Publications
159 results found
Minto L, Haller M, Haddadi H, et al., 2021, Stronger privacy for federated collaborative filtering with implicit feedback, 15th ACM Conference on Recommender Systems, Publisher: ACM, Pages: 342-350
Recommender systems are commonly trained on centrally collected userinteraction data like views or clicks. This practice however raises seriousprivacy concerns regarding the recommender's collection and handling ofpotentially sensitive data. Several privacy-aware recommender systems have beenproposed in recent literature, but comparatively little attention has beengiven to systems at the intersection of implicit feedback and privacy. Toaddress this shortcoming, we propose a practical federated recommender systemfor implicit data under user-level local differential privacy (LDP). Theprivacy-utility trade-off is controlled by parameters $\epsilon$ and $k$,regulating the per-update privacy budget and the number of $\epsilon$-LDPgradient updates sent by each user respectively. To further protect the user'sprivacy, we introduce a proxy network to reduce the fingerprinting surface byanonymizing and shuffling the reports before forwarding them to therecommender. We empirically demonstrate the effectiveness of our framework onthe MovieLens dataset, achieving up to Hit Ratio with K=10 (HR@10) 0.68 on 50kusers with 5k items. Even on the full dataset, we show that it is possible toachieve reasonable utility with HR@10>0.5 without compromising user privacy.
Malekzadeh M, Clegg R, Cavallaro A, et al., 2021, DANA: Dimension-Adaptive Neural Architecture for Multivariate Sensor Data, ACM Journal on Interactive, Mobile, Wearable and Ubiquitous Technologies (IMWUT)
Saidi SJ, Mandalari AM, Haddadi H, et al., 2021, Detecting consumer IoT devices through the lens of an ISP, Pages: 36-38
Internet of Things (IoT) devices are becoming increasingly popular and offer a wide range of services and functionality to their users. However, there are significant privacy and security risks associated with these devices. IoT devices can infringe users' privacy by ex-filtrating their private information to third parties, often without their knowledge. In this work we investigate the possibility to identify IoT devices and their location in an Internet Service Provider's network. By analyzing data from a large Internet Service Provider (ISP), we show that it is possible to recognize specific IoT devices, their vendors, and sometimes even their specific model, and to infer their location in the network. This is possible even with sparsely sampled flow data that are often the only datasets readily available at an ISP. We evaluate our proposed methodology [1] to infer IoT devices at subscriber lines of a large ISP. Given ground truth information on IoT devices location and models, we were able to detect more than 77% of the studied IoT devices from sampled flow data in the wild.
Mo F, Haddadi H, Katevas K, et al., 2021, PPFL: privacy-preserving federated learning with trusted execution environments, Mobile Systems, Applications, and Services conference, Publisher: ACM, Pages: 94-108
We propose and implement a Privacy-preserving Federated Learning (PPFL)framework for mobile systems to limit privacy leakages in federated learning.Leveraging the widespread presence of Trusted Execution Environments (TEEs) inhigh-end and mobile devices, we utilize TEEs on clients for local training, andon servers for secure aggregation, so that model/gradient updates are hiddenfrom adversaries. Challenged by the limited memory size of current TEEs, weleverage greedy layer-wise training to train each model's layer inside thetrusted area until its convergence. The performance evaluation of ourimplementation shows that PPFL can significantly improve privacy whileincurring small system overheads at the client-side. In particular, PPFL cansuccessfully defend the trained model against data reconstruction, propertyinference, and membership inference attacks. Furthermore, it can achievecomparable model utility with fewer communication rounds (0.54x) and a similaramount of network traffic (1.002x) compared to the standard federated learningof a complete model. This is achieved while only introducing up to ~15% CPUtime, ~18% memory usage, and ~21% energy consumption overhead in PPFL'sclient-side.
