| Presentation | 2002/3/8 Experimental study for replacing statistics of posterior with propagated belief in the EM algorithm on lattice model Akihiro MINAGAWA, Norio TAGAWA, Toshiyuki TANAKA, |
|---|---|
| PDF Download Page | PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | In this paper, we show results replacing the covariance of the posterior probability with the variance of the belief in the expectation maximization (EM) algorithm. We consider graphical models with hidden variables as Gaussian Markov random fields (GMRF) model, which is popular as a regularization term for solving ill-posed problem and/or for obtaining a smoothed solution of observed variables in the area of image recognition and pattern recognition. To solve this problem regarded as the maximum a posteriori (MAP) estimation, calculating the inverse covariance matrix with (number of pixels) × (number of pixels) is required, and it occurs enormous computational cost. Moreover, EM algorithm is commonly used for obtaining model parameters in the above area. For reducing the computational cost, we consider applying the belief obtained from BP scheme as an approximation of the posterior probability in the EM algorithm. We show some experimental results for estimator of model parameters in two different graphical models with GMRF. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | belief propagation / GMRF / MAP estimator / EM algorithm / MFA (mean field approximation) |
| Paper # | PRMU2001-287 |
| Date of Issue |
| Conference Information | |
| Committee | PRMU |
|---|---|
| Conference Date | 2002/3/8(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 | Pattern Recognition and Media Understanding (PRMU) |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Experimental study for replacing statistics of posterior with propagated belief in the EM algorithm on lattice model |
| Sub Title (in English) | |
| Keyword(1) | belief propagation |
| Keyword(2) | GMRF |
| Keyword(3) | MAP estimator |
| Keyword(4) | EM algorithm |
| Keyword(5) | MFA (mean field approximation) |
| 1st Author's Name | Akihiro MINAGAWA |
| 1st Author's Affiliation | Graduate School of Engineering, Tokyo Metropolitan University() |
| 2nd Author's Name | Norio TAGAWA |
| 2nd Author's Affiliation | Graduate School of Engineering, Tokyo Metropolitan University |
| 3rd Author's Name | Toshiyuki TANAKA |
| 3rd Author's Affiliation | Graduate School of Engineering, Tokyo Metropolitan University |
| Date | 2002/3/8 |
| Paper # | PRMU2001-287 |
| Volume (vol) | vol.101 |
| Number (no) | 713 |
| Page | pp.pp.- |
| #Pages | 8 |
| Date of Issue | |