Browsing by Author "Marchal, Samuel"
Now showing 1 - 20 of 23
Results Per Page
Sort Options
Item Access Control for Implantable Medical Devices(IEEE Computer Society, 2021) Camara, Carmen; Peris-Lopez, Pedro; De Fuentes, Jose Maria; Marchal, Samuel; Department of Computer Science; Professorship Aura Tuomas; Universidad Carlos III de MadridThe telemetry incorporate in the new generation of Implantable Medical Devices (IMDs) allows remote access and re-programming without interfering with the daily routine of their holders. Despite the benefits of this new feature, such remote access raises new threats related to the access of unauthorized entities to IMDs. Cardiac implants represent the most deployed types of IMD nowadays. Current solutions, to control their remote access, usually use a single feature for authentication. However, this feature is easily replicable, making these authentication schemes vulnerable to attacks. To overcome this limitation, we propose in this article a distance bounding protocol to manage access control of IMDs: ACIMD. ACIMD combines two security mechanisms, namely, identity verification (authentication) and proximity verification (distance checking). The authentication mechanism, formally and informally verified, conforms to the ISO/IEC 9798-2 standard. The distance checking is performed using the whole Electrocardiogram (ECG) signal and relies on the correlation coefficient (comparing an external versus an internal ECG signal) in the Hadamard domain. We evaluate the accuracy and security of ACIMD access control using ECG signals of 199 individuals recorded over 24 hours while considering three adversary strategies. Our results show that ACIMD is 92.92% accurate.Item Adversary Detection in Online Machine Learning Systems(2020-03-16) Szyller, Sebastian; Marchal, Samuel; Perustieteiden korkeakoulu; Asokan, N.Machine learning applications have become increasingly popular. At the same time, model training has become an expensive task in terms of computational power, amount of data, and human expertise. As a result, models now constitute intellectual property and business advantage to model owners and thus, their confidentiality must be preserved. Recently, it was shown that models can be stolen via model extraction attacks that do not require physical white-box access to the model but merely a black-box prediction API. Stolen model can be used to avoid paying for the service or even to undercut the offering of the legitimate model owner. Hence, it deprives the victim of the accumulated business advantage. In this thesis, we introduce two novel defense methods designed to detect distinct classes of model extraction attacks.Item AuDI: Towards autonomous IoT device-type identification using periodic communications(IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC, 2019-06-01) Marchal, Samuel; Miettinen, Markus; Nguyen, Thien Duc; Sadeghi, Ahmad-Reza; Asokan, N.; Department of Computer Science; Adj. Prof Asokan N. group; Helsinki Institute for Information Technology (HIIT); Technische Universität DarmstadtIoT devices are being widely deployed. But the huge variance among them in the level of security and requirements for network resources makes it unfeasible to manage IoT networks using a common generic policy. One solution to this challenge is to define policies for classes of devices based on device type. In this paper, we present AuDI, a system for quickly and effectively identifying the type of a device in an IoT network by analyzing their network communications. AuDI models the periodic communication traffic of IoT devices using an unsupervised learning method to perform identification. In contrast to prior work, AuDI operates autonomously after initial setup, learning, without human intervention nor labeled data, to identify previously unseen device types. AuDI can identify the type of a device in any mode of operation or stage of lifecycle of the device. Via systematic experiments using 33 off-the-shelf IoT devices, we show that AuDI is effective (98.2% accuracy).Item Automated Deauthentication using Web Transaction Analysis(2017-10-04) Tomsu, Radek; Marchal, Samuel; Perustieteiden korkeakoulu; Asokan, N.Companies commonly provide work related devices enabled with Internet connection to their employees. Usually, all the company's incoming and outgoing Internet traffic is checked by some protection system, eg. by a firewall. Commonly deployed protection systems use static rules that ``allow'' or ``block'' the traffic. However, these rules can not detect changes in user behaviors. Modeling user behavior may be beneficial if it is sufficiently unique with respect to activities of other users or attackers. An automated deauthentication system that is able to recognize if behavior of an active user corresponds to the behavior of an authorized and expected user is proposed in the thesis. The system can recognize an innocent attacker in more than 50\% cases and a legitimate user in more than 95\% cases. The system is expected to work