Abstract
With remarkable performance and extensive applications, reinforcement learning is becoming one of the most popular learning techniques. Often, the policy \(\pi ^*\) released by reinforcement learning model may contain sensitive information, and an adversary can infer demographic information through observing the output of the environment. In this paper, we formulate differential privacy in reinforcement learning contexts, design mechanisms for \(\epsilon \)-greedy and Softmax in the K-armed bandit problem to achieve differentially private guarantees. Our implementation and experiments illustrate that the output policies are under good privacy guarantees with a tolerable utility cost.
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 subscriptionsNotes
- 1.
- 2.
To distinguish the \(\epsilon \) in differential privacy and \(\epsilon \)-greedy, the \(\epsilon \) in \(\epsilon \)-greedy will be replaced by \(\epsilon _{rl}\) in the remainder of the article, namely \(\epsilon _{rl}\)-greedy.
References
Abadi, M., et al.: Deep learning with differential privacy. In: Proceedings of the 2016 ACM SIGSAC Conference on Computer and Communications Security, pp. 308–318. ACM (2016)
Balle, B., Gomrokchi, M., Precup, D.: Differentially private policy evaluation. In: International Conference on Machine Learning, pp. 2130–2138 (2016)
Berry, D.A., Fristedt, B.: Bandit Problems: Sequential Allocation of Experiments (Monographs on Statistics and Applied Probability), vol. 5, pp. 71–87. Chapman and Hall, London (1985)
Dwork, C.: Differential privacy: a survey of results. In: Agrawal, M., Du, D., Duan, Z., Li, A. (eds.) TAMC 2008. LNCS, vol. 4978, pp. 1–19. Springer, Heidelberg (2008). https://doi.org/10.1007/978-3-540-79228-4_1
Dwork, C.: Differential privacy. In: van Tilborg, H.C.A., Jajodia, S. (eds.) Encyclopedia of Cryptography and Security, pp. 338–340. Springer, Boston (2011). https://doi.org/10.1007/978-1-4419-5906-5
Dwork, C.: A firm foundation for private data analysis. Commun. ACM 54(1), 86–95 (2011)
Dwork, C., McSherry, F., Nissim, K., Smith, A.: Calibrating noise to sensitivity in private data analysis. In: Halevi, S., Rabin, T. (eds.) TCC 2006. LNCS, vol. 3876, pp. 265–284. Springer, Heidelberg (2006). https://doi.org/10.1007/11681878_14
Dwork, C., Roth, A., et al.: The algorithmic foundations of differential privacy. Found. Trends® Theoret. Comput. Sci. 9(3–4), 211–407 (2014)
Erlingsson, Ú., Pihur, V., Korolova, A.: RAPPOR: randomized aggregatable privacy-preserving ordinal response. In: Proceedings of the 2014 ACM SIGSAC Conference on Computer and Communications Security, pp. 1054–1067. ACM (2014)
Fredrikson, M., Jha, S., Ristenpart, T.: Model inversion attacks that exploit confidence information and basic countermeasures. In: Proceedings of the 22nd ACM SIGSAC Conference on Computer and Communications Security, pp. 1322–1333. ACM (2015)
Friedman, A., Schuster, A.: Data mining with differential privacy. In: Proceedings of the 16th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD 2010, pp. 493–502. ACM, New York (2010)
Gittins, J., Glazebrook, K., Weber, R.: Multi-Armed Bandit Allocation Indices. Wiley, Hoboken (2011)
Jaakkola, T., Singh, S.P., Jordan, M.I.: Reinforcement learning algorithm for partially observable Markov decision problems. In: Advances in Neural Information Processing Systems, pp. 345–352 (1995)
Ji, Z., Lipton, Z.C., Elkan, C.: Differential privacy and machine learning: a survey and review. arXiv preprint arXiv:1412.7584 (2014)
Kasiviswanathan, S.P., Lee, H.K., Nissim, K., Raskhodnikova, S., Smith, A.: What can we learn privately? SIAM J. Comput. 40(3), 793–826 (2011)
Kuleshov, V., Precup, D.: Algorithms for multi-armed bandit problems. arXiv preprint arXiv:1402.6028 (2014)
LeCun, Y., Bengio, Y., Hinton, G.: Deep learning. Nature 521(7553), 436 (2015)
Liu, Q., Li, P., Zhao, W., Cai, W., Yu, S., Leung, V.C.M.: A survey on security threats and defensive techniques of machine learning: a data driven view. IEEE Access 6, 12103–12117 (2018). https://doi.org/10.1109/ACCESS.2018.2805680
Mishra, N., Thakurta, A.: (Nearly) optimal differentially private stochastic multi-arm bandits. In: Proceedings of the Thirty-First Conference on Uncertainty in Artificial Intelligence, pp. 592–601. AUAI Press (2015)
Nissim, K., Raskhodnikova, S., Smith, A.: Smooth sensitivity and sampling in private data analysis. In: Proceedings of the Thirty-ninth Annual ACM Symposium on Theory of Computing, pp. 75–84. ACM (2007)
Sakuma, J., Kobayashi, S., Wright, R.N.: Privacy-preserving reinforcement learning. In: Proceedings of the 25th International Conference on Machine learning, pp. 864–871. ACM (2008)
