A game-theoretic approach to adversarial linear Gaussian classification

For conditionally Gaussian data points and linear support vector machine classifiers, the optimization problems of the adversary and the classifier are rewritten as convex problems and best response dynamics are used to learn an equilibrium of the game.