| Project/Area Number |
14540104
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
General mathematics (including Probability theory/Statistical mathematics)
|
|---|
| Research Institution | TOKYO INSTITUTE OF TECHNOLOGY |
|---|
Principal Investigator |
MASE Shigeru Tokyo Institute of Technology, Department of Math.and Comp.Sci., Professor, 大学院・情報理工学研究科, 教授 (70108190)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
KANAMORI Takafumi Tokyo Institute of Technology, Department of Math.and Comp.Sci., Research Associate, 大学院・情報理工学研究科, 助手 (60334546)
|
|---|
| Project Period (FY) |
2002 – 2004
|
| Project Status |
Completed (Fiscal Year 2004)
|
| Budget Amount *help |
¥2,000,000 (Direct Cost: ¥2,000,000)
Fiscal Year 2004: ¥700,000 (Direct Cost: ¥700,000)
Fiscal Year 2003: ¥600,000 (Direct Cost: ¥600,000)
Fiscal Year 2002: ¥700,000 (Direct Cost: ¥700,000)
|
|---|
| Keywords | Bayesian network / LBP algorithm / linkage analysis / MAPP estimator / credit-rating / Cayley tree / phase transition / learning theory / Belief Propagation / Unwrappedネットワーク / マルコフ確率場 / 相転移 / バウンダリ確率則 / Loopy Belief Propagation / マーカー遺伝子 / ゲノム解析 / 遺伝連鎖解析 / アルゴリズム / 家系図 / 多座位遺伝子データ |
|---|
| Research Abstract |
Stochastic networks with and without are common frameworks in many genetical problems. In particular, They are basis of linkage analysis of family genetic data. The LBP (Loopy Belief Propagation) algorithm is an efficient algorithm for estimating marginal probabilities of nodes of stochastic networks with loops. In order to apply this algorithm to linkage analysis, we studied the following basic theoretical problem :
(1)Taking Cayley trees as examples, convergence and marginal probability recovery problems were studied. Using theoretical and numerical results, we show that the convergence is closely related with the existence of phase transitions. Ising models on Cayley trees have two kind of phase transitions. LBP converges on one phase transition region, but does not converge on another phase transition region. If converged, beliefs may not coincide with true marginal probabilities. Nevertheless, it is observed that states that both give highest values are coincide.
(2)As an application of stochastic networks, we consider an application of credit-rating of companies. It is shown that a naive Bayesian networks can give better predictions than common subjective networks employed by analysts.
(3)With an applications to bioinformatics in mind, Kanamori studied some properties of learning theory. In particular, boosting methods.
|
