Federated learning (FL) enables the training of a global model across multiple data owners (clients) without sharing raw data. This distributed architecture is orchestrated by a central server that aggregates the local models from the clients. In cases where the server is trusted but not all network nodes, differential privacy (DP) can be used to privatize the aggregated model by adding noise. However, this may affect convergence across the FL rounds. In this work, we build on the notion of metric-privacy as a design principle to calibrate the noise added by the server under a global-DP setting, with the objective of mitigating its impact on the convergence of the aggregated model. We do not enforce metric-privacy as a formal guarantee, but rather use it to guide noise calibration. We compare our approach with vanilla FL and global-DP by analyzing the impact on six aggregation strategies and applying it to a medical imaging use case, simulating different scenarios with homogeneous and non-i.i.d. clients. Finally, we introduce the client inference attack (CIA), where a semi-honest client tries to find whether another client participated in the training and study how it can be mitigated using DP and metric-aware noise calibration. Our experiments show that metric-privacy aware noise calibration strategy improves the accuracy compared to standard DP in all the scenarios analyzed, while achieving a comparable success rate against CIA. These results indicate that metric-privacy inspired noise calibration can deliver a superior utility-privacy trade-off in medical-imaging federated settings.

We propose DocVXQA, a novel framework for visually self-explainable document question answering, where the goal is not only to produce accurate answers to questions but also to learn visual heatmaps that highlight critical regions, offering interpretable justifications for the model decision. To integrate explanations into the learning process, we quantitatively formulate explainability principles as explicit learning criteria. Unlike conventional relevance map methods that solely emphasize regions relevant to the answer, our context-aware DocVXQA delivers explanations that are contextually sufficient yet representation-efficient. This fosters user trust while achieving a balance between predictive performance and interpretability in document visual question answering applications. Extensive experiments, including human evaluation, provide strong evidence supporting the effectiveness of our method.

Federated Learning (FL) enables clients to train a joint model without disclosing their local data. Instead, they share their local model updates with a central server that moderates the process and creates a joint model. However, FL is susceptible to a series of privacy attacks. Recently, the source inference attack (SIA) has been proposed where an honest-but-curious central server tries to identify exactly which client owns a specific data record.

Federated learning (FL) is a particular type of distributed, collaborative machine learning, where participating clients process their data locally, sharing only updates of the training process. Generally, the goal is the privacy-aware optimization of a statistical model's parameters by minimizing a cost function of a collection of datasets which are stored locally by a set of clients. This process exposes the clients to two issues: leakage of private information and lack of personalization of the model. To mitigate the former, differential privacy and its variants serve as a standard for providing formal privacy guarantees. But often the clients represent very heterogeneous communities and hold data which are very diverse. Therefore, aligned with the recent focus of the FL community to build a framework of personalized models for the users representing their diversity, it is of utmost importance to protect the clients' sensitive and personal information against potential threats. To address this goal we consider $d$-privacy, also known as metric privacy, which is a variant of local differential privacy, using a metric-based obfuscation technique that preserves the topological distribution of the original data. To cope with the issues of protecting the privacy of the clients and allowing for personalized model training, we propose a method to provide group privacy guarantees exploiting some key properties of $d$-privacy which enables personalized models under the framework of FL. We provide theoretical justifications to the applicability and experimental validation on real-world datasets to illustrate the working of the proposed method.

Digital data are an essential resource for intelligent decision making. As the value of digital data increases, digital markets, where data owner and consumers can deal with data, have also been attracting attention as a means to obtain data. However, the collection of digital data can lead to privacy breaches, which are a substantial impediment that hinders an individual's willingness to provide data. Differential privacy, which is a de facto standard for privacy protection in statistical databases, can be applied to solve the privacy violation problem. To apply differential privacy to the data market, the amount of noise and corresponding data price must be determined; however, this matter has not yet been studied. In this work, we propose a fair negotiation method that can set the appropriate price and noise parameter in the differentially private data market environment. We suggest a data market framework with a market manager that acts as a broker between the data provider and consumer. We also propose a negotiation technique to determine the data price and noise parameter ε using Rubinstein bargaining considering social welfare to prevent unfair transactions. We validate that the proposed negotiation technique can determine an appropriate level of ε and unit price without unfair trade to either the data provider and the consumer.

As the value of digital data increases, the data market is in the spotlight as a means of obtaining a personal information. However, the collection of personal information makes a serious privacy violation and it is a serious problem in the use of personal data. Differential privacy, which is a de-facto standard for privacy protection in statistical databases, can be applied to solve the privacy violation problem. To apply differential privacy to the data market, the amount of noise and corresponding data price should be determined between the provider and consumer. However, this matter has not yet been studied. In this work, we introduce a Privata which is a differentially private data market framework to set the appropriate price and noise parameter in the data market environment. The Privata is based on negotiation technique using Rubinstein bargaining considering social welfare to prevent unfair transactions. We explain the Privata overview and negotiation technique in Privata, and show the Privata implementation.

Digital data on the Web are nowadays regarded significant sources of information for marketing and user profiling, etc. However, digital data are risky sources of privacy violation. To address privacy breaches, we can use differential privacy, which has become the de facto standard for privacy protection in statistical databases. However, problems need to be solved, including those related to noise parameter configuration, even before differential privacy can be applied into the real world. In this study, we introduce a linkage attack to identify a user with different nicknames for each subservice on a hue online portal service. In addition, we propose a configuration technique for the upper bound of noise parameter ε to prevent linkage attack. We demonstrate the linkage attack with experiments by using real-world online portal service data. Finally, we validate the proposed configuration technique.

The number of users who uses location-based services (LBS) which utilize a user's location information to provide beneficial services has been increasing rapidly along with the proliferation of mobile devices such as the smartphone. However, LBS users have expressed concerns regarding their privacy because the collected location information can bring about privacy violations if used maliciously by the LBS provider. Thus, developing privacy-preserving techniques to prevent the exposure of the user's location information is important for the wide application of LBS. Consequently, many researches are being conducted on topics such as location k-anonymity and pseudonym to prevent privacy threats. However, existing research is difficult to apply in real-world LBS because of several limitations such as additional computation and legacy system change cost. In this paper, we propose a P2P architecture to protect the user's location privacy in LBS using personal data storage (PDS) to overcome the drawback in existing techniques. We also introduce a collaborative caching technique, which shares extra query results among users to mitigate performance degradation, that is, for preserving location privacy. Simulation results show that the proposed technique achieves sufficient privacy protection without a trusted third party for anonymization while minimizing system performance degradation.

The increasing number of mobile device users indicates the expansion of personalized location-based services (LBS). Despite their proliferation, the risk of violating users’ privacy by exposing user's location information remains. For this reason, many studies have researched to prevent privacy violation in LBS. However, previous researches only focused on protecting users’ location information without considering semantic location privacy violation through contextual information. In this paper, we explain the process of inferring a user's behavior using semantic information which includes spatial and temporal information. We also suggest a privacy preserving technique to prevent exposure of sensitive behavior in semantic LBS. We implement an android application to validate the proposed technique. In accordance with the experimental results, the proposed b-diversity technique is validated to prevent exposure of sensitive behavior and also minimizing data utilization degradation.