A general sparse computing adaptation method and system for multi-source heterogeneous data

By implementing a unified format processing and dynamic routing strategy for multi-source heterogeneous data, the efficiency and accuracy issues of sparse computation on multi-source heterogeneous data are solved, achieving hardware and software decoupling and stability, and improving computational efficiency and compatibility.

CN122240990APending Publication Date: 2026-06-19SHANGHAI DEV CENT OF COMP SOFTWARE TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SHANGHAI DEV CENT OF COMP SOFTWARE TECH
Filing Date
2026-05-13
Publication Date
2026-06-19

AI Technical Summary

Technical Problem

Existing technologies struggle to effectively handle sparse computation on multi-source heterogeneous data, resulting in low utilization of computing resources and memory bandwidth, high adaptation costs across modalities and hardware platforms, and an inability to detect changes in the sparsity of input data during runtime.

Method used

By acquiring multi-source heterogeneous data, parsing it into a generalized coordinate sequence in a unified format, constructing a unified sparse batch structure, dynamically reconstructing dense operator nodes into dynamic routing nodes with dual-path execution capabilities, and combining a hardware performance evaluation mechanism to monitor sparsity indicators in real time and dynamically select the optimal computation path.

Benefits of technology

It achieves hardware and software decoupling, reduces invalid computation and memory usage, ensures the stability and compatibility of the algorithm, dynamically selects the optimal computation path, and improves the efficiency and accuracy of sparse computation adaptation.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122240990A_ABST
    Figure CN122240990A_ABST
Patent Text Reader

Abstract

This application discloses a general sparse computation adaptation method and system for multi-source heterogeneous data, relating to the technical field of multi-source data computation. The method includes: acquiring multi-source heterogeneous data; parsing the multi-source heterogeneous data to obtain a generalized coordinate sequence in a unified format; selecting a set of valid positions from the generalized coordinate sequence using initially preset sparsification parameters to construct a unified sparse batch structure; parsing the computation graph of the target neural network and identifying at least one dense operator node; reconstructing the dense operator node into a dynamic routing node with dual-path execution capability; real-time monitoring of the sparsity index of the unified sparse batch structure; and, combined with data dimension and hardware performance evaluation mechanisms, evaluating the estimated comprehensive performance under the sparse computation kernel path to obtain an evaluation result; and dynamically selecting an execution path based on the evaluation result. This application can effectively improve the efficiency and accuracy of the sparse computation adaptation process.
Need to check novelty before this filing date? Find Prior Art