Learning methods for kernel subspace methods and their applications

通过约束逼近问题，将子空间方法扩展为特征提取方法。子空间方法是约束近似框架的一种实现。子空间方法具有秩约束，但通过将秩约束替换为其他约束，可以实现具有不同性质的特征提取器。我们还提出了子集核主成分分析，以避免核化过程中的大量计算。

We have extended the subspace methods to feature extraction methods by constrained approximation problems. The subspace methods are sorts of realizations of the constrained approximation framework. The subspace methods have the rank constraint, however, by replacing the constraint to the other constraints, feature extractors that have various properties can be realized. We also have proposed the subset kernel principal component analysis to avoid large calculations in kernelization.