Entropy and margin maximization for structured output learning

Abstract. We consider the problem of training discriminative struc-tured output predictors, such as conditional random fields (CRFs) and structured support vector machines (SSVMs). A generalized loss func-tion is introduced, which jointly maximizes the entropy and the margin of the solution. The CRF and SSVM emerge as special cases of our frame-work. The probabilistic interpretation of large margin methods reveals insights about margin and slack rescaling. Furthermore, we derive the corresponding extensions for latent variable models, in which training operates on partially observed outputs. Experimental results for mul-ticlass, linear-chain models and multiple instance learning demonstrate that the generalized loss can improve accuracy of the resulting classifiers.