Inference method and system for dynamic branch selection of deep learning model

A technology of deep learning and reasoning methods, applied in the field of artificial intelligence deep learning reasoning, can solve problems such as feature redundancy and insufficient reasoning speed of edge computing equipment, and achieve the effect of improving reasoning speed and reducing computing redundancy
CN112446439AActive Publication Date: 2021-03-05魔视智能科技(上海)有限公司
8 Cites 2 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
魔视智能科技(上海)有限公司
Publication Date
2021-03-05

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The invention provides an inference method and system for dynamic branch selection of a deep learning model, and aims to solve the problem of redundancy of deep features of a convolutional neural network in a decoding part in the inference process of a traditional deep learning model, and can automatically select corresponding decoding branches according to the category of primary features, each decoding branch is trained through the primary feature input and the corresponding output of the corresponding branch category to obtain different feature weight sets, and the method can focus on the processing of the primary features of the corresponding branch category and the recognition of deep features to complete more complex classification or representation generation, therefore, the computational redundancy is reduced, the reasoning speed is increased, and the actual application requirements of edge computing equipment are met.
Need to check novelty before this filing date? Find Prior Art

Description

technical field

[0001] The invention belongs to the technical field of artificial intelligence deep learning reasoning, and in particular relates to a reasoning method and system for dynamic branch selection of a deep learning model applied on an edge computing device. Background technique

[0002] The current edge computing terminal applications have more and more demands for deep learning vision algorithms, but due to the cost of current edge terminals, the running inference delay of high-precision and complex models cannot meet the practical application, and the accuracy of simple models cannot meet the requirements. This is because For data-intensive and complex tasks, complex convolutional neural network models are usually required to fit the tasks well, and models that are too lightweight will underfit.

[0003] In addition, during the inference process of deep learning visual models, there will be redundancy in the deep features generated by the convolutional neural n...

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