together with other authentication systems.Item Automatic Ownership Change Detection for IoT devices(2018-08-20) Valiev, Artur; Marchal, Samuel; Perustieteiden korkeakoulu; N., AsokanConsidering the constant increases in Internet Of Things (IoT) smart home devices prevalence, their ownership is likely to change. This introduces novel privacy issues. Smart home devices store owner’s sensitive information, which needs to be handled securely in case of change in device ownership. Currently employed smart home devices cannot detect changes in their ownership, which raises a great number of privacy and security issues. To address this problem, we propose a system called FoundIoT for automatic detection of IoT device ownership change. FoundIoT provides a technique to detect change of ownership based on device context, which is inferred by monitoring wireless communication channels. Finally, we present a prototype implementation of FoundIoT for the proposed automatic ownership change detection technique. We show that FoundIoT achieves a satisfactory performance. The implementation is supported by a wide range of IoT devices and demonstrates a high speed (up to 1 minute 39 seconds) and 100% accuracy of ownership change detection.Item chowniot: Enhancing IoT privacy by automated handling of ownership change(Springer, 2019-01-01) Khan, Md Sakib Nizam; Marchal, Samuel; Buchegger, Sonja; Asokan, N.; Department of Computer Science; Slamanig, Daniel; Krenn, Stephan; Fischer-Hübner, Simone; Pierson, Jo; Kosta, Eleni; Adj. Prof Asokan N. group; Professorship Aura Tuomas; Helsinki Institute for Information Technology (HIIT); KTH Royal Institute of TechnologyConsidering the increasing deployment of smart home IoT devices, their ownership is likely to change during their life-cycle. IoT devices, especially those used in smart home environments, contain privacy-sensitive user data, and any ownership change of such devices can result in privacy leaks. The problem arises when users are either not aware of the need to reset/reformat the device to remove any personal data, or not trained in doing it correctly as it can be unclear what data is kept where. In addition, if the ownership change is due to theft or loss, then there is no opportunity to reset. Although there has been a lot of research on security and privacy of IoT and smart home devices, to the best of our knowledge, there is no prior work specifically on automatically securing ownership changes. We present a system called for securely handling ownership change of IoT devices. combines authentication (of both users and their smartphone), profile management, data protection by encryption, and automatic inference of ownership change. For the latter, we use a simple technique that leverages the context of a device. Finally, as a proof of concept, we develop a prototype that implements inferring ownership change from changes in the WiFi SSID. The performance evaluation of the prototype shows that has minimal overhead and is compatible with the dominant IoT boards on the market.Item Detecting organized eCommerce fraud using scalable categorical clustering(2019) Marchal, Samuel; Szyller, Sebastian; Department of Computer Science; Professorship Aura Tuomas; Adj. Prof Asokan N. groupOnline retail, eCommerce, frequently falls victim to fraud conducted by malicious customers (fraudsters) who obtain goods or services through deception. Fraud coordinated by groups of professional fraudsters that place several fraudulent orders to maximize their gain is referred to as organized fraud. Existing approaches to fraud detection typically analyze orders in isolation and they are not effective at identifying groups of fraudulent orders linked to organized fraud. These also wrongly identify many legitimate orders as fraud, which hinders their usage for automated fraud cancellation. We introduce a novel solution to detect organized fraud by analyzing orders in bulk. Our approach is based on clustering and aims to group together fraudulent orders placed by the same group of fraudsters. It selectively uses two existing techniques, agglomerative clustering and sampling to recursively group orders into small clusters in a reasonable amount of time. We assess our clustering technique on real-world orders placed on the Zalando website, the largest online apparel retailer in Europe1. Our clustering processes 100,000s of orders in a few hours and groups 35-45% of fraudulent orders together. We propose a simple technique built on top of our clustering that detects 26.2% of fraud while raising false alarms for only 0.1% of legitimate orders.Item Distributed and scalable parsing solution for telecom network data(2020-01-20) Khan, Muhammad; Marchal, Samuel; Vaje, Toivo; Perustieteiden korkeakoulu; Aura, TuomasThe growing usage of mobile devices and the introduction of 5G networks have increased the significance of network data for the telecom business. The success of telecom organizations can depend on employing efficient data engineering techniques for