Privacy Risks Research Articles

A Survey on Energy-Efficient Design for Federated Learning over Wireless Networks

Federated learning (FL) has emerged as a decentralized, cutting-edge framework for training models across distributed devices, such as smartphones, IoT devices, and local servers while preserving data privacy and security. FL allows devices to collaboratively learn from shared models without exchanging sensitive data, significantly reducing privacy risks. With these benefits, the deployment of FL over wireless communication systems has gained substantial attention in recent years. However, implementing FL in wireless environments poses significant challenges due to the unpredictable and fluctuating nature of wireless channels. In particular, the limited energy resources of mobile and IoT devices, many of which operate on constrained battery power, make energy management a critical concern. Optimizing energy efficiency is therefore crucial for the successful deployment of FL in wireless networks. However, existing reviews on FL predominantly focus on framework design, wireless communication, and security/privacy concerns, while paying limited attention to the system’s energy consumption. To bridge this gap, this article delves into the foundational principles of FL and highlights energy-efficient strategies tailored for various wireless architectures. It provides a comprehensive overview of FL principles and introduces energy-efficient designs, including resource allocation techniques and communication architectures, tailored to address the unique challenges of wireless communications. Furthermore, we explore emerging technologies aimed at enhancing energy efficiency and discuss future challenges and opportunities for continued research in this field.

Balancing Between Privacy and Utility for Affect Recognition Using Multitask Learning in Differential Privacy-Added Federated Learning Settings: Quantitative Study.

The rise of wearable sensors marks a significant development in the era of affective computing. Their popularity is continuously increasing, and they have the potential to improve our understanding of human stress. A fundamental aspect within this domain is the ability to recognize perceived stress through these unobtrusive devices. This study aims to enhance the performance of emotion recognition using multitask learning (MTL), a technique extensively explored across various machine learning tasks, including affective computing. By leveraging the shared information among related tasks, we seek to augment the accuracy of emotion recognition while confronting the privacy threats inherent in the physiological data captured by these sensors. To address the privacy concerns associated with the sensitive data collected by wearable sensors, we proposed a novel framework that integrates differential privacy and federated learning approaches with MTL. This framework was designed to efficiently identify mental stress while preserving private identity information. Through this approach, we aimed to enhance the performance of emotion recognition tasks while preserving user privacy. Comprehensive evaluations of our framework were conducted using 2 prominent public datasets. The results demonstrate a significant improvement in emotion recognition accuracy, achieving a rate of 90%. Furthermore, our approach effectively mitigates privacy risks, as evidenced by limiting reidentification accuracies to 47%. This study presents a promising approach to advancing emotion recognition capabilities while addressing privacy concerns in the context of empathetic sensors. By integrating MTL with differential privacy and federated learning, we have demonstrated the potential to achieve high levels of accuracy in emotion recognition while ensuring the protection of user privacy. This research contributes to the ongoing efforts to use affective computing in a privacy-aware and ethical manner.

“The value of privacy is not as high as finding my person” — Privacy risks as abstractions

Online dating has emerged as a key platform for digital engagement among young people, serving as a common medium for building relationships. As multimedia platforms, dating apps encourage users to translate their notions of identity construction, self-expression, and communication into engagement and participation in the digital world. Dating apps are one of several technologies that are changing existing norms of relationship building, self-presentation, and ideals of romance. Dating apps foster a unique form of engagement because to facilitate a date users must reveal sensitive details about their demographics, location, values, and lifestyle that they might not otherwise display on social media. On dating apps privacy concerns are unusually self-conscious. This unique digital environment constitutes a strategic research site where research can explore how privacy concerns influence the process of personal information (PI) disclosure. Findings reveal that users perceive other users as threats to their privacy, which makes them carefully consider what they reveal. In contrast, risks associated with app providers, such as selling data to advertisers, are viewed as abstract, and not behavior-altering. This discrepancy in risk is attributed to inadequate data literacy and agency over personal data. This paper concludes with suggestions for future research and policy that would improve the quality of youth engagement.