Mandalari AM, Dubois DJ, Kolcun R, et al., 2021, Blocking without breaking: identification and mitigation ofnon-essential IoT traffic, Publisher: arXiv
Despite the prevalence of Internet of Things (IoT) devices, there is littleinformation about the purpose and risks of the Internet traffic these devicesgenerate, and consumers have limited options for controlling those risks. A keyopen question is whether one can mitigate these risks by automatically blockingsome of the Internet connections from IoT devices, without rendering thedevices inoperable. In this paper, we address this question by developing arigorous methodology that relies on automated IoT-device experimentation toreveal which network connections (and the information they expose) areessential, and which are not. We further develop strategies to automaticallyclassify network traffic destinations as either required (i.e., their trafficis essential for devices to work properly) or not, hence allowing firewallrules to block traffic sent to non-required destinations without breaking thefunctionality of the device. We find that indeed 16 among the 31 devices wetested have at least one blockable non-required destination, with the maximumnumber of blockable destinations for a device being 11. We further analyze thedestination of network traffic and find that all third parties observed in ourexperiments are blockable, while first and support parties are neitheruniformly required or non-required. Finally, we demonstrate the limitations ofexisting blocklists on IoT traffic, propose a set of guidelines forautomatically limiting non-essential IoT traffic, and we develop a prototypesystem that implements these guidelines.
Aloufi R, Haddadi H, Boyle D, 2021, Configurable privacy-preserving automatic speech recognition, Publisher: arXiv
Voice assistive technologies have given rise to far-reaching privacy andsecurity concerns. In this paper we investigate whether modular automaticspeech recognition (ASR) can improve privacy in voice assistive systems bycombining independently trained separation, recognition, and discretizationmodules to design configurable privacy-preserving ASR systems. We evaluateprivacy concerns and the effects of applying various state-of-the-arttechniques at each stage of the system, and report results using task-specificmetrics (i.e. WER, ABX, and accuracy). We show that overlapping speech inputsto ASR systems present further privacy concerns, and how these may be mitigatedusing speech separation and optimization techniques. Our discretization moduleis shown to minimize paralinguistics privacy leakage from ASR acoustic modelsto levels commensurate with random guessing. We show that voice privacy can beconfigurable, and argue this presents new opportunities for privacy-preservingapplications incorporating ASR.
Crabtree A, Haddadi H, Mortier R, 2021, Privacy by design for the Internet of Things, Privacy by Design for the Internet of Things: Building Accountability and Security, Pages: 1-18, ISBN: 9781839531408
- Cite
- Citations: 2
Zhan Y, Haddadi H, 2021, MoSen: activity modelling in multiple-occupancy smart homes, Publisher: arXiv
Smart home solutions increasingly rely on a variety of sensors for behavioralanalytics and activity recognition to provide context-aware applications andpersonalized care. Optimizing the sensor network is one of the most importantapproaches to ensure classification accuracy and the system's efficiency.However, the trade-off between the cost and performance is often a challenge inreal deployments, particularly for multiple-occupancy smart homes or carehomes. In this paper, using real indoor activity and mobility traces, floor plans,and synthetic multi-occupancy behavior models, we evaluate severalmulti-occupancy household scenarios with 2-5 residents. We explore and quantifythe trade-offs between the cost of sensor deployments and expected labelingaccuracy in different scenarios. Our evaluation across different scenarios showthat the performance of the desired context-aware task is affected by differentlocalization resolutions, the number of residents, the number of sensors, andvarying sensor deployments. To aid in accelerating the adoption of practicalsensor-based activity recognition technology, we design MoSen, a framework tosimulate the interaction dynamics between sensor-based environments andmultiple residents. By evaluating the factors that affect the performance ofthe desired sensor network, we provide a sensor selection strategy and designmetrics for sensor layout in real environments. Using our selection strategy ina 5-person scenario case study, we demonstrate that MoSen can significantlyimprove overall system performance without increasing the deployment costs.
