公有数据上的高效隐私保护外包计算：EPOC框架与SHED加密
183 浏览量
更新于2024-08-28
收藏 1.11MB PDF 举报
本文主要探讨了公开数据上的高效隐私保护外包计算(Efficient Privacy-Preserving Outsourced Computation, EPOC)，这是一种新兴的研究领域，特别是在云计算背景下，如何确保用户在将数据和计算任务委托给云服务的同时，能够有效地保护函数及其结果的隐私性。作者Ximeng Liu、Baodong Qin、Robert H. Deng和Yingjiu Li提出了EPOC框架，旨在平衡隐私保护程度和性能需求。
EPOC框架的核心是Switchable Homomorphic Encryption with Partial Decryption (SHED)加密系统，它允许对多维度的公共数据进行计算而无需完全暴露原始数据。SHED的特点在于提供了灵活的加密处理能力，使得数据能够在保持部分解密功能的同时，保护数据的隐私。
为了优化通信效率和降低计算复杂度，文章引入了两种编码技术。首先，Message Pre-coding Technique通过对加密消息进行预处理，提高了加密数据的处理效率。其次，Message Extending and Coding Technique针对复合订单群中的加密信息，设计了一种扩展和编码策略，进一步提升了计算的性能。
此外，文中还提出了Secure Exponent Calculation Protocol with Public Base (SEPB)，作为EPOC框架内的关键子协议。SEPB负责执行安全的指数运算，确保在保护隐私的同时，可以顺利完成计算任务。
安全性分析表明，EPOC实现了对公开数据的高效且隐私保护的外包计算目标。它通过巧妙地结合高级加密技术和有效的编码策略，使得用户可以在享受云计算便利的同时，有效保障个人数据的隐私不受侵犯。这种创新的方法对于数据密集型应用，如大数据分析和人工智能，具有重要意义，有助于推动隐私保护与云计算之间的和谐发展。
3
c
1
= (T
2
· g
s
)
θ
1
= g
(r+s)θ
1
mod N and c
2
= g
s
mod N ,
then sends {T
1
, c
1
, c
2
} to proxy.
Once received {T
1
, c
1
, c
2
}, the proxy computes (c
1
)
θ
2
=
h
r+s
and its inverse (h
r+s
)
−1
mod N . Then, it computes
T
0
1
= T
1
(h
r+s
)
−1
mod N
2
= E
pk
+
(mh
−s
).
and c
0
2
= (c
2
)
θ
2
= g
sθ
2
mod N , and sends {T
0
1
, c
0
2
} to the
server.
The server uses {T
0
1
, c
0
2
} to compute (c
0
2
)
θ
1
= h
s
mod N .
Then, the ADD ciphertext can be recovered by calculating
(T
0
1
)
h
s
= E
pk
+
(mh
−s
· h
s
) = E
pk
+
(m).
III. SYSTEM & PRIVACY MODELS
In this section, we formalize the EPOC system and privacy
models.
A. System Model
Our system consists of four types of generic entities: Trusted
Authority (TA), User, Public Cloud Platform (PCP) and Private
Computation Cloud (PCC).
1) Trusted Authority (TA): TA is assumed to be trusted by
all the other entities to distribute and manage all the private
& public keys in the system.
2) User: The goal of a user is to get computed results
over public database according to a function of his choice.
The user sends a query to PCP, specifying the function in a
privacy-preserving manner. The PCP computes the results and
returns them back to the user. Note that the results can only
be decrypted with the help of PCC, i.e., the user cannot get
the finally results without the authorization of PCC.
3) Private Computation Cloud (PCC) : PCC provides
computation service to the user, e.g., it can process encrypted
data such as multiplication over the encrypted data. Also, PCC
has the ability to partially decrypt ciphertexts which are sent
from PCP, do some calculations over the plaintexts, and then
re-encrypt the data with corresponding user’s public key.
4) Public Cloud Platform (PCP) : PCP contains unlimited
data storage space which stores and manages all the public
data. PCP stores all the intermediate and final results in
encrypted form. Furthermore, PCP has some computation
power to perform certain calculations over encrypted data.