transforming raw network data into useful information by analytics and machine learning (ML). Elisa Oyj., a Finnish telecommunications company, receives massive amounts of network data from network equipment manufactured by various vendors. The effectiveness of data analytics depends on efficient data engineering processes. This thesis presents a scalable data parsing solution that leverages Spark, a distributed programming framework, for parallelizing parsing routines from an existing parsing solution. We design and deploy this solution as a component of the organization's data engineering pipeline to enable automation of data-centric operations. Experimental results indicate that the efficiency of the proposed solution is heavily dependent on the individual file size distribution. The proposed parsing solution demonstrates reliability, scalability, and speed during empirical evaluation and processes a 24-hour network data within 3 hours. The main outcome of the project is an optimized setup with the minimum number of data partitions to ensure zero failures and thus minimum execution time. A smaller execution time leads to lower costs of the continuously running infrastructure provisioned on the cloud.Item DNN model extraction attacks using prediction interfaces(2018-08-20) Dmitrenko, Alexey; Juuti, Mika; Marchal, Samuel; Perustieteiden korkeakoulu; Asokan, N.Machine learning (ML) and deep learning methods have become common and publicly available, while ML security to date struggles to cope with rising threats. One rising threat is model extraction attacks where adversaries are able to reproduce a target model close to perfection. The attack is widely deployable since the attacker needs only to have access to predictions to perform this attack. Stolen ML models could either be used for personal advantage to abuse paid prediction services or to create transferable adversarial examples that can be used to undermine the integrity of prediction services, i.e. prediction quality. This is a significant threat in several application areas, such as in autonomous driving, which rely heavily of computer vision via deep neural networks. In this thesis, we reproduce existing model extraction attacks and evaluate novel techniques to extract deep neural network (DNN) classifiers. We introduce new synthetic query generation strategies, and demonstrate their efficiency at extracting models for creating transferable targeted adversarial examples from stolen DNNs.Item Enhancing Privacy in IoT Devices through Automated Handling of Ownership Change(2017-08-28) Khan, Md; Marchal, Samuel; Perustieteiden korkeakoulu; Asokan, N.Considering the increasing deployment of IoT devices, their ownership is likely to change during their life cycle. Personal IoT devices used in smart home environment contain privacy sensitive user data. Ownership change of such devices can introduce threats against privacy sensitive data handled by them. To address this problem, we present a system called chownIoT for securely handling ownership change of IoT devices. chownIoT introduces a privacy enhancement protocol that leverages authentication and data encryption for protecting owner privacy. We also present an owner profile management scheme for better management of owners during the life cycle of a device. For automatic detection of ownership change, we use a simple technique which leverage the context of a device. Finally, we present a prototype that implements chownIoT including the privacy enhancement protocol and the owner profile management scheme.Item FLAME: Taming Backdoors in Federated Learning(2022) Nguyen, Thien Duc; Rieger, Phillip; Chen, Huili; Yalame, Hossein; Möllering, Helen; Fereidooni, Hossein; Marchal, Samuel; Miettinen, Markus; Mirhoseini, Azalia; Zeitouni, Shaza; Koushanfar, Farinaz; Sadeghi, Ahmad Reza; Schneider, Thomas; Department of Computer Science; Professorship Aura Tuomas; Technische Universität Darmstadt; University of California, San Diego; Alphabet Inc.Federated Learning (FL) is a collaborative machine learning approach allowing participants to jointly train a model without having to share their private, potentially sensitive local datasets with others. Despite its benefits, FL is vulnerable to so-called backdoor attacks, in which an adversary injects manipulated model updates into the federated model aggregation process so that the resulting model will provide targeted false predictions for specific adversary-chosen inputs. Proposed defenses against backdoor attacks based on detecting and filtering out malicious model updates consider only very specific and limited attacker models, whereas defenses based on differential privacy-inspired noise injection significantly deteriorate the benign performance of the aggregated model. To address these deficiencies, we introduce FLAME, a defense framework that estimates the sufficient amount of noise to be injected to ensure the elimination of backdoors. To minimize the required amount of noise, FLAME uses a model clustering