Target informed client recruitment for efficient federated learning in healthcare

BackgroundModern machine learning and deep learning methods have been widely incorporated in decision making processes in healthcare in the form of decision support mechanisms. In healthcare, data are abundant but typically not centrally available and, therefore, require some form of aggregation to facilitate training procedures. Aggregating sensitive data poses a significant privacy risk, which is why, both in Europe and the United States, legal frameworks regulate the treatment of such data. Whilst these measures protect the individual behind the data, they pose a significant challenge that results in extensive legal administration related to data sharing efforts. Federated learning (FL) offers a way to mitigate these challenges by allowing to learn models in distributed fashion, eliminating the need to aggregate data for the purpose of training. However, FL comes with a new set of challenges related to communication overhead, client selection and efficiency of the FL training procedure, among others.MethodsIn this work, we extend on a previously proposed client recruitment approach by incorporating knowledge on the local hardware such that it becomes possible to recruit a subset of clients for the federation based on the construct of client-level representativeness, which is expressed in terms of the local target distribution divergence, sample size, and the underlying hardware.ResultsWe show that, for prominent, medical regression and classification tasks, the recruitment approach yields results that are on par, or better, compared to the central and federated approaches. The proposed approach requires a mere fraction of the data for training and reduces the training time by a factor of 3-4. In addition, we show that excluded clients can still significantly benefit from the resulting federated model through local fine-tuning.ConclusionsBy expressing the representativeness of clients in function of the deviation in the local target distribution, the sample size and efficiency of the underlying hardware, we are able to define a recruitment approach that yields a subset of clients for the federation resulting in significantly reduced training time, without harming predictive performance, whilst improving the privacy preserving characteristics compared to the standard FL and central approaches.

Balancing conflicting interests in ‘sharenting’: the corporate social responsibility of social media platforms

ABSTRACT Parents frequently share personal information about their children on social media, a practice referred to as ‘sharenting’. While this may not result in immediate harm, it presents significant privacy risks that could have long-term implications for children. Given that parents have the right to share content, and sharenting has become a normalised aspect of daily life, completely prohibiting the practice is neither realistic nor practical. A more balanced approach can be found in the ‘right to be forgotten’ principle, which enables parents to continue sharing while granting children the right to request the removal of content that may compromise their privacy. This paper critically examines the content removal processes employed by social media platforms, with a particular focus on Facebook, in managing requests related to sharenting. Drawing from corporate social responsibility frameworks, this paper argues that social media platforms should take a more proactive role in addressing the privacy risks of sharenting. In a similar vein, the paper proposes specific actions platforms could implement to effectively mitigate and prevent these risks.

Problematising Sharenting of Supermoms in a Neoliberal Context

The representation of motherhood using digital platforms, which we call sharenting, can shape the socially constructed motherhood identity, i.e., supermoms in modern societies. Despite building a positive identity as mothers, supermoms may unintentionally violate their children’s privacy, instrumentalise, and even monetise the children. We examine sharenting and supermom identities as manifestations of neoliberal repositioning women as mothers and employees with little consideration of their unarticulated struggles. Drawing on a netnographic study of social media practices, we analysed fifteen working supermom accounts with over two hundred thousand followers, where child-related content was shared and posted between 2019 and 2023. We find that sharenting is often employed to construct the identity of supermoms, yet it frequently leads to the exploitation of children for profit-driven purposes, thereby increasing their vulnerability. We contribute to the literature by offering empirical evidence that the supermom identity is a by-product of the neoliberal system, which normalises time-intensive demands on women in the workplace and caregiving roles. Furthermore, we highlight the adverse consequences of sharenting and provide critical insights into the privacy and security risks associated with child-related digital content.