Crabtree A, Lodge T, McAuley D, et al., 2021, Building accountability into the Internet of Things, Privacy by Design for the Internet of Things: Building Accountability and Security, Pages: 127-152, ISBN: 9781839531408
Crabtree A, Haddadi H, Mortier R, 2021, Privacy by Design for the Internet of Things: Building accountability and security, ISBN: 9781839531408
Privacy by design is a proactive approach that promotes privacy and data protection compliance throughout project lifecycles when storing or accessing personal data. Privacy by design is essential for the Internet of Things (IoT) as privacy concerns and accountability are being raised in an increasingly connected world. What becomes of data generated, collected or processed by the IoT is clearly an important question for all involved in the development, manufacturing, applications and use of related technologies. But this IoT concept does not work well with the ‘big data’ trend of aggregating pools of data for new applications. Developers need to address privacy and security issues and legislative requirements at the design stage, and not as an afterthought. In this edited book, the authors draw on a wealth of interdisciplinary research to delineate the challenges of building accountability into the Internet of Things and solutions for delivering on this critical societal challenge. This advanced book brings together legal-tech scholars, computer scientists, human computer interaction researchers and designers, and social scientists to address these challenges and elaborate solutions. It articulates the accountability principle in law and how it impacts IoT development, presents empirical studies of accountability in action and its implications for IoT development, brings technological responses to the requirements of GDPR and ways of building accountability into the IoT, and covers compliant IoT application development, privacy-preserving data analytics, human-centred IoT security, human-data interaction, and the methodological challenge of understanding and responding to the adoption of future technologies in everyday life.
Mortier R, Haddadi H, Servia S, et al., 2021, Distributed data analytics, Privacy by Design for the Internet of Things: Building Accountability and Security, Pages: 181-210, ISBN: 9781839531408
- Cite
- Citations: 1
Pooranian Z, Conti M, Haddadi H, et al., 2021, Online Advertising Security: Issues, Taxonomy, and Future Directions, IEEE COMMUNICATIONS SURVEYS AND TUTORIALS, Vol: 23, Pages: 2494-2524
- Author Web Link
- Cite
- Citations: 3
Zhan Y, Haddadi H, 2021, MoSen: Sensor Network Optimization in Multiple-Occupancy Smart Homes, 19th IEEE International Conference on Pervasive Computing and Communications (IEEE PerCom), Publisher: IEEE, Pages: 384-388
Kolcun R, Popescu DA, Safronov V, et al., 2020, The case for retraining of ML models for IoT device identification at the edge, Publisher: arXiv
Internet-of-Things (IoT) devices are known to be the source of many securityproblems, and as such they would greatly benefit from automated management.This requires robustly identifying devices so that appropriate network securitypolicies can be applied. We address this challenge by exploring how toaccurately identify IoT devices based on their network behavior, usingresources available at the edge of the network. In this paper, we compare the accuracy of five different machine learningmodels (tree-based and neural network-based) for identifying IoT devices byusing packet trace data from a large IoT test-bed, showing that all models needto be updated over time to avoid significant degradation in accuracy. In orderto effectively update the models, we find that it is necessary to use datagathered from the deployment environment, e.g., the household. We thereforeevaluate our approach using hardware resources and data sources representativeof those that would be available at the edge of the network, such as in an IoTdeployment. We show that updating neural network-based models at the edge isfeasible, as they require low computational and memory resources and theirstructure is amenable to being updated. Our results show that it is possible toachieve device identification and categorization with over 80% and 90% accuracyrespectively at the edge.
