Presentation 2006/12/15
Cross-validation EM Algorithm for Robust Parameter Estimation
Takahiro SHINOZAKI, Mari OSTENDORF,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) A new maximum likelihood training algorithm is proposed that compensates for weaknesses of the EM algorithm by using cross-validation likelihood in the expectation step to avoid overtraining. By using a set of sufficient statistics associated with a partitioning of the training data, as in parallel EM, the algorithm has the same order of computational requirements as the original EM algorithm. Analyses using a GMM with artificial data show the proposed algorithm is more robust for overtraining than the conventional EM algorithm. Large vocabulary recognition experiments on Mandarin broadcast news data show that the method makes better use of more parameters and gives lower recognition error rates than EM training.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) EM training / overtraining / cross-validation
Paper # NLC2006-61,SP2006-117
Date of Issue

Conference Information
Committee NLC
Conference Date 2006/12/15(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Natural Language Understanding and Models of Communication (NLC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Cross-validation EM Algorithm for Robust Parameter Estimation
Sub Title (in English)
Keyword(1) EM training
Keyword(2) overtraining
Keyword(3) cross-validation
1st Author's Name Takahiro SHINOZAKI
1st Author's Affiliation Academic Center for Computing and Media Studies [South Building], Kyoto University()
2nd Author's Name Mari OSTENDORF
2nd Author's Affiliation Department of Electrical Engineering, University of Washington
Date 2006/12/15
Paper # NLC2006-61,SP2006-117
Volume (vol) vol.106
Number (no) 442
Page pp.pp.-
#Pages 6
Date of Issue