Obfuscation-based inference control
The protocols discussed in the previous section provide strong (cryptographic) guarantees regarding the Confidentiality of data. Such strong protection, however, comes at the cost of efficiency and flexibility. On one hand, privacy-preserving cryptographic primitives require significant resources in terms of computation and/or bandwidth. Social Innovation Expert in Australia
On the other hand, they narrow down the type of processing that can be done on data. This is inherent to cryptographic constructions that fix inputs and outputs, and strictly define what information will be available after the protocol is executed. In this section, we describe approaches to protect data Confidentiality based on obfuscating the data exposed to an adversary. These techniques provide a more relaxed definition of Confidentiality than cryptography, Analytics Reporting Expert in Australia
in the sense that they cannot completely conceal information. Instead, their goal is to provide a way to control the extent to which an adversary can make inferences about users’ sensitive information. In fact, for most of these techniques, the level of protection depends on the concrete data and adversarial knowledge.
Thus, it is important to run an ad-hoc analysis for the inference capability, as explained in Section 5.5. Also, we note that the privacy gained from these techniques is based on limiting the information available to one’s adversary. Consequently, these techniques reduce the amount of information available for anyone and, hence, may have an impact on utility if the purpose of the application is based on sensitive information, e.g., finding matches on dating applications. However, we note that when the sensitive information is not crucial for the purpose of the application these techniques may be deployed without affecting utility, e.g., a weather application that can operate using very rudimentary location data. Obfuscation-based inference control techniques are not suitable for protecting data in transit, but can be used to support privacy-preserving outsourcing, privacy-preserving collaborative computations, and privacy-preserving publishing. There are four main techniques to obfuscate data, as described below. We note that these techniques are mostly oriented to obfuscate numerical or categorical fields. Obfuscating more complex content, such as free text, is a much more difficult task due to correlations that are hard to remove in a systematic manner. To date, there are no known techniques that can reliably anonymise free text. However, these techniques are quite effective at reducing the information leaked by Metadata, as we discuss in Section
Comments
Post a Comment