Maali E, Boyle D, Haddadi H, 2020, Towards identifying IoT traffic anomalies on the home gateway: Poster abstract, Pages: 735-736
The number of IoT devices continues to grow despite the alarming rate of identification of security and privacy issues. There is widespread concern that development of IoT devices is performed without sufficient attention paid to security and privacy issues. Consequently, networks have a higher probability of incorporating vulnerable IoT devices that may be easy to compromise to launch cyber attacks. Inclusion of IoT devices paves the way for a new category of anomalies to be introduced to networks. Traditional anomaly detection techniques (e.g., semi-supervised and signature-based methods), however, are likely inefficient in detecting IoT-based anomalies. This is because these techniques require static signatures of known attacks, specialized hardware, or full packet inspection. They are also expensive, and may be inaccurate or unscalable. Vulnerable IoT devices can be used to perform destructive attacks or invade privacy. The ability to find anomalies in IoT traffic has the potential to assist with early detection and deployment of countermeasures to thwart such attacks. Thus, new techniques for detecting infected IoT devices are needed to mitigate the associated security and privacy risks. In this research, we investigate the possibility to identify IoT traffic using a combination of behavioural profile, predefined blocklist and device fingerprint. Such a system may be able to detect anomalous and/or malicious devices and/or traffic reliably and quickly. Initial results show that for our implementation of such a system, IoT traffic can be identified using device behaviour profile, fingerprint, and contacted destinations. This work takes the first step towards designing and evaluating iDetector, a framework that can detect anomalous behaviour within IoT networks. In our experiments, iDetector was able to correctly identify 80 - 90% of all captured traffic traversing a home gateway.
Aloufi R, Haddadi H, Boyle D, 2020, Privacy-preserving Voice Analysis via Disentangled Representations, CCSW 2020 - Proceedings of the 2020 ACM SIGSAC Conference on Cloud Computing Security Workshop, Pages: 1-14
Voice User Interfaces (VUIs) are increasingly popular and built into smartphones, home assistants, and Internet of Things (IoT) devices. Despite offering an always-on convenient user experience, VUIs raise new security and privacy concerns for their users. In this paper, we focus on attribute inference attacks in the speech domain, demonstrating the potential for an attacker to accurately infer a target user's sensitive and private attributes (e.g. their emotion, sex, or health status) from deep acoustic models. To defend against this class of attacks, we design, implement, and evaluate a user-configurable, privacy-aware framework for optimizing speech-related data sharing mechanisms. Our objective is to enable primary tasks such as speech recognition and user identification, while removing sensitive attributes in the raw speech data before sharing it with a cloud service provider. We leverage disentangled representation learning to explicitly learn independent factors in the raw data. Based on a user's preferences, a supervision signal informs the filtering out of invariant factors while retaining the factors reflected in the selected preference. Our experimental evaluation over five datasets shows that the proposed framework can effectively defend against attribute inference attacks by reducing their success rates to approximately that of guessing at random, while maintaining accuracy in excess of 99% for the tasks of interest. We conclude that negotiable privacy settings enabled by disentangled representations can bring new opportunities for privacy-preserving applications.
Zhao Y, Liu H, Li H, et al., 2020, Semi-supervised Federated Learning for Activity Recognition
Training deep learning models on in-home IoT sensory data is commonly used torecognise human activities. Recently, federated learning systems that use edgedevices as clients to support local human activity recognition have emerged asa new paradigm to combine local (individual-level) and global (group-level)models. This approach provides better scalability and generalisability and alsooffers better privacy compared with the traditional centralised analysis andlearning models. The assumption behind federated learning, however, relies onsupervised learning on clients. This requires a large volume of labelled data,which is difficult to collect in uncontrolled IoT environments such as remotein-home monitoring. In this paper, we propose an activity recognition system that usessemi-supervised federated learning, wherein clients conduct unsupervisedlearning on autoencoders with unlabelled local data to learn generalrepresentations, and a cloud server conducts supervised learning on an activityclassifier with labelled data. Our experimental results show that using a longshort-term memory autoencoder and a Softmax classifier, the accuracy of ourproposed system is higher than that of both centralised systems andsemi-supervised federated learning using data augmentation. The accuracy isalso comparable to that of supervised federated learning systems. Meanwhile, wedemonstrate that our system can reduce the number of needed labels and the sizeof local models, and has faster local activity recognition speed thansupervised federated learning does.
