VIP内容

主题: Locally Differentially Private (Contextual) Bandits Learning

摘要:

首先,我们提出了一种简单的黑盒归约框架,该框架可以解决带有LDP保证的大量无背景的bandits学习问题。根据我们的框架,我们可以通过单点反馈(例如 private bandits凸优化等)改善private bandits学习的最佳结果,并在LDP下获得具有多点反馈的BCO的第一结果。 LDP保证和黑盒特性使我们的框架在实际应用中比以前专门设计的和相对较弱的差分专用(DP)上下文无关强盗算法更具吸引力。此外,我们还将算法扩展到在(ε,δ)-LDP下具有遗憾约束ō(T~3/4 /ε)的广义线性bandits,这被认为是最优的。注意,给定DP上下文线性bandits的现有Ω(T)下界,我们的结果表明LDP和DP上下文bandits之间的根本区别。

成为VIP会员查看完整内容
0
12

最新内容

Virtual network functions (VNFs) have been widely deployed in mobile edge computing (MEC) to flexibly and efficiently serve end users running resource-intensive applications, which can be further serialized to form service function chains (SFCs), providing customized networking services. To ensure the availability of SFCs, it turns out to be effective to place redundant SFC backups at the edge for quickly recovering from any failures. The existing research largely overlooks the influences of SFC popularity, backup completeness and failure rate on the optimal deployment of SFC backups on edge servers. In this paper, we comprehensively consider from the perspectives of both the end users and edge system to backup SFCs for providing popular services with the lowest latency. To overcome the challenges resulted from unknown SFC popularity and failure rate, as well as the known system parameter constraints, we take advantage of the online bandit learning technique to cope with the uncertainty issue. Combining the Prim-inspired method with the greedy strategy, we propose a Real-Time Selection and Deployment(RTSD) algorithm. Extensive simulation experiments are conducted to demonstrate the superiority of our proposed algorithms.

0
0
下载
预览

最新论文

Virtual network functions (VNFs) have been widely deployed in mobile edge computing (MEC) to flexibly and efficiently serve end users running resource-intensive applications, which can be further serialized to form service function chains (SFCs), providing customized networking services. To ensure the availability of SFCs, it turns out to be effective to place redundant SFC backups at the edge for quickly recovering from any failures. The existing research largely overlooks the influences of SFC popularity, backup completeness and failure rate on the optimal deployment of SFC backups on edge servers. In this paper, we comprehensively consider from the perspectives of both the end users and edge system to backup SFCs for providing popular services with the lowest latency. To overcome the challenges resulted from unknown SFC popularity and failure rate, as well as the known system parameter constraints, we take advantage of the online bandit learning technique to cope with the uncertainty issue. Combining the Prim-inspired method with the greedy strategy, we propose a Real-Time Selection and Deployment(RTSD) algorithm. Extensive simulation experiments are conducted to demonstrate the superiority of our proposed algorithms.

0
0
下载
预览
参考链接
Top
微信扫码咨询专知VIP会员