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

Neural network automatic pruning method based on evolutionary algorithm

A technology of neural network and evolutionary algorithm, applied in the field of neural network, can solve problems such as taking a long time, taking a lot of time, and difficult to achieve results, and achieve the effect of improving compression efficiency, small complexity, and fast convergence speed

Inactive Publication Date: 2020-08-07
XIAMEN UNIV
View PDF0 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Despite the remarkable progress, the traditional neural network compression method is difficult to achieve the best results, and it is highly dependent on the manual design of human experts, and it takes a lot of time to design the compressed network structure
For example, pruning based on the L1 norm needs to manually design the pruned target network according to the sensitivity. Designing such a network takes a lot of time, which grows linearly with the number of layers of the network

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
  • Neural network automatic pruning method based on evolutionary algorithm
  • Neural network automatic pruning method based on evolutionary algorithm
  • Neural network automatic pruning method based on evolutionary algorithm

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0023] Such as figure 1 As shown, the present invention discloses a kind of neural network automatic pruning method based on evolutionary algorithm, and it comprises the following steps:

[0024] Step 1. Given a calibrated image-label pair set, divide the image-label pair set into a training sample set, a test photo sample set and a verification sample set;

[0025] Given the neural network to be compressed and its overall compression rate;

[0026] Given the evolutionary algebra of the evolutionary algorithm, and the original population of the initial evolutionary algorithm, the population is a set, which contains multiple vectors, each vector represents an individual, and each vector represents the compression rate of each layer of the neural network;

[0027] Step 2. Search for the width scaling factor of the neural network by binary search, so that the size of the neural network model and the number of floating-point calculations during inference meet the overall compress...

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

The invention relates to a neural network model automatic pruning method based on an evolutionary algorithm, which provides an automatic pruning scheme based on the evolutionary algorithm, and can prune a neural network more quickly and accurately compared with schemes based on reinforcement learning and the like. The method is also suitable for a relatively large data set, and the speed and the precision are optimal.

Description

technical field [0001] The invention relates to the field of neural networks, in particular to an evolutionary algorithm-based automatic pruning method for neural networks. Background technique [0002] In recent years, with the development of artificial intelligence and deep learning, people have begun to experience exponential growth in customized deep learning network structures. Users more hope that deep learning can generate customized network structures and parameters for their own hardware, which leads to the generation of neural network automatic compression systems. Given a dataset and compression ratio, neural autocompression aims to find the best compression in a huge search space through a search algorithm. Network compression has been applied with success to various computer vision tasks, such as image classification, language modeling, and semantic segmentation. [0003] At present, neural network compression methods are mainly divided into three methods: mod...

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): G06K9/62G06N3/08
CPCG06N3/08G06F18/241
Inventor 纪荣嵘唐浪
Owner XIAMEN UNIV