and weight clipping approach. This ensures that FLAME can maintain the benign performance of the aggregated model while effectively eliminating adversarial backdoors. Our evaluation of FLAME on several datasets stemming from application areas including image classification, word prediction, and IoT intrusion detection demonstrates that FLAME removes backdoors effectively with a negligible impact on the benign performance of the models.Item Know Your Phish: Novel Techniques for Detecting Phishing Sites and Their Targets(IEEE, 2016) Marchal, Samuel; Saari, Kalle; Singh, Nidhi; Asokan, N.; Tietotekniikan laitos; Department of Computer Science; Secure Systems; Perustieteiden korkeakoulu; School of SciencePhishing is a major problem on the Web. Despite the significant attention it has received over the years, there has been no definitive solution. While the state-of-the-art solutions have reasonably good performance, they require a large amount of training data and are not adept at detecting phishing attacks against new targets. In this paper, we begin with two core observations: (a) although phishers try to make a phishing webpage look similar to its target, they do not have unlimited freedom in structuring the phishing webpage, and (b) a webpage can be characterized by a small set of key terms, how these key terms are used in different parts of a webpage is different in the case of legitimate and phishing webpages. Based on these observations, we develop a phishing detection system with several notable properties: it requires very little training data, scales well to much larger test data, is language-independent, fast, resilient to adaptive attacks and implemented entirely on client-side. In addition, we developed a target identification component that can identify the target website that a phishing webpage is attempting to mimic. The target detection component is faster than previously reported systems and can help minimize false positives in our phishing detection system.Item Know Your Phish: Novel Techniques for Detecting Phishing Sites and Their Targets(IEEE Computer Society, 2016-08-10) Marchal, Samuel; Saari, Kalle; Singh, Nidhi; Asokan, N; Department of Computer Science; University of HelsinkiPhishing is a major problem on the Web. Despite the significant attention it has received over the years, there has been no definitive solution. While the state-of-the-art solutions have reasonably good performance, they require a large amount of training data and are not adept at detecting phishing attacks against new targets. In this paper, we begin with two core observations: (a) although phishers try to make a phishing webpage look similar to its target, they do not have unlimited freedom in structuring the phishing webpage, and (b) a webpage can be characterized by a small set of key terms, how these key terms are used in different parts of a webpage is different in the case of legitimate and phishing webpages. Based on these observations, we develop a phishing detection system with several notable properties: it requires very little training data, scales well to much larger test data, is language-independent, fast, resilient to adaptive attacks and implemented entirely on client-side. In addition, we developed a target identification component that can identify the target website that a phishing webpage is attempting to mimic. The target detection component is faster than previously reported systems and can help minimize false positives in our phishing detection system.Item Malicious Entity Categorization using Graph modeling(2016-10-27) Srinivaasan, Gayathri; Marchal, Samuel; Ranta-aho, Perttu; Perustieteiden korkeakoulu; Asokan, NToday, malware authors not only write malicious software but also employ obfuscation, polymorphism, packing and endless such evasive techniques to escape detection by Anti-Virus Products (AVP). Besides the individual behavior of malware, the relations that exist among them play an important role for improving malware detection. This work aims to enable malware analysts at F-Secure Labs to explore various such relationships between malicious URLs and file samples in addition to their individual behavior and activity. The current detection methods at F-Secure Labs analyze unknown URLs and file samples independently without taking into account the correlations that might exist between them. Such traditional classification methods perform well but are not efficient at identifying complex multi-stage malware that hide their activity. The interactions between malware may include any type of network activity, dropping, downloading, etc. For instance, an unknown downloader that connects to a malicious website which in turn drops a malicious payload, should indeed be blacklisted. Such analysis can help block the malware infection at its source and also comprehend the whole infection chain. The outcome of this proof-of-concept study is a system that detects new malware using graph modeling to infer their relationship to known malware as part