Deep neural network compression method based on reinforcement learning

A technology of deep neural network and reinforcement learning, which is applied in the direction of neural learning methods, biological neural network models, neural architectures, etc., can solve the problems of time-consuming models, inability to directly run volume deep neural network models, and inflexibility, etc., to achieve improved compression Scale, wide application, reduction in storage space and computation requirements

Active Publication Date: 2020-01-24
SICHUAN PANOVASIC TECH
View PDF3 Cites 5 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] The purpose of the present invention is to provide a deep neural network compression method based on reinforcement learning, which is used to solve the problem that the deep neural network model with huge volume

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
  • Deep neural network compression method based on reinforcement learning

Examples

Experimental program
Comparison scheme
Effect test

Example Embodiment

[0036] Example 1:

[0037] A deep neural network compression method based on reinforcement learning, including:

[0038] Step S100: Training a reinforcement learning agent based on the trained convolutional neural network, the reinforcement learning agent is used to identify network sparsity, and a compression method is used to compress the model;

[0039] Step S200: Train the compressed model again, optimize the parameters of the model, and obtain the final compressed model.

[0040] This method involves an input module, used to read the trained convolutional neural network model, including network structure, variable value, variable index and other information, as the data source of the next module;

[0041] The reinforcement learning module, based on the trained convolutional neural network model, trains a reinforcement learning agent that can recognize the sparsity of the network;

[0042] The compression module uses the reinforcement learning agent to judge the sparsity of each laye...

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 discloses a deep neural network compression method based on reinforcement learning, and the method comprises the steps: S100, training a reinforcement learning agent based on a trained convolutional neural network, the reinforcement learning agent being used for recognizing the sparsity of a network, and employing a compression method to compress a model; and S200, training the compressed model again, optimizing parameters of the model, and obtaining a final compression model. According to the invention, a huge and complex deep neural network model is compressed. The compressed model can run on a platform with limited hardware resources, such as smart home equipment and the like. Compared with an original model, the compressed model has the advantages that the requirements for the storage space and the operand are greatly reduced. Meanwhile, the performance of the model is still kept at the original level, wide application of the image recognition technology in daily lifecan be achieved, and convenience and safety of life are improved.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to a deep neural network compression method based on reinforcement learning. Background technique [0002] In the field of image recognition and face recognition, using deep learning technology and convolutional neural network can achieve very good recognition results, and its recognition accuracy is much higher than traditional image recognition methods. During the training process, the convolutional neural network receives a large number of training image samples, uses the convolutional layer to extract the features in the image layer by layer, and adjusts the parameters in the network through the direction propagation algorithm, so as to minimize the output error. Due to its extremely high accuracy, deep convolutional neural networks have gradually become the mainstream method in the field of image recognition. However, image recognition methods based on deep learning ...

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
IPC IPC(8): G06N3/04G06N3/08G06K9/62
CPCG06N3/082G06N3/045G06F18/241
Inventor 胡章一彭凝多唐博
Owner SICHUAN PANOVASIC TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap