Bregman Conditional Random Fields: Sequence Labeling with Parallelizable Inference Algorithms
This addresses faster inference for sequence labeling tasks, though it appears incremental as an extension of existing CRF methods.
The paper tackles sequence labeling by proposing Bregman conditional random fields (BCRF), which enable fast parallelizable inference algorithms using iterative Bregman projections. The approach achieves comparable results to standard CRFs while being faster, and outperforms mean field in highly constrained settings.
We propose a novel discriminative model for sequence labeling called Bregman conditional random fields (BCRF). Contrary to standard linear-chain conditional random fields, BCRF allows fast parallelizable inference algorithms based on iterative Bregman projections. We show how such models can be learned using Fenchel-Young losses, including extension for learning from partial labels. Experimentally, our approach delivers comparable results to CRF while being faster, and achieves better results in highly constrained settings compared to mean field, another parallelizable alternative.