Conformal Lyapunov Optimization: 确定性可靠性约束下的最优资源分配
Conformal Lyapunov Optimization: Optimal Resource Allocation under Deterministic Reliability Constraints
AI总结:
本文提出 conformal Lyapunov optimization (CLO) 框架,通过结合传统 LO 与 O-CRC 机制,在网络化系统中实现确定性长期可靠性约束下的平均长期目标优化,并在图像分割实验中有效控制假阴性率并最小化能耗与精度损失加权和。
摘要:
本文引入了 conformal Lyapunov optimization (CLO),这是一种用于网络化系统的新型资源分配框架,能在满足确定性长期可靠性约束的同时,优化平均长期目标。与在平均长期约束下处理资源分配任务的传统 Lyapunov optimization (LO) 不同,CLO 提供了正式的最坏情况确定性可靠性保证。这是通过将标准 LO 优化框架与 online conformal risk control (O-CRC)(一种控制长期风险的自适应更新机制)相集成来实现的。CLO 的有效性通过针对网络计算架构中图像分割任务的分层边缘推理实验得到了验证。具体而言,仿真结果证实,CLO 能够控制可靠性约束(通过网络中所有分割决策的假阴性率来衡量),同时最小化能耗与精度损失的加权和,其中后者代表假阳性率。
英文摘要:
This paper introduces conformal Lyapunov optimization (CLO), a novel resource allocation framework for networked systems that optimizes average long-term objectives, while satisfying deterministic long-term reliability constraints. Unlike traditional Lyapunov optimization (LO), which addresses resource allocation tasks under average long-term constraints, CLO provides formal worst-case deterministic reliability guarantees. This is achieved by integrating the standard LO optimization framework with online conformal risk control (O-CRC), an adaptive update mechanism controlling long-term risks. The effectiveness of CLO is verified via experiments for hierarchal edge inference targeting image segmentation tasks in a networked computing architecture. Specifically, simulation results confirm that CLO can control reliability constraints, measured via the false negative rate of all the segmentation decisions made in the network, while at the same time minimizing the weighted sum of energy consumption and precision loss, with the latter accounting for the rate of false positives.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。