Backdoor Attack Against Dataset Distillation in Natural Language Processing

Dataset distillation has become an important technique for enhancing the efficiency of data when training machine learning models. It finds extensive applications across various fields, including computer vision (CV) and natural language processing (NLP). However, it essentially consists of a deep neural network (DNN) model, which remain susceptible to security and privacy vulnerabilities (e.g., backdoor attacks). Existing studies have primarily focused on optimizing the balance between computational efficiency and model performance, overlooking the accompanying security and privacy risks. This study presents the first backdoor attack targeting NLP models trained on distilled datasets. We introduce malicious triggers into synthetic data during the distillation phase to execute a backdoor attack on downstream models trained with these data. We employ several widely used datasets to assess how different architectures and dataset distillation techniques withstand our attack. The experimental findings reveal that the attack achieves strong performance with a high (above 0.9 and up to 1.0) attack success rate (ASR) in most cases. For backdoor attacks, high attack performance often comes at the cost of reduced model utility. Our attack maintains high ASR while maximizing the preservation of downstream model utility, as evidenced by results showing that the clean test accuracy (CTA) of the backdoored model is very close to that of the clean model. Additionally, we performed comprehensive ablation studies to identify key factors affecting attack performance. We tested our attack method against five defense strategies, including NAD, Neural Cleanse, ONION, SCPD, and RAP. The experimental results show that these defense methods are unable to reduce the attack success rate without compromising the model’s performance on normal tasks. Therefore, these methods cannot effectively defend against our attack.

Privacy-Preserving Sequential Recommendation with Collaborative Confusion

Sequential recommendation has attracted a lot of attention from both academia and industry, however the privacy risks associated with gathering and transferring users’ personal interaction data are often underestimated or ignored. Existing privacy-preserving studies are mainly applied to traditional collaborative filtering or matrix factorization rather than sequential recommendation. Moreover, these studies are mostly based on differential privacy or federated learning, which often lead to significant performance degradation, or have high requirements for communication. In this work, we address privacy-preserving from a different perspective. Unlike existing research, we capture collaborative signals of neighbor interaction sequences and directly inject indistinguishable items into the target sequence before the recommendation process begins, thereby increasing the perplexity of the target sequence. Even if the target interaction sequence is obtained by attackers, it is difficult to discern which ones are the actual user interaction records. To achieve this goal, we introduce a novel sequential recommender system called CLOUD ( C o L laborative-c O nfusion seq U ential recommen D er), which incorporates a collaborative confusion mechanism to modify the raw interaction sequences before conducting recommendation. Specifically, CLOUD first calculates the similarity between the target interaction sequence and other neighbor sequences to find similar sequences. Then, CLOUD considers the shared representation of the target sequence and similar sequences to determine the operation to be performed: keep, delete, or insert. A copy mechanism is designed to make items from similar sequences have a higher probability to be inserted into the target sequence. Finally, the modified sequence is used to train the recommender and predict the next item. We conduct extensive experiments on three benchmark datasets. The experimental results show that CLOUD achieves a maximum modification rate of 66.57% on interaction sequences, and obtains over 99% recommendation accuracy compared to the state-of-the-art sequential recommendation methods. This proves that CLOUD can effectively protect user privacy at minimal recommendation performance cost, which provides a new solution for privacy-preserving for sequential recommendation. Our implementation is available at https://github.com/weiwang0927/CLOUD .

Honor the Contract? Effects of Algorithmic Recommendation System Features on Perceived Benefits, Privacy Risk, and Continuance Intention to Use TikTok

Algorithmic recommendations, characterized by features including accuracy, familiarity, novelty, and transparency, have been used in TikTok for a long time. However, the impact of these features on users’ privacy calculus and continuance usage intention remains unclear. Survey data from 625 Chinese users was analyzed using the least-squares partial structural equation modelling. Results showed that four features positively affect users’ perceived benefits; perceived familiarity positively affects privacy risk; perceived effectiveness of privacy policy moderates the relationship between perceived accuracy and privacy risk. Moreover, perceived benefits positively influence continuance intention to use, whereas privacy risk has a negative impact. This study shed light on why users are willing to compromise privacy for the benefits originating from features of algorithmic recommendations. Furthermore, this study provided valuable insights for platforms and practitioners concerning the design and development of effective algorithmic recommendations, as well as improving the effectiveness of privacy policy.