of the malware classification services at F-Secure.Item Mitigating Threats in IoT Network using Device Isolation(2018-03-19) Thapa, Manish; Marchal, Samuel; Perustieteiden korkeakoulu; N., AsokanIn recent years, the proliferation of the Internet of Things (IoT) is seen across various sectors. There is a sharp inclination towards using IoT devices in both home and office premises. Many traditional manufacturers are enhancing their traditional appliances into IoT devices. With the myriad of devices in the market, there also exist vulnerable devices which can be exploited by adversaries. Several security solutions are trying to address different areas of security such as network security, privacy, threat detection, etc. IoT Sentinel is one such novel system that can identify device types based on their pattern of communication. IoT Sentinel proposes several isolation levels that can be used to control the traffic of devices identified as vulnerable. IoT Sentinel uses a Software-defined Networking (SDN) component for controlling the traffic flow for devices and isolating them. In this thesis, we develop a solution to extend IoT Sentinel for device isolation, which is not dependent on SDN. The goal is to build a generic and deployable solution for network segmentation and device isolation that is suitable for home networks. The system divides the network into isolated subnets and places new devices into appropriate subnets. Communication between the subnets is controlled using a firewall thereby isolating them. We dynamically configure a DHCP server to place (lease IP address) new IoT devices identified by IoT Sentinel into appropriate subnets based on their level of vulnerability. Using our solution, we can confine vulnerable devices. Thus, the solution minimizes the damage that could be caused by vulnerable devices present in a network. Finally, we evaluate the developed solution for its security requirement of device isolation. We also present the performance evaluation of our solution based on time-delay and throughput analysis. We observe that our solution adds an acceptable delay to the existing IoT Sentinel processes. We also observe that the system throughput is not significantly affected by firewall rules in a home network scenario.Item On Designing and Evaluating Phishing Webpage Detection Techniques for the Real World(2018) Marchal, Samuel; Asokan, N.; Department of Computer Science; Adj. Prof Asokan N. group; Helsinki Institute for Information Technology (HIIT)While a plethora of apparently foolproof detection techniques have been developed to cope with phishing, it remains a continuing problem with an increasing number of attacks and victims. This is due to a gap between the reported experimental detection accuracy of solutions from the academic literature and their actual effectiveness in real-world scenarios. For instance, design choices made while only considering how to maximize the accuracy of phishing detection sometimes has the unintended effect of constraining deployability or limiting usability. We hope to raise awareness about practices causing this gap and present a set of guidelines for the design and evaluation of phishing webpage detection techniques. These guidelines can improve the effectiveness of phishing detection techniques in real-world scenarios and foster technology transfer. They also facilitate unbiased comparison of evaluation results of different detection techniques.Item Privacy Preserving Deep Neural Network Prediction using Trusted Hardware(2018-11-07) Reuter, Max; Paverd, Andrew; Marchal, Samuel; Perustieteiden korkeakoulu; Asokan, NIn recent years machine learning has gained a lot of attention not only in the scientific community but also in user-facing applications. Today, many applications utilise machine learning to take advantage of its capabilities. With such applications, users actively or passively input data that is used by state-of-the-art algorithms to generate accurate predictions. Due to the extensive work necessary to fine-tune these algorithms for a specific task, they are predominantly executed in the cloud where they can be protected from competitors or malicious users. As a result, users' privacy might be at risk as their data is sent to and processed by remote cloud services. Depending on the application, users might expose highly sensitive data, meaning a malicious provider could harvest extensive amounts of personal data from its users. In order to protect user privacy without compromising the confidentiality guarantees of traditional solutions, we propose using trusted hardware for privacy preserving deep neural network predictions. Our solution consists of a hardware-backed prediction service and a client device that connects to said service. All machine learning computations executed by the prediction service that depend on input data are protected by a trusted hardware component, called a Trusted Execution Environment. This can be verified by users via remote attestation to ensure their data