APPARATUS AND METHOD FOR DEEP NEURAL NETWORK MODEL PARAMETER REDUCTION USING SPARSITY REGULARIZED RACTORIZED MATRIX

摘要

The present invention relates to an apparatus and method for reducing the number of deep neural network model parameters. The deep neural network model parameter reduction apparatus based on the sparse limited decomposition matrix according to the present invention includes a processor storing a memory and a program in which a deep neural network model parameter reduction program is stored, and the processor expresses a hidden layer using a decomposed matrix of full rank. , Training using the method of applying the sparsity constraint to make the diagonal matrix value to 0, and determining the rank of each hidden layer of the deep neural network model according to the degree of the sparsity constraint.