Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Deep learning model optimization method based on network addition/modification

An optimization method and network technology, applied in the field of optimization of deep learning models

Pending Publication Date: 2020-11-03
深圳慕智科技有限公司
View PDF2 Cites 2 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Today's anti-sample defense methods are still unable to achieve comprehensive defense, and can only be effective against specific attack methods

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 learning model optimization method based on network addition/modification
  • Deep learning model optimization method based on network addition/modification
  • Deep learning model optimization method based on network addition/modification

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0022] Embodiments of the present invention are described below through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification.

[0023] This patent implements the optimization of the deep neural network model through adversarial training, and the specific key technologies involved include deep convolutional neural network (CNN), adversarial example (Adversarial Example), adversarial example defense technology, etc.

[0024] 1. Adversarial sample generation

[0025] In the present invention, we use the convolutional neural network as the main optimization target to generate adversarial examples for the data set in the form of pictures. Convolutional neural networks are a class of feed-forward neural networks that include convolutional computations and have deep structures. The convolutional neural network has the ability of representation learning, and can per...

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 deep learning model optimization method based on network addition / modification, in particular to an optimization method integrating a model level in the field of deep learning model optimization, and specifies an optimization strategy for adversarial samples by adopting an evaluation feedback mechanism. A defense strategy is formulated by evaluating a model optimized byusing a defense method and evaluating a feedback mechanism, and an optimal defense means is selected for coping with different attack methods. In order to optimize the model for the adversarial sampleattack method, the defense strategy of the model level is to modify the network, and the structure of the original DNN model is modified in the training stage, or the original model is not changed, and an external model is used as an additional network, so the defended DNN classifier can detect the adversarial sample or identify the adversarial sample as a correct label.

Description

technical field [0001] The invention belongs to the field of intelligent software testing, and in particular relates to the optimization of deep learning models. For the model that needs to be optimized, the model-level optimization method is used to modify the model or add the network to realize the optimization of the model, so that the model can obtain higher accuracy. Background technique [0002] In recent years, deep learning theory and technology have continued to mature, and good application results have been achieved in artificial intelligence, big data analysis, and security detection. played a key role. However, while bringing convenience, deep learning itself also has certain security issues. Such as the adversary's attack and the data privacy problem, which caused great concern in the field of security. The image field is a field where deep learning testing develops rapidly. This article discusses the optimization of deep learning models from the perspective ...

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): G06N3/04G06N3/08G06K9/62G06F11/36
CPCG06N3/08G06F11/3672G06N3/044G06N3/045G06F18/241
Inventor 房春荣顾逸飞吕军刘佳玮
Owner 深圳慕智科技有限公司
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com