Search

This paper presents a new dataset, including behavioral, biometric, and environmental data, obtained from 23 subjects each spending 1 week to 2 months in smart rooms in Tokyo, Japan. The approximate duration of the experiment is 2 years. This dataset includes personal data, such as the use of home appliances, heartbeat rate, sleep status, temperature, and illumination. Although there are many datasets that publish these data individually, datasets that publish them all at once, tied to individual pseudo IDs, are valuable. The number of days for which data were obtained was 488, the number of records was 18,418,359, and the total size of the obtained data was 2.76 GB. This dataset can be used for machine learning and analysis for tips on getting a good night's sleep, for example.

Available data may differ from true data in many cases due to sensing errors, especially for the Internet of Things (IoT). Although privacy-preserving data mining has been widely studied during the last decade, little attention has been paid to data values containing errors. Differential privacy, which is the de facto standard privacy metric, can be achieved by adding noise to a target value that must be protected. However, if the target value already contains errors, there is no reason to add extra noise. In this paper, a novel privacy model called true-value-based differential privacy (TDP) is proposed. This model applies traditional differential privacy to the “true value” unknown by the data owner or anonymizer but not to the “measured value” containing errors. Based on TDP, the amount of noise added by differential privacy techniques can be reduced by approximately 20% by our solution. As a result, the error of generated histograms can be reduced by 40.4% and 29.6% on average according to mean square error and Jensen–Shannon divergence, respectively. We validate this result on synthetic and five real data sets. Moreover, we proved that the privacy protection level does not decrease as long as the measurement error is not overestimated.

Efforts to counter COVID-19 reaffirmed the importance of rich medical, behavioral, and sociological data. To make data available to many researchers who can conduct statistical analyses and machine learning, personally identifiable information must be excluded to protect individual privacy. It is essential to remove explicit identifiers, sample population data, and apply differential privacy, the de facto standard privacy metric. Despite the general belief that the risk of re-identification is insignificant when these techniques are applied, this study shows that even after applying these techniques, the risk of being re-identified is highly significant for some data. This study proposes in detail an algorithm for estimating the number of people in a population who have certain attribute values based on incomplete, differentially private databases. If the estimated number is one, the probability that only one person with that attribute value is present in the population is high, which means that there is a high probability of re-identification. Therefore, this study concludes that the re-identification risk must be evaluated even after applying state-of-the-art techniques to protect privacy.

With the development of IoT technology, personal data are being collected in many places. These data can be used to create new services, but consideration must be given to the individual’s privacy. We can safely collect personal data while adding noise by applying differential privacy. However, because such data are very noisy, the accuracy of machine learning trained by the data greatly decreased. In this study, our objective is to build a highly accurate machine learning model using these data. We focus on the decision tree machine learning algorithm, and, instead of applying it as is, we use a preprocessing technique wherein pseudodata are generated using a copula while removing the effect of noise added by differential privacy. In detail, the proposed novel protocol consists of three steps: generating a covariance matrix from the differentially private numerical data, generating a discrete cumulative distribution function from differentially private numerical data, and generating copula-based numerical samples. Simulation results using synthetic and real datasets verify the utility of the proposed method not only for the decision tree algorithm but also for other machine learning algorithms such as deep neural networks. This method will help create machine learning models, such as recommendation systems, using differential privacy data.

A sensor device can be used to detect target events at low cost. Moreover, there is a significant risk of sensor nodes being compromised or captured within large wireless sensor networks (WSNs). The transmission of valid event messages to users by a WSN can be hindered by network congestion due to false event messages in a compromised node. Several methods are available for detecting false event messages. However, they rely on long bits of authentication codes and hence do not provide a fundamental solution to prevent network congestion. In the proposed method, various hashing vectors, which are spaceefficient data structures that can determine whether the given data are an element of a set, are created based on the correct combination of authentication codes and placed in each node in advance. An event message contains an XOR of the authentication codes, and each node verifies it based on its hashing vector. If a node is acquired illegally, the information of the hashing vector and the XOR information of the authentication codes assigned to the correct event message is compromised, so we propose an algorithm to update the information securely. Compared to existing research, the number of hops required to detect a false event message increases by only about one hop, but the amount of traffic that a malicious node can generate can be reduced by about 60% or more. In other words, the proposed method effectively reduces the amount of traffic an attacker can generate with false event messages, which also reduces the overall network congestion.

Currently, many global organizations collect personal data for marketing, recommendation system improvement, and other purposes. Some organizations collect personal data securely based on a technique known as $\epsilon$-local differential privacy (LDP). Under LDP, a privacy budget is allocated to each user in advance. Each time the user's data are collected, the user's privacy budget is consumed, and their privacy is protected by ensuring that the remaining privacy budget is greater than or equal to zero. Existing research and organizations assume that each individual's data are completely unrelated to other individuals' data. However, this assumption does not hold in a situation where interaction data between users are collected from them. In this case, each user's privacy is not sufficiently protected because the privacy budget is actually overspent. In this study, the issue of local differential privacy for person-to-person interactions is clarified. We propose a mechanism that satisfies LDP in a person-to-person interaction scenario. Mathematical analysis and experimental results show that the proposed mechanism can maintain high data utility while ensuring LDP compared to existing methods.

