Neural network processor for Winograd convolution

Abstract

The present invention provides a neural network processor for Winograd convolution. The processor includes a neuron conversion module configured to perform the conversion operation V= [BTdB] of the neuron matrix; a weight conversion module configured to perform the conversion operation U= [GgGT] of the weight matrix; a point multiplication module configured to perform a point multiplication operation of the matrices U and V to obtain a point multiplication result matrix; a post-matrix conversion module configured to perform the conversion operation F=ATMA for the dot-multiplication result matrix,wherein d denotes a neuron matrix, g denotes a weight matrix, and G, B, and A denote the transformation matrixes corresponding to the weight matrix g, the neuron matrix d, and the dot multiplication result matrix M respectively. The neural network processor of the present invention can provide the computational efficiency and reduce the running power consumption.

Description

technical field [0001] The invention relates to the technical field of neural networks, in particular to a neural network processor oriented to Winograd convolution. Background technique [0002] In recent years, deep learning technology has developed rapidly and has been widely used in solving advanced abstract cognitive problems, such as image recognition, speech recognition, natural language understanding, weather prediction, gene expression, content recommendation and intelligent robots, and has become an academic Research hotspots in the world and industry. [0003] Deep neural network is one of the perception models with the highest level of development in the field of artificial intelligence. It simulates the neural connection structure of the human brain by establishing a model, and describes the data characteristics hierarchically through multiple transformation stages, providing images, videos, audios, etc. Large-scale data processing tasks bring breakthrough prog...

AI Technical Summary

Problems solved by technology

Therefore, researchers currently propose a Winograd-based convolution operation method, which can complete equivalent convolution operation tasks and greatly reduce the multiplication of the convolution operation process by performing specific matrix transformations on the input feature map and weights. Since the prediction process of most neural network processor chips in practical applications uses a fixed neural network model, the Winograd convolution output paradigm used is usually also a fixed mode. For Winograd convolution with a fixed output paradigm, its operation process It is very clear that there is a large room for optimization, how to design a Winograd convolutional neural network accelerator for a fixed output paradigm has become a research focus

Images