Dynamic conditional random fields: factorized probabilistic models for labeling and segmenting sequence data

@inproceedings{citeulike:1014358, abstract = {In sequence modeling, we often wish to represent complex interaction between labels, such as when performing multiple, cascaded labeling tasks on the same sequence, or when long-range dependencies exist. We present dynamic conditional random fields (DCRFs), a generalization of linear-chain conditional random fields (CRFs) in which each time slice contains a set of state variables and edges---a distributed state representation as in dynamic Bayesian networks (DBNs)---and parameters are tied across slices. Since exact inference can be intractable in such models, we perform approximate inference using several schedules for belief propagation, including tree-based reparameterization (TRP). On a natural-language chunking task, we show that a DCRF performs better than a series of linear-chain CRFs, achieving comparable performance using only half the training data.}, author = {Sutton, Charles and Rohanimanesh, Khashayar and Mccallum, Andrew }, citeulike-article-id = {1014358}, keywords = {approximations, crf, graphical-models, machine-learning}, pages = {99+}, posted-at = {2007-11-07 17:05:49}, priority = {2}, title = {Dynamic conditional random fields: factorized probabilistic models for labeling and segmenting sequence data}, url = {http://citeseer.ist.psu.edu/687819.html}, year = {2004} }

TY - CONF ID - citeulike:1014358 L3 - citeulike-article-id:1014358 TI - Dynamic conditional random fields: factorized probabilistic models for labeling and segmenting sequence data SP - 99 N2 - In sequence modeling, we often wish to represent complex interaction between labels, such as when performing multiple, cascaded labeling tasks on the same sequence, or when long-range dependencies exist. We present dynamic conditional random fields (DCRFs), a generalization of linear-chain conditional random fields (CRFs) in which each time slice contains a set of state variables and edges---a distributed state representation as in dynamic Bayesian networks (DBNs)---and parameters are tied across slices. Since exact inference can be intractable in such models, we perform approximate inference using several schedules for belief propagation, including tree-based reparameterization (TRP). On a natural-language chunking task, we show that a DCRF performs better than a series of linear-chain CRFs, achieving comparable performance using only half the training data. KW - approximations KW - crf KW - graphical-models KW - machine-learning AU - Sutton, Charles AU - Rohanimanesh, Khashayar AU - Mccallum, Andrew PY - 2004/// UR - http://citeseer.ist.psu.edu/687819.html ER -