Saidi SJ, Mandalari AM, Kolcun R, et al., 2020, A Haystack Full of Needles: Scalable Detection of IoT Devices in the Wild, Pages: 87-100
Consumer Internet of Things (IoT) devices are extremely popular, providing users with rich and diverse functionalities, from voice assistants to home appliances. These functionalities often come with significant privacy and security risks, with notable recent large-scale coordinated global attacks disrupting large service providers. Thus, an important first step to address these risks is to know what IoT devices are where in a network. While some limited solutions exist, a key question is whether device discovery can be done by Internet service providers that only see sampled flow statistics. In particular, it is challenging for an ISP to efficiently and effectively track and trace activity from IoT devices deployed by its millions of subscribers - -all with sampled network data. In this paper, we develop and evaluate a scalable methodology to accurately detect and monitor IoT devices at subscriber lines with limited, highly sampled data in-the-wild. Our findings indicate that millions of IoT devices are detectable and identifiable within hours, both at a major ISP as well as an IXP, using passive, sparsely sampled network flow headers. Our methodology is able to detect devices from more than 77% of the studied IoT manufacturers, including popular devices such as smart speakers. While our methodology is effective for providing network analytics, it also highlights significant privacy consequences.
Saidi SJ, Mandalari AM, Kolcun R, et al., 2020, A haystack full of needles: scalable detection of IoT devices in the wild, Publisher: arXiv
Consumer Internet of Things (IoT) devices are extremely popular, providingusers with rich and diverse functionalities, from voice assistants to homeappliances. These functionalities often come with significant privacy andsecurity risks, with notable recent large scale coordinated global attacksdisrupting large service providers. Thus, an important first step to addressthese risks is to know what IoT devices are where in a network. While somelimited solutions exist, a key question is whether device discovery can be doneby Internet service providers that only see sampled flow statistics. Inparticular, it is challenging for an ISP to efficiently and effectively trackand trace activity from IoT devices deployed by its millions of subscribers--all with sampled network data. In this paper, we develop and evaluate a scalable methodology to accuratelydetect and monitor IoT devices at subscriber lines with limited, highly sampleddata in-the-wild. Our findings indicate that millions of IoT devices aredetectable and identifiable within hours, both at a major ISP as well as anIXP, using passive, sparsely sampled network flow headers. Our methodology isable to detect devices from more than 77% of the studied IoT manufacturers,including popular devices such as smart speakers. While our methodology iseffective for providing network analytics, it also highlights significantprivacy consequences.
Siracusano G, Galea S, Sanvito D, et al., 2020, Running neural networks on the NIC, Publisher: arXiv
In this paper we show that the data plane of commodity programmable (NetworkInterface Cards) NICs can run neural network inference tasks required by packetmonitoring applications, with low overhead. This is particularly important asthe data transfer costs to the host system and dedicated machine learningaccelerators, e.g., GPUs, can be more expensive than the processing taskitself. We design and implement our system -- N3IC -- on two different NICs andwe show that it can greatly benefit three different network monitoring usecases that require machine learning inference as first-class-primitive. N3ICcan perform inference for millions of network flows per second, whileforwarding traffic at 40Gb/s. Compared to an equivalent solution implemented ona general purpose CPU, N3IC can provide 100x lower processing latency, with1.5x increase in throughput.
