arXiv subjects

Tânia Carvalho

Publications and source records attributed to Tânia Carvalho.

12 recordsLinked to original sources

arXiv2026

SoK: Rethinking Membership Inference in Tabular Machine Learning

Membership Inference Attacks (MIAs) are currently the dominant approach to evaluating privacy in machine learning applications. Despite their significance, several concerns remain unexplored, particularly with regard to tabular data. Existing evaluations focus predominantly on high-dimensional datasets, leaving the privacy behavior of small-to-medium datasets unexplored. In this paper, we present a new taxonomy for MIAs and conduct a comprehensive study of the factors influencing vulnerability. We evaluate four distinct state-of-the-art attacks across diverse tabular datasets and model architectures. Our results show that datasets with higher class granularity and majority class percentage are more susceptible to MIAs. We further demonstrate that advanced MIAs are highly dependent on adversarial knowledge of the target model, while reference model-free attacks often achieve statistically comparable results. Most importantly, we show that records with unique signatures (single-outs) remain highly vulnerable, even when common defenses are applied. These findings indicate that current defenses reduce average leakage, but fail to adequately protect the most sensitive individuals.

cs.LG↗

arXiv2026

Privacy Vulnerabilities of Attention Layers in Tabular Foundation Models and Protection of High-Risk Queries

Tabular foundation models are commonly assumed to present limited privacy concerns as they are often pre-trained on large collections of synthetic data. However, these models leverage in-context learning, where sensitive records may be provided directly at inference time as labelled context examples. In this paper, we demonstrate that predictions generated via the attention mechanism leak sufficient information to enable effective Membership Inference Attacks (MIAs). To highlight this vulnerability, we propose AMIA (Attention-based Membership Inference Attack), a shadow-model-free attack that exploits the concentration of transformer attention patterns. Our results show that attention mechanisms reveal strong membership signals, which exceed classical confidence-based attacks, achieving an average gain of 7.7\%, specially in low false-positive regimes. To mitigate this risk, we introduce an inference-time defence inspired by $k$-anonymity principles. This approach reduces the uniqueness of context-key representations without introducing random noise or retraining the model. By targeting only high-risk queries identified through AMIA scores, the defence substantially reduces membership leakage of this attack by an average of 50\% and 25\% against confidence-based attacks, while preserving predictive utility with only 3.9\% performance degradation. Beyond showing that context examples are vulnerable, we further demonstrate that fine-tuning introduces an additional source of privacy risk. In particular, samples whose prediction confidence increases after fine-tuning become more susceptible to MIAs, indicating that fine-tuning can amplify memorisation and expose sensitive training information through confidence shifts.

cs.CR↗

arXiv2026

SoK: Challenges in Tabular Membership Inference Attacks

Membership Inference Attacks (MIAs) are currently a dominant approach for evaluating privacy in machine learning applications. Despite their significance in identifying records belonging to the training dataset, several concerns remain unexplored, particularly with regard to tabular data. In this paper, first, we provide an extensive review and analysis of MIAs considering two main learning paradigms: centralized and federated learning. We extend and refine the taxonomy for both. Second, we demonstrate the efficacy of MIAs in tabular data using several attack strategies, also including defenses. Furthermore, in a federated learning scenario, we consider the threat posed by an outsider adversary, which is often neglected. Third, we demonstrate the high vulnerability of single-outs (records with a unique signature) to MIAs. Lastly, we explore how MIAs transfer across model architectures. Our results point towards a general poor performance of these attacks in tabular data which contrasts with previous state-of-the-art. Notably, even attacks with limited attack performance can still successfully expose a large portion of single-outs. Moreover, our findings suggest that using different surrogate models makes MIAs more effective.

cs.LG↗

arXiv2025

Privacy-Driven Network Data for Smart Cities

A smart city is essential for sustainable urban development. In addition to citizen engagement, a smart city enables connected infrastructure, data-driven decision making and smart mobility. For most of these features, network data plays a critical role, particularly from public Wi-Fi infrastructures, where cities can benefit from optimized services such as public transport management and the safety and efficiency of large events. One of the biggest concerns in developing a smart city is using secure and private data. This is particularly relevant in the case of Wi-Fi network data, where sensitive information can be collected. This paper specifically addresses the problem of sharing secure data to enhance the quality of the Wi-Fi network in a city. Despite the high importance of this type of data, related work focuses on improving the safety of mobility patterns, targeting only the protection of MAC addresses. On the opposite side, we provide a practical methodology for safeguarding all attributes in real Wi-Fi network data. This study was developed in collaboration with a multidisciplinary team of legal experts, data custodians and technical privacy specialists, resulting in high-quality data. On top of that, we show how to integrate the legal considerations for secure data sharing. Our approach promotes data-driven innovation and privacy awareness in the context of smart city initiatives, which have been tested in a real scenario.