Generative AI model privacy: a survey

The rapid progress of generative AI models has yielded substantial breakthroughs in AI, facilitating the generation of realistic synthetic data across various modalities. However, these advancements also introduce significant privacy risks, as the models may inadvertently expose sensitive information from their training data. Currently, there is no comprehensive survey work investigating privacy issues, e.g., attacking and defending privacy in generative AI models. We strive to identify existing attack techniques and mitigation strategies and to offer a summary of the current research landscape. Our survey encompasses a wide array of generative AI models, including language models, Generative Adversarial Networks, diffusion models, and their multi-modal counterparts. It indicates the critical need for continued research and development in privacy-preserving techniques for generative AI models. Furthermore, we offer insights into the challenges and discuss the open problems in the intersection of privacy and generative AI models.

Generative AI and future education: a review, theoretical validation, and authors' perspective on challenges and solutions.

Generative AI (Gen AI), exemplified by ChatGPT, has witnessed a remarkable surge in popularity recently. This cutting-edge technology demonstrates an exceptional ability to produce human-like responses and engage in natural language conversations guided by context-appropriate prompts. However, its integration into education has become a subject of ongoing debate. This review examines the challenges of using Gen AI like ChatGPT in education and offers effective strategies. To retrieve relevant literature, a search of reputable databases was conducted, resulting in the inclusion of twenty-two publications. Using Atlas.ti, the analysis reflected six primary challenges with plagiarism as the most prevalent issue, closely followed by responsibility and accountability challenges. Concerns were also raised about privacy, data protection, safety, and security risks, as well as discrimination and bias. Additionally, there were challenges about the loss of soft skills and the risks of the digital divide. To address these challenges, a number of strategies were identified and subjected to critical evaluation to assess their practicality. Most of them were practical and align with the ethical and pedagogical theories. Within the prevalent concepts, "ChatGPT" emerged as the most frequent one, followed by "AI," "student," "research," and "education," highlighting a growing trend in educational discourse. Moreover, close collaboration was evident among the leading countries, all forming a single cluster, led by the United States. This comprehensive review provides implications, recommendations, and future prospects concerning the use of generative AI in education.

Continual Learning From a Stream of APIs.

Continual learning (CL) aims to learn new tasks without forgetting previous tasks. However, existing CL methods require a large amount of raw data, which is often unavailable due to copyright considerations and privacy risks. Instead, stakeholders usually release pre-trained machine learning models as a service (MLaaS), which users can access via APIs. This paper considers two practical-yet-novel CL settings: data-efficient CL (DECL-APIs) and data-free CL (DFCL-APIs), which achieve CL from a stream of APIs with partial or no raw data. Performing CL under these two new settings faces several challenges: unavailable full raw data, unknown model parameters, heterogeneous models of arbitrary architecture and scale, and catastrophic forgetting of previous APIs. To overcome these issues, we propose a novel data-free cooperative continual distillation learning framework that distills knowledge from a stream of APIs into a CL model by generating pseudo data, just by querying APIs. Specifically, our framework includes two cooperative generators and one CL model, forming their training as an adversarial game. We first use the CL model and the current API as fixed discriminators to train generators via a derivative-free method. Generators adversarially generate hard and diverse synthetic data to maximize the response gap between the CL model and the API. Next, we train the CL model by minimizing the gap between the responses of the CL model and the black-box API on synthetic data, to transfer the API's knowledge to the CL model. Furthermore, we propose a new regularization term based on network similarity to prevent catastrophic forgetting of previous APIs. Our method performs comparably to classic CL with full raw data on the MNIST and SVHN datasets in the DFCL-APIs setting. In the DECL-APIs setting, our method achieves 0.97×, 0.75× and 0.69× performance of classic CL on the more challenging CIFAR10, CIFAR100, and MiniImageNet, respectively.

Actionability of Synthetic Data in a Heterogeneous and Rare Health Care Demographic: Adolescents and Young Adults With Cancer.

