读书笔记（七）--IoT Security
Li, Chao, and Balaji Palanisamy. “Privacy in Internet of Things: from Principles to Technologies.” IEEE Internet of Things Journal (2018).
A. Problem Statement
The paper targets the problem of privacy in the Internet of Things (IoT) from a technical and a legal standpoint. Particularly, how does privacy technology work at each IoT layer?
B. Problem Significance
Nowadays, all sensitive data are everywhere like connected cars and smart health application where store huge amount information about us. Not just the data from these objects, there is a problem the web traffic, metadata, between the objects in your home and the cloud services that are behind can tell your lifestyle. Most people are not happy that their privacy has been leaked, or they are worried about what their data will be used for. Nowadays, all sensitive data are everywhere like connected cars and smart health application where store huge amount information about us. Not just the data from these objects, there is a problem the web traffic, metadata, between the objects in your home and the cloud services that are behind can tell your lifestyle. Most people are not happy that their privacy has been leaked, or they are worried about what their data will be used for.
C. State of the Art
IoT was originally referred to as “embedded internet”. RFID, sensors, and actuators are embedded in physical objects which can talk to the Internet. Privacy enhancing technologies (PETs) help people prevent information from being leaked, or know more about what their information is being used for. Privacy by Design (PbD) means that put PETs directly into the design of software and system.
k-anonymous and differential privacy are introduced here when dealing with data in the perception layer. For K individuals, first, remove his identity attribute. For other attributes, see whether the attribute classification below it is greater than K. If it is greater than K, modify it. If not, just leave it. Differential privacy is the integration of individual analysis into a trend or a large population study. We can use Epsilon to process the data, adding Gaussian noise, which in my understanding is like an image plus a Gaussian blur effect.
Encryption still uses AES, RSA and SHA-2, which are well known to everyone.
Differential privacy和k-anonymous 的定义还没有能比较通俗易懂的说明白，欢迎大家评论赐教！
D. Contributions
This paper is very comprehensive about PETs, from historical development to definition as well as traditional technologies and state-of-the-art. In particular, how the various layers of IoT work and what they do to meet the privacy requirements. Their analysis has shown differential privacy and Tor demonstrate a great potential for use in the IoT.
And IoT architecture can be divided into 4 layers and PbD has 8 strategies. Each layer is responsible for implementing different strategies, as is shown in Table I.
| IoT Layered Architecture | Strategies |
|---|---|
| Perception Layer | Minimize, Aggregate, Hide, Separate |
| Networking Layer | Hide and Minimize |
| Middleware Layer | Inform, Control, Demonstrate, Enforce |
| Application Layer | — |
The paper compares k-anonymity and differential privacy with IoT specific 8 challenges in perception layer, differential privacy is a winner because of composability feature to reduce information and increase the granularity. However, for healthy devices that require accurate data, differential privacy will not work well. And the encryption of the perception layer should take into account the capacity of IoT devices. Some traditional algorithms cannot meet the requirements because of high-cost. Attribute-Based Encryption (ABE) still has its own problems with key distribution, and in the future, this aspect can comply with the design requirements of NIST.
8大策略和8大挑战请看原文
In the networking layer, the communication secured by IPsec or TLS in the transport layer can only hide the content of messages. It is still good for us to use anonymization in the layer. And Virtual Private Network (VPN) and the onion router (Tor) can help with it. The Tor design is wonderful that each node only knows the IP of its predecessor and successor. But still, need to work out voice over the internet because Tor does not support UDP.
In the middleware layer, interaction-enhancing PETs fulfilling Inform and Control strategies. As for Inform strategy, I think it is important to let people know clearly privacy policies. For Control strategy, I agree that we should focus on designing a centralized platform. A Trusted Third Party (TTP) play an important role in compliance-enhancing. TTP can distribute keys, check whether the policy preference sticks with data or not.
It talks about the risks faced by the application layer and the issues to be aware of when designing an app.
E. Remaining Questions
k-anonymous is not a good way to high-dimensional data sets, but differential privacy is only usable for large datasets because of the injected noise. What method will we use in a medium data set? Are we mixing the two technologies in the perception layer? Or still use different methods depending on
the purpose, such as general-use for differential privacy, personal data analysis still uses k-anonymous?