- [Home]
- [Research achievement]
- [Research achievement detail]
| Title | A Note on Belief Propagation Algorithm in Graphical Models with a Single Loop (in Japanese) |
|---|---|
| Authors | 岡野洋平 、小泉大城 、松嶋敏泰 |
| Released Year | 2005 |
| Format | Conference |
| Category | Channel coding |
| Jounal Name | Proceedings of the 28th Symposium on Information Theory and Its Applications |
| Jounal Page | vol.1, pp.1-4 |
| Published Year | 2005 |
| Published Month | 11 |
| Abstract （English） |
In the field of probabilistic reasoning, the peformanceof Belief Propagation Algorithm(BP) in graphicalmodels with a single loop is analyzed to some degree. Theapproximate posterior probability as the output of BP canbe calculated by the largest eigenvalue of a matrix which isproduct of all transition matrices, whereas the exact posteriorprobability can be calculated by the all eigenvaluesof the matrix. By using this fact, we examine the factorsrelated to the second eigenvalue against the largest eigenvalue,and analyze the performance of BP in Tailbiting(TB)codes whose graphical models have a single loop. |
| Note （English） |
1 |
| Manuscript | |
| Presentation |
