Fully Distributed EM for Very Large Datasets

@inproceedings{citeulike:2919014, abstract = {In EM and related algorithms, E-step compu- tations distribute easily, because data items are independent given parameters. For very large data sets, however, even storing all of the parameters in a single node for the M- step can be impractical. We present a frame- work that fully distributes the entire EM pro- cedure. Each node interacts only with pa- rameters relevant to its data, sending mes- sages to other nodes along a junction-tree topology. We demonstrate improvements over a MapReduce topology, on two tasks: word alignment and topic modeling.}, author = {Wolfe, Jason and Haghighi, Aria and Klein, Dan }, booktitle = {Proceedings of ICML}, citeulike-article-id = {2919014}, keywords = {em, machine-learning}, posted-at = {2008-07-18 15:36:16}, priority = {2}, title = {Fully Distributed EM for Very Large Datasets}, url = {http://www.cs.berkeley.edu/\~{}aria42/pubs/icml08-distributedem}, year = {2008} }

TY - CONF ID - citeulike:2919014 L3 - citeulike-article-id:2919014 TI - Fully Distributed EM for Very Large Datasets T2 - Proceedings of ICML N2 - In EM and related algorithms, E-step compu- tations distribute easily, because data items are independent given parameters. For very large data sets, however, even storing all of the parameters in a single node for the M- step can be impractical. We present a frame- work that fully distributes the entire EM pro- cedure. Each node interacts only with pa- rameters relevant to its data, sending mes- sages to other nodes along a junction-tree topology. We demonstrate improvements over a MapReduce topology, on two tasks: word alignment and topic modeling. KW - em KW - machine-learning AU - Wolfe, Jason AU - Haghighi, Aria AU - Klein, Dan PY - 2008/// UR - http://www.cs.berkeley.edu/~aria42/pubs/icml08-distributedem ER -