Next, we present the dataset and the function used in
our EPOC. Suppose a data space D defined by a set of
γ dimensions {x
1
, ··· , x
γ
} and a dataset S on D with
cardinality γ. A transaction
~
a
i
∈ S can be represented
as
~
a
i
= (a
i,1
, ··· , a
i,γ
), where a
i,k
∈ G
2p
0
q
0
is a value
on dimension d
k
(i = 1, ··· , τ, k = 1, ··· , γ). All the
transactions are stored as plaintext form in PCP and the dataset
S can be accessed by any party in the system (include the
adversary A
∗
defined in III-B). A user α’s goal is to obtain
the output of the function
F : o =
κ
X
j=1
C
j
x
t
j,1
1
···x
t
j,γ
γ
,
3
3
All the functions used in EPOC are defined over ring of integers Z, thus
the function must satisfy the following restriction: 1) the size of output of the
function must less than N, i.e. |o| < N . 2) the size of every monomials must
less N , i.e. |C
j
x
t
j,1
1
· · · x
t
j,γ
γ
| < N for every j.
over the public data stored in PCP, where C
j
∈ G
2p
0
q
0
and
t
j,k
∈ Z
N
. This kind of functions can be used for statistical
analysis. For example, the user can calculate the arithmetic
mean x = (
P
γ
i=1
x
i
)/γ across different dimensions. The stan-
dard deviation σ =
q
1
γ
P
γ
i
(x
i
− x)
2
can also be calculated
(note that X =
1
γ
P
γ
i
(x
i
− x)
2
, a special case of F, can be
computed on server side, the user can calculate
√
X by himself
to get the standard deviation). Moreover, for polynomial kernel
SVM classification, the classifier is to map the data from the
input space X to a feature space F . In the space F , the
discriminant function is f(x) =
P
n
i=1
k(x, x
i
) + b, where
k(x, x
0
) = (x
T
x
0
+ 1)
d
is a degree-d polynomial kernel.
B. Privacy Model
In our privacy model, PCP and PCC are curious-but-
honest parties in that they strictly follow the protocol, but
are interested in user’s private information, e.g., both PCP and
PCC are curious about user’s private outsourced function and
calculated results. EPOC should meet the following privacy
requirements.
• User Privacy: protect privacy of user’s function and the
output of the function from PCP, PCC and any external
adversary. That is, PCP, PCC and any external adversary
cannot learn anything about user’s function in the query
as well as anything about the output of the function.
• Privilege Separation: It can prevent the user to the abuse
of privilege without the permission of the PCC and allow
PCC to efficiently revoke the user’s private key.
• Security of Data-in-transit: guarantee confidentiality of
data during transmission.
To satisfy these privacy requirements, we assume that there
exist an active adversary A
∗
in our system. The goal of A
∗
is to compromise user’s privacy
4
with the following attacking
abilities: A
∗
may eavesdrop on all the communication links to
get the challenge user’s encrypted queries, encrypted middle
results and encrypted final results. Also, A
∗
may compromise
PCP, PCC, and users simultaneously, but subject to the follow-
ing restrictions: 1) A
∗
cannot compromise PCP and PCC at
the same time. 2) A
∗
cannot compromise the challenge user.
For efficiency concern, we achieve the aforementioned
privacy requirements by defining the following three privacy
levels.
Definition 1. (Level-I Privacy): Upon completion of an EPOC
protocol, A
∗
cannot learn the coefficients C
j
(j = 1, ··· , κ)
of the outsourced function F and the final outputs.
Definition 2. (Level-II Privacy) Upon completion of an EPOC
protocol, A
∗
cannot learn the coefficients C
j
(j = 1, ··· , κ)
If the function is defined over Z
N
, the function which used in EPOC has
no restriction, i.e., the function is formalized as
F : o =
κ
X
j=1
C
j
x
t
j,1
1
· · · x
t
j,γ
γ
mod N.
4
Before attacking a user, the adversary A
∗
should issue a target and try
to guess the plaintext value of the target user’s ciphertext with some existing
resources. We call the target user as the challenge user.
剩余13页未读，继续阅读
点击了解资源详情
点击了解资源详情
点击了解资源详情
2021-03-07 上传
2021-02-25 上传
2020-02-26 上传
2022-07-01 上传
2021-02-08 上传
2021-03-07 上传
weixin_38501916
- 粉丝: 2
- 资源: 934
我的内容管理
展开
最新资源
- Multi-Task-Learning:多任务学习的论文，代码和应用程序列表
- 计算机三级-第8章 无线局域网设备安装与调试.zip
- parrot-bot:HTTP-IRC 网关
- 学习MySQL的资料和练习.zip
- VC.NET获取所有的ODBC驱动程序名称
- redstock：RedStock是产品和库存管理软件
- wnetwrap:Wininet包装器-简单的https库
- voice-commands-with-wordnet:轻松映射无数语音命令-完全脱机！
- 最新版windows jdk-17_windows-x64_bin.zip
- underscore.vim:Vim 脚本实用程序库
- VC++制作文字闪烁变色的启动窗体特效
- minecraft.github.io
- Raspberry Pi-电动糖果分配器-项目开发
- Hadoop-2.8.0-Day08-Hive函数与HQL详解-课件与资料.zip
- JavaLine:我的java学习行。 请注意
- basic-search-engine:使用BTree和位图的搜索引擎