Abstract Background The importance of privacy protection in analyses of personal data, such as genome-wide association studies (GWAS), has grown in recent years. GWAS focuses on identifying single-nucleotide polymorphisms (SNPs) associated with certain diseases such as cancer and diabetes, and the chi-squared (χ 2) hypothesis test of independence can be utilized for this identification. However, recent studies have shown that publishing the results of χ 2 tests of SNPs or personal data could lead to privacy violations. Several studies have proposed anonymization methods for χ 2 testing with ε-differential privacy, which is the cryptographic community’s de facto privacy metric. However, existing methods can only be applied to 2×2 or 2×3 contingency tables, otherwise their accuracy is low for small numbers of samples. It is difficult to collect numerous high-sensitive samples in many cases such as COVID-19 analysis in its early propagation stage. Results We propose a novel anonymization method (RandChiDist), which anonymizes χ 2 testing for small samples. We prove that RandChiDist satisfies differential privacy. We also experimentally evaluate its analysis using synthetic datasets and real two genomic datasets. RandChiDist achieved the least number of Type II errors among existing and baseline methods that can control the ratio of Type I errors. Conclusions We propose a new differentially private method, named RandChiDist, for anonymizing χ 2 values for an I×J contingency table with a small number of samples. The experimental results show that RandChiDist outperforms existing methods for small numbers of samples.

In recent times, advancements in text-to-speech technologies have yielded more natural-sounding voices. However, this has also made it easier to generate malicious fake voices and disseminate false narratives. ASVspoof stands out as a prominent benchmark in the ongoing effort to automatically detect fake voices, thereby playing a crucial role in countering illicit access to biometric systems. Consequently, there is a growing need to broaden our perspectives, particularly when it comes to detecting fake voices on social media platforms. Moreover, existing detection models commonly face challenges related to their generalization performance. This study sheds light on specific instances involving the latest speech generation models. Furthermore, we introduce a novel framework designed to address the nuances of detecting fake voices in the context of social media. This framework considers not only the voice waveform but also the speech content. Our experiments have demonstrated that the proposed framework considerably enhances classification performance, as evidenced by the reduction in equal error rate. This underscores the importance of considering the waveform and the content of the voice when tasked with identifying fake voices and disseminating false claims. [ABSTRACT FROM AUTHOR]

An increasingly popular class of software known as participatory sensing, or mobile crowdsensing, is a means of collecting people’s surrounding information via mobile sensing devices. To avoid potential undesired side effects of this data analysis method, such as privacy violations, considerable research has been conducted over the last decade to develop participatory sensing that looks to preserve privacy while analyzing participants’ surrounding information. To protect privacy, each participant perturbs the sensed data in his or her device, then the perturbed data is reported to the data collector. The data collector estimates the true data distribution from the reported data. As long as the data contains no sensing errors, current methods can accurately evaluate the data distribution. However, there has so far been little analysis of data that contains sensing errors. A more precise analysis that maintains privacy levels can only be achieved when a variety of sensing errors are considered.

Colorizing line drawings requires special skill, experience, and knowledge. Artists also spend a great deal of time and effort creating art. Given this background, research on automated line drawing colorization was recently conducted. However, the existing approaches present multiple problems, one of which is the inconsistency of the whites of the eyes (sclera) between line drawings and the results of colorizing. In particular, in line drawings, a person’s skin and sclera are often expressed in white. Hence, there are cases in which existing colorization methods cannot predict the boundary correctly. In this study, we propose automated colorization methods that use machine learning to segment sclera regions in grayscale line drawings. To improve the accuracy of previous automated colorization approaches, we implemented sclera-region detection and an automated colorizing approach on grayscale line drawings of people. In addition, we evaluated the colorization results created by our methods through a user study. Statistics show that our methods are somewhat superior to industrial application, but many of our respondents perceived little difference between the methods.

Research on deep-training captioning models that modify the natural-language contents of images and moving images has produced considerable results and attracted attention in recent years. In this research, we aim to generate recipe sentences from cooking videos acquired from YouTube. We treat this as an image-captioning task and propose two methods suitable for the work. We propose a method that adds a vector of a sentence already generated in the same recipe to the input of a captioning model. Then, we compare generated and correct sentences to calculate scores. We also propose a data-processing method to improve accuracy. We use several widely used metrics to evaluate image-captioning problems. We then train the same data with the simplest encoder–decoder model, compare it with correct recipe sentences, and calculate the metrics. The results indicate that our proposed methods help increase accuracy.

Non-deterministic behaviors of concurrent programs sometimes produce errors that depend on interleavings, such as race conditions. Unfortunately, executing all possible interleavings is usually not feasible because of their potential huge number. In our previous work, we used the existing reachability testing method to generate test cases and succeeded in reducing the number of test cases by eliminating redundant and infeasible ones. In this paper, we propose three further improvements. The first is to reduce the memory space required for generating test cases. We propose a new method by analyzing data dependency to generate only those test cases that might affect sequences of locks and shared variables. The second improvement is to generate test cases for detecting race conditions caused by accesses through reference variables. Our method can generate test cases for detecting such race conditions by creating test cases based on the data dependency of the reference variables. The third improvement is to reduce the effort involved in checking race conditions by utilizing previous test results. The new method can identify only those parts of the execution trace in which the sequence of locks and shared variables might be affected by a new test case, thus necessitating that race conditions be rechecked only for those affected parts. [ABSTRACT FROM AUTHOR]