cs.SI↗

arXiv2025

Secure Visual Data Processing via Federated Learning

As the demand for privacy in visual data management grows, safeguarding sensitive information has become a critical challenge. This paper addresses the need for privacy-preserving solutions in large-scale visual data processing by leveraging federated learning. Although there have been developments in this field, previous research has mainly focused on integrating object detection with either anonymization or federated learning. However, these pairs often fail to address complex privacy concerns. On the one hand, object detection with anonymization alone can be vulnerable to reverse techniques. On the other hand, federated learning may not provide sufficient privacy guarantees. Therefore, we propose a new approach that combines object detection, federated learning and anonymization. Combining these three components aims to offer a robust privacy protection strategy by addressing different vulnerabilities in visual data. Our solution is evaluated against traditional centralized models, showing that while there is a slight trade-off in accuracy, the privacy benefits are substantial, making it well-suited for privacy sensitive applications.

cs.CV↗

arXiv2024

Empowering Open Data Sharing for Social Good: A Privacy-Aware Approach

The Covid-19 pandemic has affected the world at multiple levels. Data sharing was pivotal for advancing research to understand the underlying causes and implement effective containment strategies. In response, many countries have promoted the availability of daily cases to support research initiatives, fostering collaboration between organisations and making such data available to the public through open data platforms. Despite the several advantages of data sharing, one of the major concerns before releasing health data is its impact on individuals' privacy. Such a sharing process should be based on state-of-the-art methods in Data Protection by Design and by Default. In this paper, we use a data set related to Covid-19 cases in the second largest hospital in Portugal to show how it is feasible to ensure data privacy while improving the quality and maintaining the utility of the data. Our goal is to demonstrate how knowledge exchange in multidisciplinary teams of healthcare practitioners, data privacy, and data science experts is crucial to co-developing strategies that ensure high utility of de-identified data.

cs.DB↗

arXiv2024

Automated Privacy-Preserving Techniques via Meta-Learning

Sharing private data for learning tasks is pivotal for transparent and secure machine learning applications. Many privacy-preserving techniques have been proposed for this task aiming to transform the data while ensuring the privacy of individuals. Some of these techniques have been incorporated into tools, whereas others are accessed through various online platforms. However, such tools require manual configuration, which can be complex and time-consuming. Moreover, they require substantial expertise, potentially restricting their use to those with advanced technical knowledge. In this paper, we propose AUTOPRIV, the first automated privacy-preservation method, that eliminates the need for any manual configuration. AUTOPRIV employs meta-learning to automate the de-identification process, facilitating the secure release of data for machine learning tasks. The main goal is to anticipate the predictive performance and privacy risk of a large set of privacy configurations. We provide a ranked list of the most promising solutions, which are likely to achieve an optimal approximation within a new domain. AUTOPRIV is highly effective as it reduces computational complexity and energy consumption considerably.

cs.LG↗

arXiv2024

Synthetic Data Outliers: Navigating Identity Disclosure

Multiple synthetic data generation models have emerged, among which deep learning models have become the vanguard due to their ability to capture the underlying characteristics of the original data. However, the resemblance of the synthetic to the original data raises important questions on the protection of individuals' privacy. As synthetic data is perceived as a means to fully protect personal information, most current related work disregards the impact of re-identification risk. In particular, limited attention has been given to exploring outliers, despite their privacy relevance. In this work, we analyze the privacy of synthetic data w.r.t the outliers. Our main findings suggest that outliers re-identification via linkage attack is feasible and easily achieved. Furthermore, additional safeguards such as differential privacy can prevent re-identification, albeit at the expense of the data utility.

cs.LG↗

arXiv2024

Differentially-Private Data Synthetisation for Efficient Re-Identification Risk Control

