Repeated substrings contribute local costs to a sequence model. The highlighted occurrences illustrate the pattern structure used by algorithms for partition functions, marginals, MAP inference, and sampling; no fitted model is shown. Research source
About this paper
Efficient inference for sequence models with pattern-based potentials: partition functions, marginals, MAP inference, and sampling.
The model assigns energy to occurrences of predefined patterns along a sequence. The algorithms improve earlier complexity bounds and make these conditional random fields more practical for sequence tagging.
Citation
BibTeX
@inproceedings{takhanov2013patternbasedcrfs,
title = {Inference algorithms for pattern-based CRFs on sequence data},
author = {Takhanov, Rustem and Kolmogorov, Vladimir},
year = {2013},
booktitle = {ICML 2013},
eprint = {1210.0508},
archivePrefix = {arXiv},
url = {https://arxiv.org/pdf/1210.0508}
}