面向隐私保护的联邦学习任务优化调度研究
Oriented privacy-preserving optimization for federated learning task scheduling
吴霁霖 1金惠 1李昕镁 2莫磊2
作者信息
- 1. 中国烟草总公司贵州省公司信息中心,贵州贵阳 550004
- 2. 东南大学,江苏南京 210096
- 折叠
摘要
[目的/意义]人工智能技术的广泛应用,引发了数据安全和个人隐私保护的问题.联邦学习作为隐私保护的新型范式被广泛应用,但是由于任务实时性和设备能效性影响联邦学习过程的效率,因此需要对边缘设备的计算性能进行优化调度.[方法/过程]基于big.LITTLE多核异构平台,考虑近似计算任务,在能量和实时性的约束下,提出了基于服务质量(QoS)优化的任务映射问题,并被建模为混合整型非线性规划问题.同时,使用线性化方法,可以将该问题等价地转换为混合整型线性规划问题,并使用优化求解器求解.[结果/结论]通过真实应用任务集验证任务映射方法的可行性,仿真结果表明,提出的最优化算法能够显著提升系统QoS.
Abstract
[Purpose/Significance]Federated learning is widely used as a new paradigm for privacy protection,but real-time performance and energy efficiency affect the efficiency of the federated learning process.Therefore,it is necessary to optimize the scheduling of edge devices.[Method/Process]The iterative tasks of federated learning are modeled as imprecise computation tasks on heterogeneous multi-core platforms,so as to satisfy real-time and energy constraints and improve the system QoS.This problem is formulated as a mixed integer nonlinear programming(MINLP)problem and can be equivalently transformed into a mixed integer linear programming(MILP)problem solved by the optimization solver.[Results/Conclusion]The simulation results show that the proposed optimization algorithm can significantly improve the QoS under re-source-limited situations.
关键词
任务调度/近似计算/服务质量/隐私保护/数据安全Key words
task scheduling/imprecise computation/QoS/privacy protection/data security引用本文复制引用
基金项目
中国烟草总公司贵州省公司科技项目(2022XM27)
出版年
2024
NETL
NSTL
万方数据