Abstract
This paper addresses the problem of non-Bayesian learning over multi-agent networks, where agents repeatedly collect partially informative observations about an unknown state of the world, and try to collaboratively learn the true state out of m alternatives. We focus on the impact of adversarial agents on the performance of consensus-based non-Bayesian learning, where non-faulty agents combine local learning updates with consensus primitives. In particular, we consider the scenario where an unknown subset of agents suffer Byzantine faults—agents suffering Byzantine faults behave arbitrarily. We propose two learning rules. In our learning rules, each non-faulty agent keeps a local variable which is a stochastic vector over the m possible states. Entries of this stochastic vector can be viewed as the scores assigned to the corresponding states by that agent. We say a non-faulty agent learns the underlying truth if it assigns one to the true state and zeros to the wrong states asymptotically.
-
In our first update rule, each agent updates its local score vector as (up to normalization) the product of (1) the likelihood of the cumulative private signals and (2) the weighted geometric average of the score vectors of its incoming neighbors and itself. Under reasonable assumptions on the underlying network structure and the global identifiability of the network, we show that all the non-faulty agents asymptotically learn the true state almost surely.
-
We propose a modified variant of our first learning rule whose complexity per iteration per agent is \(O(m^2 n \log n)\), where n is the number of agents in the network. In addition, we show that this modified learning rule works under a less restrictive network identifiability condition.
Similar content being viewed by others
Notes
This is because the upper bound f can be learned via long-time performance statistics, whereas, the actual size of \({\mathcal {F}}\) varies across executions, and may be impossible to be predicted in some applications.
References
Su, L., Vaidya, N.H.: Non-Bayesian learning in the presence of Byzantine agents. To appear in Proceedings of ACM Symposium on Distributed Computing (DISC) (2016)
Chamberland, J.-F., Veeravalli, V.V.: Decentralized detection in sensor networks. IEEE Trans. Signal Process. 51, 407–416 (2003)
Gale, D., Kariv, S.: Bayesian learning in social networks. Games Econ. Behav. 45, 329–346 (2003)
Jadbabaie, A., Molavi, P., Sandroni, A., Tahbaz-Salehi, A.: Non-Bayesian social learning. Games Econ. Behav. 76, 210–225 (2012)
Tsitsiklis, J.: Decentralized detection by a large number of sensors. Math. Control Signals Syst. 1, 167–182 (1988)
Tsitsiklis, J.N.: Decentralized detection. In: Poor, H.V., Thomas, J.B. (eds.) Advances in Statistical Signal Processing. JAI Press, Greenwich (1993)
Varshney, P.K.: Distributed Detection and Data Fusion. Springer, Berlin (2012)
Wong, E., Hajek, B.: Stochastic Processes in Engineering Systems. Springer, Berlin (2012)
Bajović, D., Jakovetić, D., Moura, J.M., Xavier, J., Sinopoli, B.: Large deviations performance of consensus+ innovations distributed detection with non-Gaussian observations. IEEE Trans. Signal Process. 60(11), 5987–6002 (2012)
Cattivelli, F.S., Sayed, A.H.: Distributed detection over adaptive networks using diffusion adaptation. IEEE Trans. Signal Process. 59(5), 1917–1932 (2011)
Jakovetić, D., Moura, J.M., Xavier, J.: Distributed detection over noisy networks: large deviations analysis. IEEE Trans. Signal Process. 60(8), 4306–4320 (2012)
Jadbabaie, A., Molavi, P., Tahbaz-Salehi, A.: Information heterogeneity and the speed of learning in social networks. Columbia Business School Research Paper (13-28) (2013)
Nedić, A., Olshevsky, A., Uribe, C.A.: Nonasymptotic convergence rates for cooperative learning over time-varying directed graphs. arXiv preprint arXiv:1410.1977 (2014)
Rad, K.R., Tahbaz-Salehi, A.: Distributed parameter estimation in networks. In: IEEE Conference on Decision and Control (CDC), pp. 5050–5055. IEEE (2010)
Shahrampour, S., Jadbabaie, A.: Exponentially fast parameter estimation in networks using distributed dual averaging. In: IEEE Conference on Decision and Control (CDC), pp. 6196–6201. IEEE (2013)
Lalitha, A., Sarwate, A., Javidi, T.: Social learning and distributed hypothesis testing. In 2014 IEEE International Symposium on Information Theory (ISIT), pp. 551–555, June 2014. Extended version at arXiv:1410.4307
Shahrampour, S., Rakhlin, A., Jadbabaie, A.: Finite-time analysis of the distributed detection problem. arXiv preprint arXiv:1512.09311 (2015)
Shahrampour, S., Rakhlin, A., Jadbabaie, A.: Distributed detection: finite-time analysis and impact of network topology. arXiv preprint arXiv:1409.8606 (2014)
