Neural network maxout layer calculation device

Abstract

The invention provides maxout layer calculation device, and aims at executing maxout layer calculation according to maxout layer operation instructions. The device is characterized by comprising a maxout layer calculation module, wherein the maxout layer calculation module is connected with a storage module and comprises a register unit, a loading / storage unit, a pruning unit and a calculation unit; the register unit is used for storing a maxout layer input data addresses which are storage addresses, in the storage module, of maxout layer input data; the loading / storage unit is used for obtaining corresponding maxout layer input data from the storage module according to a maxout layer input data address corresponding to a maxout layer calculation instruction; the pruning unit is used for compressing the maxout layer input data which is preferably stored according to features on the feature dimension so as to obtain pruned data; and the calculation unit is used for carrying out corresponding calculation on the pruned data according to the maxout layer calculation instruction so as to obtain a maxout layer calculation result.

Description

technical field [0001] The invention relates to a neural network maxout layer computing device and method, which are used to execute maxout layer computing according to maxout layer computing instructions. Background technique [0002] At present, more and more algorithms in the computer field involve maxout layer operations. Taking artificial neural network algorithms as an example, various neural network algorithms contain a large number of maxout layer operations. In the neural network, it is often necessary to prune the input data. For a maxout layer input data, the data is three-dimensional data, which are the data of the abscissa, ordinate and feature dimension respectively. The pruning strategy of the maxout operation is in the feature dimension Compressed data, ie f j =max({f i |a j ≤i≤b j}), where f i Each value on the feature dimension corresponding to a point, a j and b j Calibrate the range of data selected on the feature dimension, and finally get the val...

AI Technical Summary

Problems solved by technology

[0008] The purpose of the present invention is to provide a maxout layer computing device to solve the problems existing in the prior art such as limited inter-chip communication, insufficient on-chip cache, and insufficient flexibility in the length of the supported maxout layer.

Images