Activation processing device applied to neural networks

Abstract

The invention provides an activation processing device applied to a neural network. The activation processing device includes a searching table unit, a plurality of matching units and a plurality of calculation units; the searching table unit is used for storing a variable interval reflecting an activation function in a neural network and a mapping relationship between corresponding fitting linearfunction parameters, wherein the linear function parameters include the intercept and slope for defining linear functions; the plurality of matching units are configured to output a linear function parameter corresponding to an input variable of an activation function to be calculated on the basis of the searching table; the plurality of calculation units are one-to-one correspondingly connectedwith the plurality of matching units; and each of the plurality of calculation units is used for completing a linear operation for the input variable of the activation function to be calculated according to a linear function parameter output by the corresponding matching unit. The activation processing device of the invention can improve the activation processing efficiency of the neural network and reduce the power consumption.

Description

technical field [0001] The invention relates to the technical field of deep learning, in particular to an activation processing device applied to a neural network. 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] 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 in layers through multiple transformation stages. Breakthroughs in large-scale data processing tasks. The data processing of the neural network includes the ...

AI Technical Summary

Problems solved by technology

Activation processing usually includes a series of operations such as multiplication and addition operations. In the prior art, software or arithmetic logic units are usually used to implement activation processing. There are problems such as slow execution speed and large circuit hardware overhead, and it is difficult to achieve light weight. , Neural Network Processor with Low Power Consumption

Images