The paper develops some interesting game-theoretic results in the "No Box" setting, in which neither the attacker nor the defender have access to other party's actions. It proves that some standard attack and defense strategies constitute a Nash equilibrium in this setting. However, on the practical part, the experimental evaluation does not always demonstrate the utility of the proposed method, especially in comparison to the baselines. It would also be helpful if the authors could discuss the relationship of their method to prior game-theoretic approaches to adversarial learning in the general machine learning setting, e.g. Brückner, M., Kanzow, C. and Scheffer, T., 2012. Static prediction games for adversarial learning problems. The Journal of Machine Learning Research, 13(1), pp.2617-2654. Brückner, M. and Scheffer, T., 2011, August. Stackelberg games for adversarial prediction problems. In Proceedings of the 17th ACM SIGKDD international conference on Knowledge discovery and data mining (pp. 547-555).