BriefGPT.xyz
Jan, 2017
在线学习的差分隐私代价
The Price of Differential Privacy For Online Learning
HTML
PDF
Naman Agarwal, Karan Singh
TL;DR
本文提出了一种确保差分隐私的在线线性优化算法，其完全信息情况下的后果与epsilon无关，但在轮盘线性优化和非随机多臂匪徒的情况下，其遗憾上限是一个$ ilde{O}$函数，同时使时间复杂度在$\tilde{O}(\frac{1}{\epsilon}\sqrt{T}))$内。
Abstract
We design
differentially private algorithms
for the problem of
online linear optimization
in the full information and bandit settings with optimal $\tilde{O}(\sqrt{T})$
→