Osia SA, Shahin Shamsabadi A, Sajadmanesh S, et al., 2020, A Hybrid Deep Learning Architecture for Privacy-Preserving Mobile Analytics, IEEE INTERNET OF THINGS JOURNAL, Vol: 7, Pages: 4505-4518, ISSN: 2327-4662
- Author Web Link
- Cite
- Citations: 65
Lisi E, Malekzadeh M, Haddadi H, et al., 2020, Modelling and forecasting art movements with CGANs, Publisher: ROYAL SOC
- Author Web Link
- Cite
- Citations: 2
Shamsabadi AS, Gascon A, Haddadi H, et al., 2020, PrivEdge: from local to distributed private training and prediction, IEEE Transactions on Information Forensics and Security, Vol: 15, Pages: 3819-3831, ISSN: 1556-6013
Machine Learning as a Service (MLaaS) operators provide model training and prediction on the cloud. MLaaS applications often rely on centralised collection and aggregation of user data, which could lead to significant privacy concerns when dealing with sensitive personal data. To address this problem, we propose PrivEdge, a technique for privacy-preserving MLaaS that safeguards the privacy of users who provide their data for training, as well as users who use the prediction service. With PrivEdge, each user independently uses their private data to locally train a one-class reconstructive adversarial network that succinctly represents their training data. As sending the model parameters to the service provider in the clear would reveal private information, PrivEdge secret-shares the parameters among two non-colluding MLaaS providers, to then provide cryptographically private prediction services through secure multi-party computation techniques. We quantify the benefits of PrivEdge and compare its performance with state-of-the-art centralised architectures on three privacy-sensitive image-based tasks: individual identification, writer identification, and handwritten letter recognition. Experimental results show that PrivEdge has high precision and recall in preserving privacy, as well as in distinguishing between private and non-private images. Moreover, we show the robustness of PrivEdge to image compression and biased training data. The source code is available at https://github.com/smartcameras/PrivEdge.
Mo F, Shamsabadi AS, Katevas K, et al., 2020, DarkneTZ: towards model privacy at the edge using trusted execution environments, Publisher: arXiv
We present DarkneTZ, a framework that uses an edge device's Trusted ExecutionEnvironment (TEE) in conjunction with model partitioning to limit the attacksurface against Deep Neural Networks (DNNs). Increasingly, edge devices(smartphones and consumer IoT devices) are equipped with pre-trained DNNs for avariety of applications. This trend comes with privacy risks as models can leakinformation about their training data through effective membership inferenceattacks (MIAs). We evaluate the performance of DarkneTZ, including CPUexecution time, memory usage, and accurate power consumption, using two smalland six large image classification models. Due to the limited memory of theedge device's TEE, we partition model layers into more sensitive layers (to beexecuted inside the device TEE), and a set of layers to be executed in theuntrusted part of the operating system. Our results show that even if a singlelayer is hidden, we can provide reliable model privacy and defend against stateof the art MIAs, with only 3% performance overhead. When fully utilizing theTEE, DarkneTZ provides model protections with up to 10% overhead.
Lisi E, Malekzadeh M, Haddadi H, et al., 2020, Modeling and forecasting art movements with CGANs, Royal Society Open Science, Vol: 7, ISSN: 2054-5703
Conditional generative adversarial networks (CGANs) are a recent and popular method for generating samples from a probability distribution conditioned on latent information. The latent information often comes in the form of a discrete label from a small set. We propose a novel method for training CGANs which allows us to condition on a sequence of continuous latent distributions f(1), …, f(K). This training allows CGANs to generate samples from a sequence of distributions. We apply our method to paintings from a sequence of artistic movements, where each movement is considered to be its own distribution. Exploiting the temporal aspect of the data, a vector autoregressive (VAR) model is fitted to the means of the latent distributions that we learn, and used for one-step-ahead forecasting, to predict the latent distribution of a future art movement f(K+1). Realizations from this distribution can be used by the CGAN to generate ‘future’ paintings. In experiments, this novel methodology generates accurate predictions of the evolution of art. The training set consists of a large dataset of past paintings. While there is no agreement on exactly what current art period we find ourselves in, we test on plausible candidate sets of present art, and show that the mean distance to our predictions is small.
Malekzadeh M, Clegg RG, Cavallaro A, et al., 2020, Privacy and utility preserving sensor-data transformations, Pervasive and Mobile Computing, Vol: 63, Pages: 1-13, ISSN: 1574-1192
Sensitive inferences and user re-identification are major threats to privacywhen raw sensor data from wearable or portable devices are shared withcloud-assisted applications. To mitigate these threats, we propose mechanismsto transform sensor data before sharing them with applications running onusers' devices. These transformations aim at eliminating patterns that can beused for user re-identification or for inferring potentially sensitiveactivities, while introducing a minor utility loss for the target application(or task). We show that, on gesture and activity recognition tasks, we canprevent inference of potentially sensitive activities while keeping thereduction in recognition accuracy of non-sensitive activities to less than 5percentage points. We also show that we can reduce the accuracy of userre-identification and of the potential inference of gender to the level of arandom guess, while keeping the accuracy of activity recognition comparable tothat obtained on the original data.
