Super-network search space construction method and device and electronic equipment

A technology of search space and construction method, applied in the field of neural network search and computer vision, it can solve problems such as inability to automatically search for models, affecting the speed and accuracy of models, and achieve the effect of satisfying search constraints and improving search efficiency.

Active Publication Date: 2019-12-20
BEIJING BAIDU NETCOM SCI & TECH CO LTD
View PDF4 Cites 4 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] At present, since the model depth in the hypernetwork is often fixed, under the constraints of model size and speed, the fixed model depth makes it impossible to automatically search for models of any depth.
Furthermore, it affects the speed and accuracy of the model searched in the search space of the super 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
  • Super-network search space construction method and device and electronic equipment
  • Super-network search space construction method and device and electronic equipment
  • Super-network search space construction method and device and electronic equipment

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0061] In a specific embodiment, such as figure 1 As shown, a hypernetwork search space construction method is provided, including:

[0062] Step S10: in the multi-layer feature map of the deep neural network, select two-layer feature maps;

[0063] Step S20: adding a connection layer between the selected two layers of feature maps, the connection layer is used to merge any two layers of feature maps;

[0064] Step S30: Construct the search space of the hypernetwork by using the connection layer.

[0065] In one example, a deep neural network may include an input layer, a hidden layer, and an output layer, each layer has a plurality of neurons, and neurons in each layer are connected to each other. In computer vision, the hidden layer is mainly used to extract features, so it can also be called feature layer. Such as figure 2 As shown, in the convolutional neural network, in each convolutional layer, the data exists in three-dimensional form. The convolutional neural net...

Embodiment 2

[0089] In another specific embodiment, such as Figure 6 As shown, a hypernetwork search space construction device 100 is provided, including:

[0090] The feature map selection module 110 is used to select two layers of feature maps in the multi-layer feature map of the deep neural network;

[0091] A connection layer adding module 120, configured to add a connection layer between the selected two-layer feature maps, the connection layer is used for;

[0092] The search space construction module 130 is configured to construct the search space of the hypernetwork by using the connection layer.

[0093] In one embodiment, as Figure 7 As shown, a hypernetwork search space construction device 200 is provided, and the feature map selection module 110 includes:

[0094] A feature map generation unit 1101, configured to input the feature map of the first layer to the multi-layer operation layer, and output the feature map of the Mth layer, where M is greater than or equal to 2; ...

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 super-network search space construction method and device and electronic equipment, and relates to the field of neural network search. According to the specific implementation scheme, two layers of feature maps are selected from multiple layers of feature maps of a deep neural network; a connecting layer is added between the two layers of selected feature maps, wherein the connecting layer is used for combining the two layers of selected feature maps; and a search space of the super network is built by utilizing the connection layer. The number of the feature layers is changed by adding the connecting layers, and the depth of the deep neural network is also changed. A new search space is constructed by using a connection layer, so that a deep neural network of anydepth can be obtained under the constraint conditions of any model size and model speed and precision.

Description

technical field [0001] The present application relates to the field of computer vision, in particular to the field of neural network search. Background technique [0002] Deep learning technology has achieved great success in many directions. In recent years, NAS technology (Neural Architecture Search, neural network architecture search) has become a research hotspot. NAS uses algorithms instead of tedious manual operations to automatically search for the best neural network architecture in a massive search space. The steps of performing the architecture search of the neural network include: firstly, the search space is defined and the search space is determined. Then, determine the search strategy according to the optimization algorithm adopted, such as reinforcement learning, evolutionary algorithm, Bayesian optimization and other algorithms. Finally, the search finds the speed of the model structure as well as the performance of the model. [0003] Hypernetwork-based s...

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): G06K9/00G06K9/46G06K9/62
CPCG06V20/58G06V10/454G06F18/2414Y02D10/00
Inventor 希滕张刚温圣召
Owner BEIJING BAIDU NETCOM SCI & TECH CO LTD
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