Bayesian Games for Adversarial Regression Problems

Michael Großhans, Christoph Sawade, Michael Brückner, Tobias Scheffer
Proceedings of the 30th International Conference on Machine Learning, PMLR 28(3):55-63, 2013.

Abstract

We study regression problems in which an adversary can exercise some control over the data generation process. Learner and adversary have conflicting but not necessarily perfectly antagonistic objectives. We study the case in which the learner is not fully informed about the adversary’s objective; instead, any knowledge of the learner about parameters of the adversary’s goal may be reflected in a Bayesian prior. We model this problem as a Bayesian game, and characterize conditions under which a unique Bayesian equilibrium point exists. We experimentally compare the Bayesian equilibrium strategy to the Nash equilibrium strategy, the minimax strategy, and regular linear regression.

Cite this Paper

BibTeX
@InProceedings{pmlr-v28-grosshans13, title = {Bayesian Games for Adversarial Regression Problems}, author = {Großhans, Michael and Sawade, Christoph and Brückner, Michael and Scheffer, Tobias}, booktitle = {Proceedings of the 30th International Conference on Machine Learning}, pages = {55--63}, year = {2013}, editor = {Dasgupta, Sanjoy and McAllester, David}, volume = {28}, number = {3}, series = {Proceedings of Machine Learning Research}, address = {Atlanta, Georgia, USA}, month = {17--19 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v28/grosshans13.pdf}, url = {https://proceedings.mlr.press/v28/grosshans13.html}, abstract = {We study regression problems in which an adversary can exercise some control over the data generation process. Learner and adversary have conflicting but not necessarily perfectly antagonistic objectives. We study the case in which the learner is not fully informed about the adversary’s objective; instead, any knowledge of the learner about parameters of the adversary’s goal may be reflected in a Bayesian prior. We model this problem as a Bayesian game, and characterize conditions under which a unique Bayesian equilibrium point exists. We experimentally compare the Bayesian equilibrium strategy to the Nash equilibrium strategy, the minimax strategy, and regular linear regression.} }
Endnote
%0 Conference Paper %T Bayesian Games for Adversarial Regression Problems %A Michael Großhans %A Christoph Sawade %A Michael Brückner %A Tobias Scheffer %B Proceedings of the 30th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2013 %E Sanjoy Dasgupta %E David McAllester %F pmlr-v28-grosshans13 %I PMLR %P 55--63 %U https://proceedings.mlr.press/v28/grosshans13.html %V 28 %N 3 %X We study regression problems in which an adversary can exercise some control over the data generation process. Learner and adversary have conflicting but not necessarily perfectly antagonistic objectives. We study the case in which the learner is not fully informed about the adversary’s objective; instead, any knowledge of the learner about parameters of the adversary’s goal may be reflected in a Bayesian prior. We model this problem as a Bayesian game, and characterize conditions under which a unique Bayesian equilibrium point exists. We experimentally compare the Bayesian equilibrium strategy to the Nash equilibrium strategy, the minimax strategy, and regular linear regression.
RIS
TY - CPAPER TI - Bayesian Games for Adversarial Regression Problems AU - Michael Großhans AU - Christoph Sawade AU - Michael Brückner AU - Tobias Scheffer BT - Proceedings of the 30th International Conference on Machine Learning DA - 2013/05/26 ED - Sanjoy Dasgupta ED - David McAllester ID - pmlr-v28-grosshans13 PB - PMLR DP - Proceedings of Machine Learning Research VL - 28 IS - 3 SP - 55 EP - 63 L1 - http://proceedings.mlr.press/v28/grosshans13.pdf UR - https://proceedings.mlr.press/v28/grosshans13.html AB - We study regression problems in which an adversary can exercise some control over the data generation process. Learner and adversary have conflicting but not necessarily perfectly antagonistic objectives. We study the case in which the learner is not fully informed about the adversary’s objective; instead, any knowledge of the learner about parameters of the adversary’s goal may be reflected in a Bayesian prior. We model this problem as a Bayesian game, and characterize conditions under which a unique Bayesian equilibrium point exists. We experimentally compare the Bayesian equilibrium strategy to the Nash equilibrium strategy, the minimax strategy, and regular linear regression. ER -
APA
Großhans, M., Sawade, C., Brückner, M. & Scheffer, T.. (2013). Bayesian Games for Adversarial Regression Problems. Proceedings of the 30th International Conference on Machine Learning, in Proceedings of Machine Learning Research 28(3):55-63 Available from https://proceedings.mlr.press/v28/grosshans13.html.

Related Material