Research on rare diseases and atypical health care demographics is often slowed by high interparticipant heterogeneity and overall scarcity of data. Synthetic data (SD) have been proposed as means for data sharing, enlargement, and diversification, by artificially generating real phenomena while obscuring the real patient data. The utility of SD is actively scrutinized in health care research, but the role of sample size for actionability of SD is insufficiently explored. We aim to understand the interplay of actionability and sample size by generating SD sets of varying sizes from gradually diminishing amounts of real individuals' data. We evaluate the actionability of SD in a highly heterogeneous and rare demographic: adolescents and young adults (AYAs) with cancer. A population-based cross-sectional cohort study of 3,735 AYAs was subsampled at random to produce 13 training data sets of varying sample sizes. We studied four distinct generator architectures built on the open-source Synthetic Data Vault library. Each architecture was used to generate SD of varying sizes on the basis of each aforementioned training subsets. SD actionability was assessed by comparing the resulting SD with their respective real data against three metrics-veracity, utility, and privacy concealment. All examined generator architectures yielded actionable data when generating SD with sizes similar to the real data. Large SD sample size increased veracity but generally increased privacy risks. Using fewer training participants led to faster convergence in veracity, but partially exacerbated privacy concealment issues. SD is a potentially promising option for data sharing and data augmentation, yet sample size plays a significant role in its actionability. SD generation should go hand-in-hand with consistent scrutiny, and sample size should be carefully considered in this process.

Cascade Vertical Federated Learning Towards Straggler Mitigation and Label Privacy Over Distributed Labels

Vertical federated learning (VFL) enables collaborative machine learning on vertically partitioned data with privacy-preservation. Most VFL methods face three daunting challenges in real-world applications. First, most existing VFL methods assume that at least one party holds the complete set of labels of all data samples. However, this assumption often violates the nature of many practical scenarios, where the parties only have partial labels. Second, the heterogeneity and dynamic of computational and communication resources in participated parties may cause the straggler problem and slow down training convergence. Third, the confidential label information could be exposed through malicious parties during VFL. To address these challenges, we propose a novel VFL algorithm named Cascade Vertical Federated Learning (CVFL), in which partitioned labels can be fully utilized to train neural networks with privacy-preservation. To mitigate the straggler problem, we design a novel optimization objective to increase straggler's contribution to the trained models. To mitigate the label privacy risks, we design a novel defense approach to protect the label privacy of CVFL. We conduct comprehensive experiments and the results demonstrate the effectiveness and efficiency of CVFL. Further, the proposed defense approach can achieve a better tradeoff between label privacy and model utility than two widely-used defense approaches.

Re-identification of anonymised MRI head images with publicly available software: investigation of the current risk to patient privacy

Facial recognition software (FRS) has historically been perceived as lacking the capability to identify individuals from cross-sectional medical images. Utilising such data for identification purposes was considered infeasible due to the substantial computational power and specialised technical expertise it would require. However, recent advancements in accessible artificial intelligence-based (AI-based) software and open-source tools have made these applications widely available and easy to use, raising new privacy concerns. This proof-of-concept was designed as a cross-sectional study and included participants with a verified online presence. Standard magnetic resonance imaging (MRI) head scans were performed on these participants, from which three-dimensional rendering (3DR) images were created using free and publicly available software. These images were used for face searches by free and publicly available FRS. Different head orientations and hairstyles were applied to the 3DR images to assess whether non-facial features influenced the FRS results. All results were obtained between the 10th of February 2024 and the 1st of March2024. Face searches of 3DR images in a database containing over 800 million images from the World Wide Web (WWW) yielded correct matches for 50% of the participants in less than 10min. The user-friendly software required minimal computational knowledge or resources, making this process broadly accessible. Modifying elements such as hairstyles or the orientation of the 3DR to better resemble actual photographs of the participants improved FRS matches. Current existing FRS can swiftly and accurately identify individuals from MRI head scans. This poses a significant privacy risk for participants in enrolled clinical trials and highlights the urgent need for improved data protection measures and increased sensitivity to ensure participant confidentiality. There was no funding source for this study.

Invisible boundaries: Balancing Image metadata privacy with forensic Imperatives

Metadata provides a wealth of information that is invaluable for digital forensic investigations, but it also poses significant privacy risks. The varying degrees of metadata retention across social media platforms have significant implications for digital forensic investigations. Metadata plays a crucial role in constructing timelines, verifying data authenticity, and providing evidence of interactions. In this research, we investigate the handling of EXIF metadata by various social media platforms and assess the implications for user privacy and digital forensic investigations. Our findings underscore the critical need for specialized software to manage and remove metadata from images uploaded to social media, preserving user privacy while retaining necessary data for legal and forensic purposes. To address these concerns, we implement a framework for encrypting and decrypting metadata, ensuring that sensitive information can be protected before images are uploaded to social media platforms. This research aims to encapsulate the findings and emphasize the need for balanced solutions that respect user privacy while supporting forensic investigations.

