Learning Structured Models with the AUC Loss and Its Generalizations
Nir Rosenfeld, Ofer Meshi, Daniel Tarlow, Amir Globerson · 2014
Many problems involve the prediction of mul-tiple, possibly dependent labels. The struc-tured output prediction framework builds predictors that take these dependencies into account and use them to improve accuracy. In many such tasks, performance is evalu-ated by the Area Under the ROC Curve (AUC). While a framework for optimizing the AUC loss for unstructured models exists, it does not naturally extend to structured mod-els. In this work, we propose a representa-tion and learning formulation for optimizing structured models over the AUC loss, show how our approach generalizes the unstruc-tured case, and provide algorithms for solv-ing the resulting inference and learning prob-lems. We also explore several new variants of the AUC measure which naturally arise from our formulation. Finally, we empirically show the utility of our approach in several do-mains. 1