Abstract
Batch Reinforcement Learning (Batch RL) consists in training a policy using trajectories collected with another policy, called the behavioural policy. Safe policy improvement (SPI) provides guarantees with high probability that the trained policy performs better than the behavioural policy, also called baseline in this setting. Previous work shows that the SPI objective improves mean performance as compared to using the basic RL objective, which boils down to solving the MDP with maximum likelihood (Laroche et al. 2019). Here, we build on that work and improve more precisely the SPI with Baseline Bootstrapping algorithm (SPIBB) by allowing the policy search over a wider set of policies. Instead of binarily classifying the state-action pairs into two sets (the uncertain and the safe-to-train-on ones), we adopt a softer strategy that controls the error in the value estimates by constraining the policy change according to the local model uncertainty. The method can take more risks on uncertain actions all the while remaining provably-safe, and is therefore less conservative than the state-of-the-art methods. We propose two algorithms (one optimal and one approximate) to solve this constrained optimization problem and empirically show a significant improvement over existing SPI algorithms both on finite MDPS and on infinite MDPs with a neural network function approximation.
K. Nadjahi and R. Laroche—Equal contribution.
K. Nadjahi—Work done while interning at Microsoft Research Montréal.
Finite MDPs code available at https://github.com/RomainLaroche/SPIBB.
SPIBB-DQN code available at https://github.com/rems75/SPIBB-DQN.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
References
Bellemare, M., Srinivasan, S., Ostrovski, G., Schaul, T., Saxton, D., Munos, R.: Unifying count-based exploration and intrinsic motivation. In: Proceedings of the 29th Advances in Neural Information Processing Systems (NIPS) (2016)
Borgwardt, K.H.: The Simplex Method: A Probabilistic Analysis. Springer, Heidelberg (1987). https://doi.org/10.1007/978-3-642-61578-8
Burda, Y., Edwards, H., Storkey, A., Klimov, O.: Exploration by random network distillation. In: Proceedings of the 7th International Conference on Learning Representations (ICLR) (2019)
Dantzig, G.: Linear Programming and Extensions. Rand Corporation Research Study. Princeton Univ. Press, Princeton (1963)
Dantzig, G.B., Thapa, M.N.: Linear Programming 2: Theory and Extensions. Springer, New York (2003). https://doi.org/10.1007/b97283
Ernst, D., Geurts, P., Wehenkel, L.: Tree-based batch mode reinforcement learning. J. Mach. Learn. Res. 6, 503–556 (2005)
Fox, L., Choshen, L., Loewenstein, Y.: Dora the explorer: directed outreaching reinforcement action-selection. In: Proceedings of the 6th International Conference on Learning Representations (ICLR) (2018)
Geist, M., Scherrer, B., Pietquin, O.: A theory of regularized Markov decision processes. In: Proceedings of the 36th International Conference on Machine Learning (ICML) (2019)
Gondzio, J.: Interior point methods 25 years later. Eur. J. Oper. Res. 218(3), 587–601 (2012)
Guez, A., Vincent, R.D., Avoli, M., Pineau, J.: Adaptive treatment of epilepsy via batch-mode reinforcement learning. In: Proceedings of the 23rd AAAI Conference on Artificial Intelligence, pp. 1671–1678 (2008)
He, K., Zhang, X., Ren, S., Sun, J.: Delving deep into rectifiers: surpassing human-level performance on imagenet classification. arXiv preprint arXiv:1502.01852 (2015)
Iyengar, G.N.: Robust dynamic programming. Math. Oper. Res. 30(2), 257–280 (2005)
Kakade, S., Langford, J.: Approximately optimal approximate reinforcement learning. In: Proceedings of the 19th International Conference on Machine Learning (ICML), vol. 2, pp. 267–274 (2002)
