In the context of classification, it is a common phenomenon that high-dimensional data such as micro-array data consist of only a few informative components. If one uses standard statistical modeling and estimation procedures with entire information, it tends to overfit the data due to noise...