2009 Fiscal Year Final Research Report
Asymptotic properties of information criteria for selecting hyper parameter
| Project/Area Number |
19700265
|
|---|
| Research Category |
Grant-in-Aid for Young Scientists (B)
|
| Allocation Type | Single-year Grants |
|---|
| Research Field |
Statistical science
|
|---|
| Research Institution | Hiroshima University |
|---|
Principal Investigator |
YANAGIHARA Hirokazu Hiroshima University, 大学院・理学研究科, 准教授 (70342615)
|
|---|
| Project Period (FY) |
2007 – 2009
|
| Keywords | モデル選択 / 情報量規準 / 罰則付き尤度 / バイアス補正 |
|---|
| Research Abstract |
We proposed two bias-corrected CV criteria, one is reducing the bias of CV to O(n^<-2k>) by the weighed sum of leave-k-out CV criteria, and the other is reducing the bias of CV to O(n^<-3>) by applying the jackknife method to an evaluation of the bias. Moreover, we proposed an unbiased C_p criterion for multivariate ridge regression under the normal assumption.
|
