判别正则化谱回归

谱回归(SR)算法是一种正则化的降维方法,通过学习获得回归框架下的嵌入函数,使其避免了稠密矩阵分解的问题.但是在谱回归的构图中,更加关注于类内信息,而忽视了很重要的类间信息.为此,提出一种新的降维算法——判别正则化谱回归(DRSR).它将数据集的判别信息和流行结构同时嵌入到正则项的构造中,期望使输出结果即保持同类样本间的内在邻近关系,同时又能将不同类的近邻样本尽可能分得开.最后,分析了这种算法的优缺点,并在两个常用的数据集(Yale和wine)上验证了算法的可行性及有效性.

Discriminatively Regularized Spectral Regression

Spectral Regression is a regularized method for dimensionality reduction.It casts the problem of learning an embedding function into a regression framework,which avoids eigen-decomposition of dense matrices.However,the intra-class information attract more attentions in constructive graph of SR in stead of the critical inter-class information.To address this issue,a novel algorithms for dimensionality reduction are presented,called Discriminatively Regularized Spectral Regression(DRSR) method.DRSR embeds the discriminative information as well as the manifold structures into the regularization term,which aims to retain the intraclass compactness and connects each data point with its neighboring points of the same class,while characterizes the interclass separability and connects the marginal points.The feasibility and effectiveness of the proposed method is then verified on two popular databases(Yale and wine) with promising results.