D09-1011 solution ( possible on trees ) or a variational approximation ( like our BP ) . Our work seeks
D09-1147 . Li et al. ( 2009 ) propose a variational approximation maximizing sentence probability
N10-1081 otherwise . For this reason , our variational approximation allows the use of recursive grammars
N10-1081 value . One of the advantages of variational approximation over sampling methods is the
D10-1120 is presented in Section 7.1 . Variational approximations to the HDP are truncated at 10
D13-1069 for summarization and use the variational approximation for inference . Experimental
N12-1014 exact objective , or a cheaper variational approximation to it , in a way that crucially
D10-1120 during inference by setting the HDP variational approximation truncation level to one . For
N06-5002 Integration -- Laplace Approximation -- Variational Approximation -- Others ( Message Passing Algorithms
N09-1027 improvements in parse quality only when a variational approximation is used to select the most likely
N09-1063 and Klein ( 2007 ) , who use a variational approximation to the most probable parse .
P06-2124 of document-pairs . Efficient variational approximation algorithms are designed for inference
N10-1061 factored representation . Our variational approximation takes the following form : Q
P06-1039 saddle-point ) approximation and the variational approximation . A third , less common , but
P06-1055 are given as the solution of a variational approximation of the original grammar . However
N10-1003 of Goodman ( 1996 ) , or in the variational approximation of Matsuzaki et al. ( 2005 )
N10-1061 Q_63 denotes all factors of the variational approximation except for the factor being updated
D11-1053 along the same lines by using a variational approximation ( see Agarwal and Chen ( 2009
D10-1004 - ferent , in reality both are variational approximations emanating from Prop. 1 , respectively
D10-1004 presenting a geometric view of the variational approximations underlying message-passing algorithms
hide detail