Learning and Inference over Constrained Output

V. Punyakanok, D. Roth, W. Yih, and D. Zimak

Abstract

We study learning structured output in a discriminative framework where values of the output variables are estimated by local classifiers. In this framework, complex dependencies among the output variables are captured by constraints and dictate which global labels can be inferred. We compare two strategies, learning independent classifiers and inference based training, by observing their behaviors in different conditions. Experiments and theoretical justification lead to the conclusion that using inference based learning is superior when the local classifiers are difficult to learn but may require many examples before any discernible difference can be observed.

Details

Publication typeInproceedings
Published inProceedings of IJCAI 2005
Pages1124-1129
> Publications > Learning and Inference over Constrained Output