Klee, V., Minty, G.J.: How good is the simplex algorithm? In: Shisha, O. (ed.) Inequalities, vol. III, pp. 159–175. Academic Press, New York (1972)
Lange, S., Gabel, T., Riedmiller, M.: Batch reinforcement learning. In: Wiering, M., van Otterlo, M. (eds.) Reinforcement Learning. Adaptation, Learning, and Optimization, vol. 12, pp. 45–73. Springer, Heidelberg (2012). https://doi.org/10.1007/978-3-642-27645-3_2
Laroche, R., Trichelair, P., Tachet des Combes, R.: Safe policy improvement with baseline bootstrapping. In: Proceedings of the 36th International Conference on Machine Learning (ICML) (2019)
Mnih, V., et al.: Human-level control through deep reinforcement learning. Nature 518(7540), 529 (2015)
Nilim, A., El Ghaoui, L.: Robust control of Markov decision processes with uncertain transition matrices. Oper. Res. 53(5), 780–798 (2005)
Paszke, A., et al.: Automatic differentiation in PyTorch. In: NIPS-W (2017)
Petrik, M., Ghavamzadeh, M., Chow, Y.: Safe policy improvement by minimizing robust baseline regret. In: Proceedings of the 29th Advances in Neural Information Processing Systems (NIPS) (2016)
Riedmiller, M.: Neural fitted Q iteration – first experiences with a data efficient neural reinforcement learning method. In: Gama, J., Camacho, R., Brazdil, P.B., Jorge, A.M., Torgo, L. (eds.) ECML 2005. LNCS (LNAI), vol. 3720, pp. 317–328. Springer, Heidelberg (2005). https://doi.org/10.1007/11564096_32
Schulman, J., Levine, S., Abbeel, P., Jordan, M., Moritz, P.: Trust region policy optimization. In: Proceedings of the 32nd International Conference on Machine Learning (ICML) (2015)
Schulman, J., Wolski, F., Dhariwal, P., Radford, A., Klimov, O.: Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347 (2017)
Simão, T.D., Spaan, M.T.J.: Safe policy improvement with baseline bootstrapping in factored environments. In: Proceedings of the 32nd AAAI Conference on Artificial Intelligence (2019)
Singh, S.P., Kearns, M.J., Litman, D.J., Walker, M.A.: Reinforcement learning for spoken dialogue systems. In: Proceedings of the 13th Advances in Neural Information Processing Systems (NIPS), pp. 956–962 (1999)
Sutton, R.S., Barto, A.G.: Reinforcement Learning: An Introduction. The MIT Press, Cambridge (1998)
Thomas, P.S.: Safe reinforcement learning. Ph.D. thesis, Stanford university (2015)
Tieleman, T., Hinton, G.: Lecture 6.5-rmsprop: divide the gradient by a running average of its recent magnitude. COURSERA: Neural Netw. Mach. Learn. 4(2), 26–31 (2012)
van Hasselt, H., Guez, A., Silver, D.: Deep reinforcement learning with double q-learning. CoRR, abs/1509.06461 (2015)
Weissman, T., Ordentlich, E., Seroussi, G., Verdu, S., Weinberger, M.J.: Inequalities for the L1 deviation of the empirical distribution. Hewlett-Packard Labs, Technical report (2003)
Author information
Authors and Affiliations
Corresponding authors
Editor information
Editors and Affiliations
1 Electronic supplementary material
Below is the link to the electronic supplementary material.
Rights and permissions
Copyright information
© 2020 Springer Nature Switzerland AG
About this paper
Cite this paper
Nadjahi, K., Laroche, R., Tachet des Combes, R. (2020). Safe Policy Improvement with Soft Baseline Bootstrapping. In: Brefeld, U., Fromont, E., Hotho, A., Knobbe, A., Maathuis, M., Robardet, C. (eds) Machine Learning and Knowledge Discovery in Databases. ECML PKDD 2019. Lecture Notes in Computer Science(), vol 11908. Springer, Cham. https://doi.org/10.1007/978-3-030-46133-1_4
Download citation
DOI: https://doi.org/10.1007/978-3-030-46133-1_4
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-46132-4
Online ISBN: 978-3-030-46133-1
eBook Packages: Computer ScienceComputer Science (R0)
