Neural network operation methods and apparatus

By replacing element-wise operations with shuffling and pooling operations in neural network operations, the dependence of the MAC arithmetic unit on weights and the limitations of hardware resources are resolved, thereby improving the execution speed and efficiency of neural network operations.

CN114154628BActive Publication Date: 2026-07-17SAMSUNG ELECTRONICS CO LTD

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
SAMSUNG ELECTRONICS CO LTD
Filing Date
2021-03-09
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

In existing technologies, MAC processors require weights when performing element-wise summation operations, and their hardware resource utilization and processing speed are limited, making it difficult to efficiently perform channel-wise MAC operations.

Method used

The elements of the matrix are shuffled by a shuffler, and pooling operations are used to replace element-wise operations, reducing the dependence on weights and improving hardware resource utilization and processing speed.

Benefits of technology

This approach improves the execution speed and efficiency of neural network operations without increasing hardware resources, reduces memory usage, and increases the utilization of the arithmetic unit.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN114154628B_ABST
    Figure CN114154628B_ABST
Patent Text Reader

Abstract

Disclosed are a neural network operation method and apparatus. A computer-implemented neural network operation method includes: storing a matrix on which an operation of a neural network is to be performed in a memory; reading the matrix on which the operation of the neural network is to be performed from the memory by a shuffler; shuffling a portion of elements of the matrix by the shuffler and sending the shuffled matrix to an operator; and performing a replacement operation of the operation based on the shuffled matrix by the operator.
Need to check novelty before this filing date? Find Prior Art