System and method for signal processing

Abstract

The embodiment of the present application discloses a signal processing method and device, including: an input module, used to receive an input signal matrix and a weight matrix; a first interleaving module, used to interleave an input signal matrix to obtain an interleaved signal matrix; a first block The module is used to block the interleaved signal matrix; the second interleaved module is used to interleave the weight matrix to obtain the interleaved weight matrix; the processing module is used to process the interleaved weight matrix to obtain a plurality of sparse block weight matrices; matrix multiplication The module is used to implement matrix multiplication corresponding to multiple sparse block weight matrices and multiple block signal matrices to obtain multiple matrix multiplication results; the output module is used to output signal processing results. The method and the related device solve the problem of unbalanced load caused by the uneven distribution of key weights, and improve the computing efficiency of the neural network.

Description

technical field [0001] This application relates to the field of artificial intelligence, in particular to systems and methods for signal processing. Background technique [0002] Artificial neural networks (ANN), also referred to as neural network (NN), is a network structure that imitates the behavioral characteristics of animal neural networks for information processing. The structure is composed of a large number of nodes (or called neurons) connected to each other, and based on a specific computing model, the purpose of processing information is achieved by learning and training input information. A neural network includes an input layer, a hidden layer and an output layer. The input layer is responsible for receiving input signals, the output layer is responsible for outputting the calculation results of the neural network, and the hidden layer is responsible for learning, training and other calculation processes. It is the memory unit of the network, and the memory of ...

AI Technical Summary

Problems solved by technology

However, the selection of the adaptive threshold is strongly related to the number of processor cores, and the sparseness scheme of different processor specifications should be adjusted accordingly, which increases the computational complexity; and the distribution of key weights cannot be uniform, by changing the threshold strong Exercising it evenly can cause neural network performance to degrade

Images