remains protected. In addition, we have built a proof-of-concept implementation of our solution using Intel Software Guard Extensions (SGX). Compared to existing solutions relying on homomorphic encryption, our proof-of-concept implementation vastly increases the set of supported machine learning algorithms. Moreover, our implementation is tightly integrated into the existing pipeline of machine learning tools by supporting the Open Neural Network Exchange (ONNX) Format. Furthermore, we focus on minimising our Trusted Computing Base (TCB), thus our proof-of-concept implementation only consists of 4,500 lines of code. Additionally, we achieve a 7x increase in throughput whilst decreasing the latency 40x compared to prior work. In our tests, SGX reduced throughput by 11% and increased latency by 21% compared to our baseline implementation without SGX.Item PrivICN: Privacy-preserving content retrieval in information-centric networking(Elsevier, 2019-02-11) Bernardini, Cesar; Marchal, Samuel; Asghar, Muhammad Rizwan; Crispo, Bruno; Department of Computer Science; Adj. Prof Asokan N. group; University of Auckland; University of Trento; Barracuda NetworksInformation-Centric Networking (ICN) has emerged as a paradigm to cope with the increasing demand for content delivery on the Internet. In contrast to the Internet Protocol (IP), the underlying architecture of ICN enables users to request contents based on their name rather than their hosting location (IP address). On the one hand, this preserves users’ anonymity since packet routing does not require source and destination addresses of the communication parties. On the other hand, semantically-rich names reveal information about users’ interests, which poses serious threats to their privacy. A curious ICN node can monitor the traffic to profile users’ or censor specific contents for instance. In this paper, we present PrivICN: a system that enhances users privacy in ICN by protecting the confidentiality of content names and content data. PrivICN relies on a proxy encryption scheme and has several features that distinguish it from existing solutions: it preserves full in-network caching benefits, it does not require end-to-end communication between consumers and providers and it provides flexible user management (addition/removal of users). We evaluate PrivICN in a real ICN network (CCNx implementation) showing that it introduces an acceptable overhead and little delay. PrivICN is publicly available as an open-source library.Item Real-Time Client-Side Phishing Prevention(2016-08-24) Armano, Giovanni; Marchal, Samuel; Perustieteiden korkeakoulu; Asokan, NIn the last decades researchers and companies have been working to deploy effective solutions to steer users away from phishing websites. These solutions are typically based on servers or blacklisting systems. Such approaches have several drawbacks: they compromise user privacy, rely on off-line analysis, are not robust against adaptive attacks and do not provide much guidance to the users in their warnings. To address these limitations, we developed a fast real-time client-side phishing prevention software that implements a phishing detection technique recently developed by Marchal et al. It extracts information from the visited webpage and detects if it is a phish to warn the user. It is also able to detect the website that the phish is trying to mimic and propose a redirection to the legitimate domain. Furthermore, to attest the validity of our solution we performed two user studies to evaluate the usability of the interface and the program's impact on user experience.Item Robust Aggregation Technique Against Poisoning Attacks in Multi-Stage Federated Learning Applications(2024) Siriwardhana, Yushan; Porambage, Pawani; Liyanage, Madhusanka; Marchal, Samuel; Ylianttila, Mika; Department of Computer Science; Professorship Aura Tuomas; University of Oulu; VTT Technical Research Centre of Finland; University College DublinFederated Learning (FL) is a distributed Machine Learning (ML) technique that allows model training without sharing data. FL is vulnerable to poisoning attacks where an adversary manipulates the learning process by providing false information to the federation. Ensuring security in FL is vital before using FL in real applications, as the consequences can be adverse. Multi-stage FL is a novel variant of FL that performs intermediate model aggregations, thereby reducing the traffic toward the FL central server. The existing robust aggregation techniques are insufficient in multi-stage FL systems. This paper proposes a novel robust aggregation algorithm against poisoning attacks in a three-layer multi-stage FL system that consists of device, edge, and cloud layers. We evaluate the proposed robust algorithm considering an Augmented Reality (AR) application with different poisoner placements and attack strategies. The evaluation results show that the proposed algorithm can effectively defend against poisoning attacks in three-layer multi-stage FL systems.
