J05-4005 global minimum . We therefore use a stochastic approximation to gradient descent . Whereas
W12-3021 sample . Since we are using a stochastic approximation to the model score , in general
W12-3021 for Computational Linguistics a stochastic approximation . We propose two strategies for
N09-1069 Stepwise EM is motivated from the stochastic approximation literature , where we think of
P15-1076 training algorithm using joint stochastic approximation and trans-dimensional mixture
E06-1028 simple " language model " , a stochastic approximation to the language of the collection
P15-1076 similarly as in related works on stochastic approximation ( Gu and Zhu , 2001 ) and stochastic
D12-1101 samples . Since we are using a stochastic approximation to the model score , in general
W13-3519 that make either deterministic or stochastic approximations ( Kurihara and Sato , 2006 ;
P15-1076 random field models . 3.2 Joint stochastic approximation Training random field models
J05-4005 shown in Equation ( 7 ) , the stochastic approximation method updates parameters incrementally
D13-1160 0t +1 is set based on taking a stochastic approximation of aO ( B ; Bt ) B = Bt . aB
P07-1096 updates at lines 13 and 14 are a stochastic approximation of gradient descent that minimizes
J05-4005 − fd ( wi ) ) ( 8 ) The stochastic approximation method can be viewed as optimizing
N09-1069 ηk . Standard results from the stochastic approximation literature state that E ∞
P15-1076 Model Estimation We develop a stochastic approximation algorithm using Markov chain
hide detail