Protecting user data privacy can be achieved via many methods, from statistical transformations to generative models. However, all of them have critical drawbacks. For example, creating a transformed data set using traditional techniques is highly time-consuming. Also, recent deep learning-based solutions require significant computational resources in addition to long training phases, and differentially private-based solutions may undermine data utility. In this paper, we propose $ε$-PrivateSMOTE, a technique designed for safeguarding against re-identification and linkage attacks, particularly addressing cases with a high \sloppy re-identification risk. Our proposal combines synthetic data generation via noise-induced interpolation with differential privacy principles to obfuscate high-risk cases. We demonstrate how $ε$-PrivateSMOTE is capable of achieving competitive results in privacy risk and better predictive performance when compared to multiple traditional and state-of-the-art privacy-preservation methods, including generative adversarial networks, variational autoencoders, and differential privacy baselines. We also show how our method improves time requirements by at least a factor of 9 and is a resource-efficient solution that ensures high performance without specialised hardware.

cs.LG↗

arXiv2023

A Three-Way Knot: Privacy, Fairness, and Predictive Performance Dynamics

As the frontier of machine learning applications moves further into human interaction, multiple concerns arise regarding automated decision-making. Two of the most critical issues are fairness and data privacy. On the one hand, one must guarantee that automated decisions are not biased against certain groups, especially those unprotected or marginalized. On the other hand, one must ensure that the use of personal information fully abides by privacy regulations and that user identities are kept safe. The balance between privacy, fairness, and predictive performance is complex. However, despite their potential societal impact, we still demonstrate a poor understanding of the dynamics between these optimization vectors. In this paper, we study this three-way tension and how the optimization of each vector impacts others, aiming to inform the future development of safe applications. In light of claims that predictive performance and fairness can be jointly optimized, we find this is only possible at the expense of data privacy. Overall, experimental results show that one of the vectors will be penalized regardless of which of the three we optimize. Nonetheless, we find promising avenues for future work in joint optimization solutions, where smaller trade-offs are observed between the three vectors.

cs.LG↗

arXiv2022

Survey on Privacy-Preserving Techniques for Data Publishing

The exponential growth of collected, processed, and shared microdata has given rise to concerns about individuals' privacy. As a result, laws and regulations have emerged to control what organisations do with microdata and how they protect it. Statistical Disclosure Control seeks to reduce the risk of confidential information disclosure by de-identifying them. Such de-identification is guaranteed through privacy-preserving techniques. However, de-identified data usually results in loss of information, with a possible impact on data analysis precision and model predictive performance. The main goal is to protect the individuals' privacy while maintaining the interpretability of the data, i.e. its usefulness. Statistical Disclosure Control is an area that is expanding and needs to be explored since there is still no solution that guarantees optimal privacy and utility. This survey focuses on all steps of the de-identification process. We present existing privacy-preserving techniques used in microdata de-identification, privacy measures suitable for several disclosure types and, information loss and predictive performance measures. In this survey, we discuss the main challenges raised by privacy constraints, describe the main approaches to handle these obstacles, review taxonomies of privacy-preserving techniques, provide a theoretical analysis of existing comparative studies, and raise multiple open issues.

cs.CR↗

arXiv2022

Towards a Data Privacy-Predictive Performance Trade-off

Machine learning is increasingly used in the most diverse applications and domains, whether in healthcare, to predict pathologies, or in the financial sector to detect fraud. One of the linchpins for efficiency and accuracy in machine learning is data utility. However, when it contains personal information, full access may be restricted due to laws and regulations aiming to protect individuals' privacy. Therefore, data owners must ensure that any data shared guarantees such privacy. Removal or transformation of private information (de-identification) are among the most common techniques. Intuitively, one can anticipate that reducing detail or distorting information would result in losses for model predictive performance. However, previous work concerning classification tasks using de-identified data generally demonstrates that predictive performance can be preserved in specific applications. In this paper, we aim to evaluate the existence of a trade-off between data privacy and predictive performance in classification tasks. We leverage a large set of privacy-preserving techniques and learning algorithms to provide an assessment of re-identification ability and the impact of transformed variants on predictive performance. Unlike previous literature, we confirm that the higher the level of privacy (lower re-identification risk), the higher the impact on predictive performance, pointing towards clear evidence of a trade-off.

cs.LG↗