Salem, A., Zhang, Y., Humbert, M., Berrang, P., Fritz, M., Backes, M.: ML-Leaks: model and data independent membership inference attacks and defenses on machine learning models. arXiv preprint arXiv:1806.01246 (2018)
Shokri, R., Stronati, M., Song, C., Shmatikov, V.: Membership inference attacks against machine learning models. In: 2017 IEEE Symposium on Security and Privacy (SP), pp. 3–18. IEEE (2017)
Sun, S.: A survey of multi-view machine learning. Neural Comput. Appl. 23(7–8), 2031–2038 (2013)
Sutton, R.S., Barto, A.G.: Reinforcement Learning: An Introduction. MIT Press, Cambridge (2018)
Tang, X., Zhu, L., Shen, M., Du, X.: When homomorphic cryptosystem meets differential privacy: training machine learning classifier with privacy protection. arXiv preprint arXiv:1812.02292 (2018)
Tossou, A.C., Dimitrakakis, C.: Algorithms for differentially private multi-armed bandits. In: Thirtieth AAAI Conference on Artificial Intelligence (2016)
Wu, X., Fredrikson, M., Jha, S., Naughton, J.F.: A methodology for formalizing model-inversion attacks. In: 2016 IEEE 29th Computer Security Foundations Symposium (CSF), pp. 355–370. IEEE (2016)
Acknowledgement
This research was financially supported by the National Key Research and Development Plan (2018YFB1004101), China Postdoctoral Science Foundation Funded Project (2019M650606), Key Lab of Information Network Security, Ministry of Public Security (C19614), Special Fund on Education and Teaching Reform of BESTI (jy201805), the Fundamental Research Funds for the Central Universities (328201910), Key Laboratory of Network Assessment Technology of Institute of Information Engineering, Chinese Academy of Sciences.
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Appendices
Appendices
A. Proof of Algorithm 1
B. Analysis on Total Time Steps
We analyze the total time steps n needed to get a accurate approximation of \(q_*(a)=\mathbb {E}_t[R_t\mid A_t=a]\) for every action a. The analysis is presented in two aspects. The first aspect is to consider how many times m we need to select action a to get an accurate approximation of \(q_*(a)\). The second aspect is to analyze the value of n needed to guarantee m times sampling of a. We start with the first aspect.
Recall that w.l.o.g. \(R_t\) is assumed to be within \([0,\varLambda ]\). Consider the Doob martingale \(B_i=\mathbb {E}[\tfrac{1}{m}(X_1+X_2+\cdots +X_m)\mid X_1,X_2,\dotsc ,X_i]\), where \(X_i\) is the numerical reward received when we select action a. Note that \(X_i\)’s are i.i.d. The stochastic process \(B_0,B_1,\dotsc \) is a martingale w.r.t. \(X_i\) as \(\mathbb {E}(|B_j|)\le \varLambda <\infty \) and
holds. Note that \(B_0=\mathbb {E}[\tfrac{1}{m}(X_1+X_2+\cdots +X_m)]=q_*(a)\) and \(B_m=\tfrac{1}{m}(X_1+X_2+\cdots +X_m)\), which is just Q(a) in the algorithm. We also have \(|B_{j+1}-B_j|\le \tfrac{\varLambda }{m}\) as \(X_i\)’s are independent and
holds. According to the Azuma-Hoeffding inequality, we then have \(P(|Q(a)-q_*(a)|\ge \lambda \varLambda )\le 2\exp {(-\tfrac{(\lambda \varLambda )^2}{2\varLambda ^2/m})}=2\exp {(-\tfrac{\lambda ^2m}{2})}\), where \(\lambda \in (0,1)\).
Now we consider the second aspect. Instead of counting the number of times a particular action a is selected, we consider the number of times selecting a when \(rand()<\epsilon _{rl}\), which servers as a lower bound of the actual counting. Denote the latter as \(Y_i\). So in every time step, action a is selected with probability \(\tfrac{\epsilon _{rl}}{K}\). Applying Chernoff bound, if the total time steps \(n=\tfrac{2mK}{\epsilon _{rl}}\) we have \(P(Y_i<m)\le \exp {(-\tfrac{m}{4})}\). Applying the union bound we have the probability that every action is selected at least m times is at least \(1-K\exp {(-\tfrac{m}{4})}\). With conditional probability the final probability that every estimate Q(a) is within \(\lambda \varLambda \) of \(q_*(a)\) is
if we choose \(m=\tfrac{2}{\lambda ^2}(c'+1)\ln K\), where \(c'\), c are constants. In a word if the total time steps \(n\ge \tfrac{4K}{\epsilon _{rl}\lambda ^2}(c'+1)\ln K\) then w.h.p. every action value estimate is within a preferable range of the true action value.
C. Traverse in Algorithm 2
Rights and permissions
Copyright information
© 2020 Springer Nature Switzerland AG
About this paper
Cite this paper
Ma, P., Wang, Z., Zhang, L., Wang, R., Zou, X., Yang, T. (2020). Differentially Private Reinforcement Learning. In: Zhou, J., Luo, X., Shen, Q., Xu, Z. (eds) Information and Communications Security. ICICS 2019. Lecture Notes in Computer Science(), vol 11999. Springer, Cham. https://doi.org/10.1007/978-3-030-41579-2_39
Download citation
DOI: https://doi.org/10.1007/978-3-030-41579-2_39
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-030-41578-5
Online ISBN: 978-3-030-41579-2
eBook Packages: Computer ScienceComputer Science (R0)