Training Structural SVMs when Exact Inference is Intractable

While discriminative training (e.g., CRF, structural SVM) holds much promise for machine translation, image segmentation, and clustering, the complex inference these applications require make exact tr