A multi-agent collaborative distributed task optimization and learning system

By constructing soft-boundary state vectors and dynamic adjustment mechanisms, the problem that static segmentation schemes cannot adapt to the actual environment is solved, enabling edge agents to autonomously perceive and optimize segmentation, and improving the execution efficiency of deep learning tasks in industrial edge computing clusters.

CN122387671APending Publication Date: 2026-07-14
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Filing Date
2026-04-21
Publication Date
2026-07-14

AI Technical Summary

Technical Problem

In existing technologies, when multiple edge agents in an industrial edge computing cluster collaboratively execute deep learning tasks, static task partitioning schemes cannot be dynamically adjusted based on real-time computing saturation and collaborative waiting rate, resulting in partitioning schemes that are not compatible with the actual operating environment, leading to resource idleness or increased communication overhead.

Method used

A distributed task optimization and learning system employing multi-agent collaboration transforms traditional discrete partitioning decisions into continuous state variables by constructing soft-boundary state vectors, enabling the partitioning boundaries to be gradually adjusted during task execution. The system includes a task parsing module, a soft-boundary modeling module, a state awareness module, a boundary adjustment decision module, and a conflict resolution module, utilizing computational differences, real-time performance metrics, and trend analysis for dynamic adjustments.

Benefits of technology

It enables edge agents to autonomously perceive the degree of adaptation between segmentation granularity and collaborative state without a central scheduler, dynamically adjust segmentation boundaries, reduce communication overhead, improve the execution efficiency of distributed deep learning tasks, and avoid boundary oscillations and resource waste.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122387671A_ABST
    Figure CN122387671A_ABST
Patent Text Reader

Abstract

The application discloses a kind of multi-agent collaborative distributed task optimization and learning system, it is related to collaborative task optimization technical field.System includes: task analysis module obtains computation graph;Soft boundary modeling module constructs the soft boundary state vector including split tendency value, state confidence and direction inertia value for directed edge, based on the ratio of operator calculation amount and output data volume is corrected initially and segmented task fragment;State perception module collects local real-time running index to calculate adaptation degree score, determines bottleneck position by trend analysis and attribution inference;Boundary adjustment decision module calculates migration potential energy to select adjustment target, executes split tendency value directional increase and decrease and updates state confidence and direction inertia value;Conflict resolution module arbitrates conflict adjustment based on weighted priority;Convergence control module freezes split scheme after meeting convergence condition and locally reactivates when environment changes.The application realizes the self-adaptive adjustment of distributed task split granularity at runtime.
Need to check novelty before this filing date? Find Prior Art