Program, information processing device, and information processing method

The program adjusts stride and dilation parameters within a single convolutional neural network model to balance computational complexity and accuracy, facilitating real-time processing on devices with varying computing powers.

JP7877265B2Active Publication Date: 2026-06-22KK TOSHIBA
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
JP2023065376
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2023-04-13
Publication Date
2026-06-22
Estimated Expiration
2043-04-13

AI Technical Summary

Technical Problem

Conventional frame skipping methods for neural networks fail to provide a flexible implementation for controlling the trade-off between computational amount and accuracy using a single model.

Method used

A program that functions as a memory control unit, conversion unit, first CNN unit, and second CNN unit, allowing for the conversion of stride and dilation parameters to adjust computational complexity and accuracy using a single convolutional neural network model.

Benefits of technology

Enables flexible implementation of a convolutional neural network model to control the trade-off between computational complexity and accuracy, enabling real-time processing on devices with varying computing powers while reducing memory requirements and development costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007877265000001
    Figure 0007877265000001
  • Figure 0007877265000002
    Figure 0007877265000002
  • Figure 0007877265000003
    Figure 0007877265000003
Patent Text Reader

Abstract

To provide flexible implementation that performs control over a trade-off between computational complexity and precision using one model when a convolutional neural network model is used.SOLUTION: A program causes a computer to function as a storage control unit, a conversion unit, a first CNN unit, and a second CNN unit. The storage control unit reads a first stride parameter and a first dilation parameter from a storage device. The conversion unit uses a conversion parameter to convert the first stride parameter into a second stride parameter and also the first dilation parameter into a second dilation parameter. The first CNN unit uses at least the second stride parameter to execute first CNN processing on a feature vector. The second CNN unit uses at least the second dilation parameter to execute second CNN processing based upon the output vector of the first CNN unit as an input.SELECTED DRAWING: Figure 4
Need to check novelty before this filing date? Find Prior Art

Citation Information

Patent Citations

  • Separate conveying device of parts

    JP1989053917A

  • Parameter optimization device, parameter optimization method, and parameter optimization program

    WO2021111633A1

  • Convolutional neural network acceleration device, convolutional neural network acceleration method, and convolutional neural network acceleration program

    WO2024053120A1