Training Conditional Random Fields for Maximum Labelwise Accuracy
Abstract
We consider the problem of training a conditional random field (CRF) to maximize per-label predictive accuracy on a training set, an approach motivated by the principle of empirical risk minimization. We give a gradient-based procedure for minimizing an arbitrarily accurate approximation of the empirical risk under a Hamming loss function. In experiments with both simulated and real data, our optimization procedure gives significantly better testing performance than several current approaches for CRF training, especially in situations of high label noise.
Cite
Text
Gross et al. "Training Conditional Random Fields for Maximum Labelwise Accuracy." Neural Information Processing Systems, 2006.Markdown
[Gross et al. "Training Conditional Random Fields for Maximum Labelwise Accuracy." Neural Information Processing Systems, 2006.](https://mlanthology.org/neurips/2006/gross2006neurips-training/)BibTeX
@inproceedings{gross2006neurips-training,
title = {{Training Conditional Random Fields for Maximum Labelwise Accuracy}},
author = {Gross, Samuel S. and Russakovsky, Olga and Do, Chuong B. and Batzoglou, Serafim},
booktitle = {Neural Information Processing Systems},
year = {2006},
pages = {529-536},
url = {https://mlanthology.org/neurips/2006/gross2006neurips-training/}
}