Assessing the role of artificial intelligence in enhancing customer personalization: A study of ethical and privacy implications in digital marketing

Technological evolution, especially in artificial intelligence (AI), has significantly changed digital marketing, specifically customer-targeted experience. Machine learning, recommendation systems, and predictive analytics have made it possible for enterprises to personalize content, products, and services, among other things, on a very large scale. Nevertheless, all these advances have come with large ethical and privacy concerns that are yet to be met. This paper discusses how AI can enhance customers' personalization to analyze the moral and privacy considerations involved. The research examines how machine learning marketing strategies work in the digital environment and real-life examples of organizations that successfully implement these approaches and potential issues. It highlights how individuals might buy more products and make businesses more profitable with customized advertising techniques. However, the paper also describes the moral problems of AI that use personal data without permission, artificially intelligent algorithms that affect decision-making, and artificial intelligence decision-making. At the same time, its work has yet to be well known. The paper looks into privacy risks like data breach incidents, privacy violations, GDPR, and CCPA. This paper adopts an exploratory case study research design as it gathers academic papers, industry examples, and expert opinions to present a balanced view of the effects of AI. Based on the research, conclusions can be drawn that great opportunities AI opens for personalization should be met with safety considerations and working ethical guidelines. In the end, the paper presents guidelines that should help businesses integrate trust and compliance into creating and using AI technologies.

Network alignment in multiplex social networks using the information diffusion dynamics

Modern social networking applications (SNAs) typically rely on independent data management systems, leading to fragmented user identities and the creation of data silos. This fragmentation impedes the development of unified user profiles and limits the effectiveness of cross-platform behavioural analysis and personalized recommendations. As the demand for integrated services grows, the need for a unified user identity across platforms becomes increasingly critical. However, existing identity integration methods face significant challenges, including data isolation, privacy risks, and a lack of universal standards. To address these issues, we propose a framework that utilizes node dynamics time-series data for network alignment in multiplex social networks. Our approach employs the UIU diffusion model to simulate information diffusion dynamics across multiple platforms and uses the Expectation-Maximization (EM) algorithm for network alignment. Crucially, our method relies solely on publicly available user information from different platforms, avoiding the need for access to private user data, thereby enhancing security. Experimental results demonstrate the model’s efficacy, achieving F1 scores of 100% for interlayer links and over 90% for intralayer links on real-world datasets. These findings highlight the model’s potential for applications in social influence analysis, community detection, and recommendation systems.

AI Ethics and Transparency in Operations Management: How Governance Mechanisms Can Reduce Data Bias and Privacy Risks

AI Ethics and Transparency in Operations Management: How Governance Mechanisms Can Reduce Data Bias and Privacy Risks

Distributed Malicious Traffic Detection

With the wide deployment of edge devices, distributed network traffic data are rapidly increasing. Traditional detection methods for malicious traffic rely on centralized training, in which a single server is often used to aggregate private traffic data from edge devices, so as to extract and identify features. However, these methods face difficult data collection, heavy computational complexity, and high privacy risks. To address these issues, this paper proposes a federated learning-based distributed malicious traffic detection framework, FL-CNN-Traffic. In this framework, edge devices utilize a convolutional neural network (CNN) to process local detection, data collection, feature extraction, and training. A server aggregates model updates from edge devices using four federated learning algorithms (FedAvg, FedProx, Scaffold, and FedNova) to build a global model. This framework allows multiple devices to collaboratively train a model without sharing private traffic data, addressing the “Data Silo” problem while ensuring privacy. Evaluations on the USTC-TFC2016 dataset show that for independent and identically distributed (IID) data, this framework can reach or exceed the performance of centralized deep learning methods. For Non-IID data, this framework outperforms other neural networks based on federated learning, with accuracy improvements ranging from 2.59% to 4.73%.