Floating-point number multiplication method and computer readable storage medium

Abstract

The invention discloses a floating-point number multiplication method and a computer readable storage medium, the method comprises: establishing a first quantization lookup table and a second quantization lookup table, the first quantization lookup table being used for recording an association relationship between a floating-point number and a corresponding integer set, and the second quantizationlookup table is used for recording an association relationship between the integer and the corresponding floating-point number; obtaining two floating-point numbers to be multiplied; respectively obtaining integer sets corresponding to the two floating-point numbers according to the first quantization lookup table to obtain a first integer set and a second integer set; respectively adding each integer in the first integer set and each integer in the second integer set to obtain a third integer set; obtaining floating-point numbers corresponding to integers in the third integer set according to the second quantization lookup table; and adding the floating-point numbers corresponding to the integers to obtain a multiplication result of the two floating-point numbers to be multiplied. According to the method, the convolution operation is accelerated, and the precision loss caused by quantization is greatly reduced.

Description

technical field [0001] The invention relates to the technical field of convolutional neural networks, in particular to a method for multiplying floating-point numbers and a computer-readable storage medium. Background technique [0002] Convolutional neural network is currently the mainstream technology in the field of computer vision, and can be widely used in target detection, recognition, super-resolution, 3D reconstruction and other fields. Since the convolutional neural network model usually has the characteristics of a large number of parameters and a large amount of floating-point operations, most of the current methods require a high-performance computing platform to realize the parallel operation of convolution. To this end, researchers have gradually begun to focus on the acceleration of convolution operations, as well as quantization operations. For example, convolutional networks based on binary quantization, or sparsely represented networks, etc. However, the ...

AI Technical Summary

Problems solved by technology

However, the accuracy loss of the binary quantization method is large, and the sparse representation network is difficult to process in parallel, and it is difficult to deploy to mobile terminal applications such as mobile phones.

With the popularization of images and videos in mobile devices, there is an urgent need for high-performance target detection and recognition methods that can be deployed on cheap computing platforms to meet the needs of different application scenarios, but currently there is a lack of effective methods for mobile devices. Convolution Acceleration Scheme

Images