SUBSPACE ANALYSIS OF DEEP LEARNING ALGORITHMS USING TENSOR FLOW

Abstract: Low dimensional linear spaces can viably demonstrate the image varieties of numerous objects (human faces specifically) under factor lighting. The standard linear subspace learning algorithms incorporate Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and Locality Preserving Projection (LPP). These techniques consider an n1 × n2 picture as a high dimensional vector in Rn1×n2, while a picture spoke to in the plane is inherently a matrix. In this paper, we propose another algorithm called Tensor Subspace Analysis (TSA). TSA thinks about a picture as the second request tensor in Rn1 Rn2, where Rn1 and Rn2 are two vector spaces. TSA can generally describe the connection between the segment vectors of the picture matrix and the column vectors. TSA identifies the genetic neighborhood mathematical structure of the tensor space by learning a lower-dimensional tensor subspace. We contrast our proposed approach and PCA, LDA, and LPP techniques on two standard databases. Experimental results show that TSA accomplishes a better recognition rate while being considerably more effective