A method and device for optimizing resources of end-to-end collaborative inference combining non-orthogonal multiple access and speculative sampling, a terminal and a medium

By constructing an edge-to-edge collaborative inference system with non-orthogonal multiple access and speculative sampling, resource allocation is optimized, solving the problems of insufficient computing power of terminal devices and cloud offloading latency. This achieves low-latency and efficient resource management and improves the performance of mobile intelligent applications.

CN122420114APending Publication Date: 2026-07-17PENG CHENG LAB
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
PENG CHENG LAB
Filing Date
2026-04-29
Publication Date
2026-07-17

Smart Images

  • Figure CN122420114A_ABST
    Figure CN122420114A_ABST
Patent Text Reader

Abstract

本发明公开了一种联合非正交多址和投机采样的端边协同推理的资源优化处理方法、装置、终端及介质，包括：构建基于非正交多址和投机采样的端边协同推理系统模型；接收到各终端发送的预测数据，并使用验证模型对所述预测数据中的预测词元进行验证，将验证结果数据发送给对应的终端；根据所有终端推理请求的所需时间，建立对预测词元序列长度、非正交多址的上行和下行传输时长、终端运行算力、边缘服务器运行算力的联合优化模型；求解所述联合优化模型，并根据求解的最优解对所述端边协同推理系统模型进行资源优化处理。本发明提供了一种投机采样策略、数据上传传输时长，终端算力配置策略的联合优化方法，提高了通信及计算资源的利用效率。
Need to check novelty before this filing date? Find Prior Art