Unlock instant, AI-driven research and patent intelligence for your innovation.

Network quantization method, inference method, and network quantization device

A quantitative method and network technology, applied in neural learning methods, biological neural network models, instruments, etc., can solve problems such as poor reasoning accuracy, poor machine learning speed, and impact

Pending Publication Date: 2021-04-30
SOCIONEXT INC
View PDF1 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

The problem that arises in this case is that the quantization error increases, which adversely affects the speed of machine learning, and also adversely affects the accuracy of inference after learning.

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Network quantization method, inference method, and network quantization device
  • Network quantization method, inference method, and network quantization device
  • Network quantization method, inference method, and network quantization device

Examples

Experimental program
Comparison scheme
Effect test

Embodiment approach 1

[0037] The network quantization method and network quantization device according to Embodiment 1 will be described.

[0038] [1-1. Network quantization device]

[0039] First, use figure 1 The configuration of the network quantization device according to this embodiment will be described. figure 1 It is a block diagram showing an outline of the functional configuration of the network quantization device 10 according to this embodiment.

[0040] The network quantization device 10 is a device that quantizes the neural network 14 . That is, the network quantization device 10 is a device that converts the neural network 14 with floating point precision into a neural network with fixed point precision, that is, into a quantized network. In addition, the network quantization device 10 may not quantize all the tensors used by the neural network 14, but only needs to quantize at least a part of the tensors. Here, the tensor refers to a value represented by an n-dimensional array (...

Embodiment approach 2

[0105] The network quantization method and the like according to Embodiment 2 will be described. The difference between the network quantization method involved in this embodiment and the quantization method involved in Embodiment 1 is that, according to the statistical information of the test data set, the test data set is classified into multiple types, and different types are performed according to each type. deal with. Hereinafter, an inference method using a quantized network generated by the network quantization method, network quantization device, and network quantization method according to this embodiment will be described focusing on differences from Embodiment 1.

[0106] [2-1. Network quantization device]

[0107] First, use Figure 9 The configuration of the network quantization device according to this embodiment will be described. Figure 9 It is a block diagram showing an outline of the functional configuration of the network quantization device 110 accordin...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

This network quantization method for quantizing a neural network (14) includes: a database construction step (S20) for constructing a statistical information database (18) of tensors that are dealt with by the neural network (14) and obtained when a plurality of test data sets (12) are input to the neural network (14); a parameter generation step (S30) for generating a quantization parameter set by quantizing tensor values; and a network construction step (S40) for quantizing the neural network (14) by using the quantization parameter set (22), wherein, on the basis of the statistical information database (18), the parameter generation step (S30) sets, among the tensor values, a quantization step interval in a high frequency area including tensor values of the maximum frequency to be narrower than that in a low frequency area including tensor values having non-zero frequency and less frequency than the high frequency area.

Description

technical field [0001] The present disclosure relates to a network quantization method, a reasoning method, and a network quantization device. Background technique [0002] Machine learning has traditionally been performed using networks such as neural networks. Here, a model that uses numerical data as an input and performs some operations to obtain an output value of the numerical data is called a network. When installing a network on hardware such as a computer, in order to reduce hardware costs, it is desirable to construct a network with lower calculation accuracy while keeping the inference accuracy after installation at the same level as the floating-point accuracy. [0003] For example, when installing a network that performs all calculations with floating point precision, it is required to perform calculations with fixed point precision while maintaining inference precision due to increased hardware costs. network of. [0004] Hereinafter, a network with floating...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06N3/02
CPCG06N3/082G06N3/048G06N3/045G06F18/211G06F18/2414G06N3/08G06F18/2415G06F18/2431
Inventor 笹川幸宏
Owner SOCIONEXT INC