Monte Carlo subspace method: An incremental approach to high-dimensional data classification

This paper presents the Monte Carlo subspace method - a cost-effective classification technique for high-dimensional data by the Monte Carlo scheme. The most intensive computation in the linear subspace methods is the reduction of dimensionality of the feature space by the eigen decomposition or singular value decomposition. In the present method, the subspaces are learned by updating their orthonormal basis sets with random increment of the dimension of the feature space. The subspace learning progresses with the similarity measurement of test samples until their classification is completed. The expected advantages include reduction of the computational expense without critical loss of recognition rate especially for the high-dimensional data. The performance of the present method was experimentally verified using face recognition datasets.