Adversarial prediction: Approximating our training data and optimizing over the exact performance measure to provide greater flexibility for:
Learning under covariate shift (input distribution bias) and active learning;
Cost-sensitive classification and inductive optimization of univariate performance measures;
Learning to optimize for F-measure, discounted cumulative gain, and other multivariate performance measures; and
Structured prediction problems over sequences, trees, graphs, etc.
Inverse optimal control: Using maximum entropy structured prediction techniques to forecast future human behavior for intelligent robotics and vehicle navigation applications.