人工智能培训

搜索

机器学习论文:一种用于对抗攻击采样的量子主动学习算法(A quantum active learning algorithm for sampling agai

[复制链接]
fanjz 发表于 2019-12-9 14:34:01 | 显示全部楼层 |阅读模式
fanjz 2019-12-9 14:34:01 746 0 显示全部楼层
机器学习论文:一种用于对抗攻击采样的量子主动学习算法(A quantum active learning algorithm for sampling against adversarial  attacks)对抗性攻击代表了学习算法的严重威胁,并可能损害未来自治系统的安全性。 Khouryand Hadfield-Menell(KH)的一个定理为保证主动学习算法的鲁棒性提供了充分的条件,但有一个警告:必须知道相应分类问题的类别之间的最小距离。我们提出了一个理论框架,使我们能够将主动学习视为对最有希望的新点进行分类的样本,以便可以找到类之间的最小距离并使用定理KH。量子主动学习算法的复杂度是所用变量的多项式,如空间$ m $的维数和初始训练数据$ n $的大小。另一方面,如果有人用经典计算机复制此方法,则我们期望它将花费$ m $的指数时间,这就是所谓的“维数诅咒”的一个例子。
Adversarial attacks represent a serious menace for learning algorithms andmay compromise the security of future autonomous systems.A theorem by Khouryand Hadfield-Menell (KH), provides sufficient conditions to guarantee therobustness of active learning algorithms, but comes with a caveat: it iscrucial to know the smallest distance among the classes of the correspondingclassification problem.We propose a theoretical framework that allows us tothink of active learning as sampling the most promising new points to beclassified, so that the minimum distance between classes can be found and thetheorem KH used.The complexity of the quantum active learning algorithm ispolynomial in the variables used, like the dimension of the space $m$ and thesize of the initial training data $n$.On the other hand, if one replicatesthis approach with a classical computer, we expect that it would takeexponential time in $m$, an example of the so-called `curse of dimensionality'.机器学习论文:一种用于对抗攻击采样的量子主动学习算法(A quantum active learning algorithm for sampling against adversarial  attacks)
URL地址:https://arxiv.org/abs/1912.03283     ----pdf下载地址:https://arxiv.org/pdf/1912.03283    ----机器学习论文:一种用于对抗攻击采样的量子主动学习算法(A quantum active learning algorithm for sampling against adversarial  attacks)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

fanjz当前离线
新手上路

查看:746 | 回复:0

快速回复 返回顶部 返回列表