Osia SA, Taheri A, Shamsabadi AS, et al., 2020, Deep Private-Feature Extraction, IEEE Transactions on Knowledge and Data Engineering, Vol: 32, Pages: 54-66, ISSN: 1041-4347
We present and evaluate Deep Private-Feature Extractor (DPFE), a deep model which is trained and evaluated based on information theoretic constraints. Using the selective exchange of information between a user's device and a service provider, DPFE enables the user to prevent certain sensitive information from being shared with a service provider, while allowing them to extract approved information using their model. We introduce and utilize the log-rank privacy, a novel measure to assess the effectiveness of DPFE in removing sensitive information and compare different models based on their accuracy-privacy trade-off. We then implement and evaluate the performance of DPFEon smartphones to understand its complexity, resource demands, and efficiency trade-offs. Our results on benchmark image datasets demonstrate that under moderate resource utilization, DPFE can achieve high accuracy for primary tasks while preserving the privacy of sensitive information.
Zhao Y, Haddadi H, Skillman S, et al., 2020, Privacy-preserving Activity and Health Monitoring on Databox, 3rd ACM International Workshop on Edge Systems, Analytics and Networking (EdgeSys), Publisher: ASSOC COMPUTING MACHINERY, Pages: 49-54
- Author Web Link
- Cite
- Citations: 9
Ren J, Dubois DJ, Choffnes D, et al., 2019, Information exposure from consumer IoT devices: a multidimensional, network-informed measurement approach, ACM Internet Measurement Conference (IMC), Publisher: ASSOC COMPUTING MACHINERY, Pages: 267-279
Internet of Things (IoT) devices are increasingly found in everyday homes, providing useful functionality for devices such as TVs, smart speakers, and video doorbells. Along with their benefits come potential privacy risks, since these devices can communicate information about their users to other parties over the Internet. However, understanding these risks in depth and at scale is difficult due to heterogeneity in devices' user interfaces, protocols, and functionality.In this work, we conduct a multidimensional analysis of information exposure from 81 devices located in labs in the US and UK. Through a total of 34,586 rigorous automated and manual controlled experiments, we characterize information exposure in terms of destinations of Internet traffic, whether the contents of communication are protected by encryption, what are the IoT-device interactions that can be inferred from such content, and whether there are unexpected exposures of private and/or sensitive information (e.g., video surreptitiously transmitted by a recording device). We highlight regional differences between these results, potentially due to different privacy regulations in the US and UK. Last, we compare our controlled experiments with data gathered from an in situ user study comprising 36 participants.
Aloufi R, Haddadi H, Boyle D, 2019, Emotion filtering at the edge, Publisher: arXiv
Voice controlled devices and services have become very popular in theconsumer IoT. Cloud-based speech analysis services extract information fromvoice inputs using speech recognition techniques. Services providers can thusbuild very accurate profiles of users' demographic categories, personalpreferences, emotional states, etc., and may therefore significantly compromisetheir privacy. To address this problem, we have developed a privacy-preservingintermediate layer between users and cloud services to sanitize voice inputdirectly at edge devices. We use CycleGAN-based speech conversion to removesensitive information from raw voice input signals before regeneratingneutralized signals for forwarding. We implement and evaluate our emotionfiltering approach using a relatively cheap Raspberry Pi 4, and show thatperformance accuracy is not compromised at the edge. In fact, signals generatedat the edge differ only slightly (~0.16%) from cloud-based approaches forspeech recognition. Experimental evaluation of generated signals show thatidentification of the emotional state of a speaker can be reduced by ~91%.
This data is extracted from the Web of Science and reproduced under a licence from Thomson Reuters. You may not copy or re-distribute this data in whole or in part without the written consent of the Science business of Thomson Reuters.