Compression method and device of recurrent neural network based on intrinsic orthogonal decomposition, processor and computer readable storage medium thereof

The invention relates to an intrinsic orthogonal decomposition-based recurrent neural network compression method, which comprises the following steps of entering a step A, and recording a hidden layer value of a compressed object during long and short-term memory network reasoning; the hidden layers of all the recorded data are averaged, and an average hidden layer matrix H is obtained; entering a step B, calculating a projection vector for the obtained average hidden layer matrix H, and obtaining a sub-matrix v; and performing dimension reduction processing on the original coefficient matrix by using the sub-matrix v to obtain a new model parameter quantity. The invention also relates to a corresponding device, a processor and a computer readable storage medium. The compression method and device of the recurrent neural network based on the intrinsic orthogonal decomposition, the processor and the computer readable storage medium can be applied to a compression task of a time sequence classification model, th