Molavi, P., Tahbaz-Salehi, A., Jadbabaie, A.:. Foundations of non-Bayesian social learning. Columbia Business School Research Paper No. 15-95. SSRN: https://ssrn.com/abstract=2683607 or https://doi.org/10.2139/ssrn.2683607 (2017)
Olfati-Saber, R., Franco, E., Frazzoli, E., Shamma, J.: Belief consensus and distributed hypothesis testing in sensor networks. In: Networked Embedded Sensing and Control. Springer (2006)
Rahimian, M.A., Jadbabaie, M.A.: Learning without recall: a case for log-linear learning. IFAC-PapersOnLine 48(22), 46–51, ISSN 2405-8963 (2015). https://doi.org/10.1016/j.ifacol.2015.10.305
Su, L., Vaidya, N.H.: Asynchronous distributed hypothesis testing in the presence of crash failures University of Illinois at Urbana-Champaign, technical report. arXiv:1606.03418 (2016)
Shahin, S., Ali, J.: An online optimization approach for multi-agent tracking of dynamic parameters in the presence of adversarial noise. In: American Control Conference (ACC). IEEE (2017)
Bedi, A.S., Sarma, P., Rajawat, K.: Adversarial multi-agent target tracking with inexact online gradient descent. arXiv preprint arXiv:1710.05133 (2017)
Peaseć, M., Shostakć, R., Lamport, L.: Reaching agreement in the presence of faults. J. ACM 27(2), 228–234 (1980)
Dolev, D., Lynch, N.A., Pinter, S.S., Stark, E.W., Weihl, W.E.: Reaching approximate agreement in the presence of faults. J. ACM 33(3), 499–516 (1986)
Fekete, A.D.: Asymptotically optimal algorithms for approximate agreement. Distrib. Comput. 4(1), 9–29 (1990)
LeBlanc, H.J., Zhang, H., Sundaram, S., Koutsoukos, X.: Consensus of multi-agent networks in the presence of adversaries using only local information. In: Proceedings of the 1st International Conference on High Confidence Networked Systems, HiCoNS ’12, pp. 1–10, New York, NY, USA. ACM (2012)
Vaidya, N.H.: Iterative byzantine vector consensus in incomplete graphs. In: Distributed Computing and Networking, pp. 14–28. Springer (2014)
Vaidya, N.H., Tseng, L., Liang, G.: Iterative approximate byzantine consensus in arbitrary directed graphs. In: Proceedings of the 2012 ACM Symposium on Principles of Distributed Computing, pp. 365–374. ACM (2012)
Mendes, H., Herlihy, M.: Multidimensional approximate agreement in Byzantine asynchronous systems. In: Proceedings of the Forty-Fifth Annual ACM Symposium on Theory of Computing, STOC ’13, pp. 391–400, New York, NY, USA. ACM (2013)
Su, L., Vaidya, N.H.: Reaching approximate byzantine consensus with multi-hop communication. In: Proceedings of International Symposium on Stabilization, Safety, and Security of Distributed Systems (SSS) (2015)
Lynch, N.A.: Distributed Algorithms. Morgan Kaufmann Publishers Inc., San Francisco (1996)
Vaidya, N.H., Garg, V.K.: Byzantine vector consensus in complete graphs. In: Proceedings of the 2013 ACM Symposium on Principles of Distributed Computing, pp. 65–73. ACM (2013)
Perles, M.A., Sigron, M.: A generalization of Tverberg’s theorem. arXiv:0710.4668 (2007)
Hajnal, J., Bartlett, M.: Weak ergodicity in non-homogeneous markov chains. In: Mathematical Proceedings of the Cambridge Philosophical Society, vol. 54, pp. 233–246. Cambridge University Press, Cambridge (1958)
Wolfowitz, J.: Products of indecomposable, aperiodic, stochastic matrices. In: Proceedings of the American Mathematical Society, pp. 733–737. JSTOR (1963)
Vaidya, N.H.: Matrix representation of iterative approximate Byzantine consensus in directed graphs. arXiv:1203.1888 (2012)
Author information
Authors and Affiliations
Corresponding author
Additional information
This research is supported in part by National Science Foundation award NSF 1421918. Any opinions, findings, and conclusions or recommendations expressed here are those of the authors and do not necessarily reflect the views of the funding agencies or the U.S. government.
A short version of this manuscript [1] has been accepted to appear in the Proceedings of the International Symposium on Distributed Computing (DISC), Sep, 2016.
A proof of Eq. (21)
A proof of Eq. (21)
First it is easy to see that
Let \((\tilde{\theta _1}, \tilde{\theta _2})\) and \(\tilde{w_i}\) be the hypotheses ordered pair and the private signal such that
If \(\log \frac{\ell _i(\tilde{w_i}|\tilde{\theta _1})}{\ell _i(\tilde{w_i}|\tilde{\theta _2})}<0\), it holds that
If \(\log \frac{\ell _i(\tilde{w_i}|\tilde{\theta _1})}{\ell _i(\tilde{w_i}|\tilde{\theta _2})}\ge 0\), then
Rights and permissions
About this article
Cite this article
Su, L., Vaidya, N.H. Defending non-Bayesian learning against adversarial attacks. Distrib. Comput. 32, 277–289 (2019). https://doi.org/10.1007/s00446-018-0336-4
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00446-018-0336-4