Modular tensor sparsity preserving projection algorithm for dimension reduction

This paper proposes a modular tensor sparsity preserving projection (MTSPP) algorithm. This algorithm uniformly partitions the high-dimensional matrix data and builds third order tensor data, determines the weight of sparse reconstruction of all samples and applies it to the sparsity preserving projection of the third order tensor. Experiments finally indicate that MTSPP improves the robust performance of the global sparse representation-based dimension reduction algorithm by weighted sparse representation and spatial relationship of characteristics within the module and between modules.