the China Scholarship Council under Grant 201706465028.
This is a preview of subscription content, log in via an institution.
Buying options
Tax calculation will be finalised at checkout
Purchases are for personal use only
Learn about institutional subscriptionsReferences
Watkins, C.J., Dayan, P.: Q-learning. Mach. Learn. 8(3–4), 279–292 (1992)
Sutton, R.S., Barto, A.G.: Reinforcement Learning: An Introduction. MIT Press, Cambridge (2018)
Sutton, R.S., McAllester, D.A., Singh, S.P., Mansour, Y.: Policy gradient methods for reinforcement learning with function approximation. In: Advances in Neural Information Processing Systems, pp. 1057–1063 (2000)
Konda, V.R., Tsitsiklis, J.N.: Onactor-critic algorithms. SIAM J. Control Optim. 42(4), 1143–1166 (2003)
Bhatnagar, S., Ghavamzadeh, M., Lee, M., Sutton, R.S.: Incremental natural actor-critic algorithms. In: Advances in Neural Information Processing Systems, pp. 105–112 (2008)
Grondman, I., Busoniu, L., Lopes, G.A., Babuska, R.: A survey of actor-critic reinforcement learning: standard and natural policy gradients. IEEE Trans. Syst. Man Cybern. Part C (Appl. Rev.) 42(6), 1291–1307 (2012)
Lillicrap, T.P., et al.: Continuous control with deep reinforcement learning. arXiv preprint arXiv:1509.02971 (2015)
Silver, D., Lever, G., Heess, N., Degris, T., Wierstra, D., Riedmiller, M.: Deterministic policy gradient algorithms (2014)
Mnih, V., et al.: Playing atari with deep reinforcement learning. arXiv preprint arXiv:1312.5602 (2013)
Narasimhan, K., Kulkarni, T., Barzilay, R.: Language understanding for text-based games using deep reinforcement learning. arXiv preprint arXiv:1506.08941 (2015)
Zelinka, M.: Using reinforcement learning to learn how to play text-based games. arXiv preprint arXiv:1801.01999 (2018)
Van Hasselt, H., Guez, A., Silver, D.: Deep reinforcement learning with double q-learning. In: Thirtieth AAAI Conference on Artificial Intelligence (2016)
Wang, Z., Schaul, T., Hessel, M., Van Hasselt, H., Lanctot, M., De Freitas, N.: Dueling network architectures for deep reinforcement learning. arXiv preprint arXiv:1511.06581 (2015)
Bellemare, M.G., Dabney, W., Munos, R.: A distributional perspective on reinforcement learning. In: Proceedings of the 34th International Conference on Machine Learning, vol. 70. pp. 449–458. JMLR.org (2017)
Fortunato, M., et al.: Noisy networks for exploration. arXiv preprint arXiv:1706.10295 (2017)
Mnih, V., et al.: Human-level control through deep reinforcement learning. Nature 518(7540), 529 (2015)
Atherton, L.A., Dupret, D., Mellor, J.R.: Memory trace replay: the shaping of memory consolidation by neuromodulation. Trends Neurosci. 38(9), 560–570 (2015)
McNamara, C.G., Tejero-Cantero, Á., Trouche, S., Campo-Urriza, N., Dupret, D.: Dopaminergic neurons promote hippocampal reactivation and spatial memory persistence. Nat. Neurosci. 17(12), 1658 (2014)
Hinton, G.E.: To recognize shapes, first learn to generate images. Progr. Brain Res. 165, 535–547 (2007)
Schaul, T., Quan, J., Antonoglou, I., Silver, D.: Prioritized experience replay. arXiv preprint arXiv:1511.05952 (2015)
Adam, S., Busoniu, L., Babuska, R.: Experience replay for real-time reinforcement learning control. IEEE Trans. Syst. Man Cybern. Part C (Appl. Rev.) 42(2), 201–212 (2011)
Wawrzyński, P.: Real-time reinforcement learning by sequential actor-critics and experience replay. Neural Netw. 22(10), 1484–1497 (2009)
Hou, Y., Liu, L., Wei, Q., Xu, X., Chen, C.: A novel DDPG method with prioritized experience replay. In: 2017 IEEE International Conference on Systems, Man, and Cybernetics (SMC), pp. 316–321. IEEE (2017)
Horgan, D., et al.: Distributed prioritized experience replay. arXiv preprint arXiv:1803.00933 (2018)
Sauthoff, G., Mhl, M., Janssen, S., Giegerich, R.: Bellmans GAP a language and compiler for dynamic programming in sequence analysis. Bioinformatics 29(5), 551–560 (2013)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2019 Springer Nature Singapore Pte Ltd.
About this paper
Cite this paper
Wang, H., Zeng, F., Tu, X. (2019). Deep Q-Learning with Phased Experience Cooperation. In: Sun, Y., Lu, T., Yu, Z., Fan, H., Gao, L. (eds) Computer Supported Cooperative Work and Social Computing. ChineseCSCW 2019. Communications in Computer and Information Science, vol 1042. Springer, Singapore. https://doi.org/10.1007/978-981-15-1377-0_58
Download citation
DOI: https://doi.org/10.1007/978-981-15-1377-0_58
Published:
Publisher Name: Springer, Singapore
Print ISBN: 978-981-15-1376-3
Online ISBN: 978-981-15-1377-0
eBook Packages: Computer ScienceComputer Science (R0)