Local differential privacy is a quantitative privacy metric based on indistinguishability, and the increasing complexity of recent attacks on privacy information has fostered its adoption. WALDP is a recently proposed local differential privacy mechanism, which reduces the amount of noise using dimension reduction. We focus on this mechanism and propose another dimension reduction method DR.OR using odds ratios. It is possible to maintain the original concept, i.e., data generality, by using DR.OR as a subroutine of WALDP. Previous study focused only on support vector machines (SVMs), but we also evaluated WALDP for the logistic regression and the deep neural network, and showed that WALDP achieves high accuracy as well. In particular, the proposed DR.OR boasts high accuracy, especially for logistic regression, which has a strong relationship with odds ratios. This suggests that further performance improvements can be expected by selecting a dimension reduction method in consideration of the learning model.

With the spread of IoT devices, various data about our lives are being collected, such as heart rate, physical activity, number of steps, pulse, oxygen intake, calorie consumption, etc. If these data can be analyzed, it will be possible to learn the signs of disease. However, it is dangerous for a person’s activity status to be managed on an external server with the view of privacy. To solve this problem, local differential privacy (LDP), which is a technique for randomly adding local noise to data, has been proposed. While ensuring privacy by LDP is certainly important, it degrades the usefulness of the analysis of data with added noise. In this paper, we propose a new mechanism to protect data privacy in both phases of training and testing based on LDP. We also make sure feasibility of our mechanism in two cases of breast cancer screening data and ionosphere data set.

A huge number of documents such as news articles, public reports, and personal essays has been released on websites and social media. Once documents including privacy-sensitive information are published, the risk of privacy breaches increases; thus, documents should be carefully checked before publication. In many cases, human experts redact or sanitize documents before publishing; however, this approach is sometimes inefficient with regard to its cost and accuracy. Furthermore, critical privacy risks may remain in the documents. In this paper, we present a generalized adversary model and apply it to document data. This paper devises an attack algorithm for documents, which uses a web search engine, and proposes a privacy-preserving algorithm against the attacks. We evaluate the privacy risks for real accident reports from schools and court documents. As experiments using the real reports, we show that human-sanitized documents still include privacy risks, and our proposal would contribute to risk reduction.

Time-sequence data is high dimensional and con- tains a lot of information, which can be utilized in various fields, such as insurance, finance, and advertising. Personal data including time-sequence data is often converted to anonymized datasets, which need to strike a balance between both privacy and utility. In this paper, we consider low-rank matrix decomposition as one of the anonymization methods and evaluate its efficiency. We convert time-sequence datasets to matrices and evaluate both privacy and utility. The record IDs in time-sequence data are changed at regular intervals to reduce re-identification risk. However, since individuals tend to behave in a similar fashion over periods of time, there remains a risk of record linkage even if record IDs are different. Hence, we evaluate the re- identification and linkage risks as privacy risks of time-sequence data. Our experimental results show that matrix decomposition is a viable anonymization method and it can achieve better utility than existing anonymization methods.

Secure two-party comparison plays a crucial role in many privacy-preserving applications, such as privacy-preserving data mining and machine learning. In particular, the available comparison protocols with the appropriate input/output configuration have a significant impact on the performance of these applications. In this paper, we firstly describe a taxonomy of secure two-party comparison protocols which allows us to describe the different configurations used for these protocols in a systematic manner. This taxonomy leads to a total of 216 types of comparison protocols.We then describe conversions among these types. While these conversions are based on known techniques and have explicitly or implicitly been considered previously, we show that a combination of these conversion techniques can be used to convert a perhaps less-known two-party comparison protocol by Nergiz et al. (IEEE SocialCom 2010) into a very efficient protocol in a configuration where the two parties hold shares of the values being compared, and obtain a share of the comparison result. This setting is often used in multi-party computation protocols, and hence in many privacy-preserving applications as well. We furthermore implement the protocol and measure its performance. Our measurement suggests that the protocol outperforms the previously proposed protocols for this input/output configuration, when off-line pre-computation is not permitted.

The privacy of individuals included in the datasets must be preserved when sensitive datasets are published. Anonymization algorithms such as k-anonymization have been proposed in order to reduce the risk of individuals in the dataset being identified. k-anonymization is the most common technique of modifying attribute values in a dataset until at least k identical records are generated. There are many algorithms that can be used to achieve k-anonymity. However, existing algorithms have the problem of information loss due to a tradeoff between data quality and anonymity. In this paper, we propose a novel method of constructing a generalization hierarchy for k anonymization algorithms. Our method analyses the correlation between attributes and generates an optimal hierarchy according to the correlation. The effect of the proposed scheme has been verified using the actual data: the average of k of the datasets is 83:14, and it is around 1=3 of the value obtained by conventional methods.