Impersonation attacks against web authentication servers have been increasing in complexity over the last decade. Tunnelling services, such as VPNs or proxies, can be for instance used to faithfully impersonate victims in foreign countries. In this paper we study the detection of user authentication attacks involving network tunnelling geolocation deception. For that purpose we explore different models to profile a user based on network latencies. We design a classical machine learning model and a deep learning model to profile web resource loading times collected on client-side. In order to test our approach we profiled network latencies for 86 real users located around the globe. We show that our proposed novel network profiling is able to detect up to 88.3% of attacks using VPN tunneling schemes.

No abstract

Risk based authentication has been advocated as complement to traditional authentication mechanisms in order to raise the bar against attackers in possession of stolen credentials. Behavioral biometrics has received attention in the literature in the past decade, however the best results have been obtained in the so-called continuous setting and with enough training data, usually spanning several hours of user interaction. In this paper we explore the more challenging scenario of behavioral biometrics as an effective risk-based authentication technique using both mouse and keyboard information at login time (static authentication), assuming only between 3 and 7 login sessions per user for training. In a controlled but realistic experiment with 89 subjects we achieve a FRR of 10.73% and FAR of 23.34% for a model trained using only 5 login attempts, each performed in less than 30 seconds on average. We also evaluate our prototype with 2000 users from production data in the banking domain.

Objective and Impact Statement: We apply a deep learning (DL) segmentation method and automate the extraction of imaging markers for neonatal lung structure using magnetic resonance imaging (MRI) in order to inform clinical care with robust and quantifiable information about the neonatal lung. Introduction: Quantification of lung structural information in a standardized fashion is crucial to inform diagnostic processes that enable personalized treatment and monitoring strategies. Increased efficiency and accuracy in image quantification is especially needed in prematurely born infants, for whom long-term survival is critically determined by acute and chronic pulmonary complications, currently diagnosed based on clinical criteria due to the lack of routinely applicable diagnostic tools. Methods: We prospectively enrolled 107 premature infants in two clinical centers with and without chronic lung disease, i.e., Bronchopulmonary Dysplasia (BPD) to perform quiet-breathing lung MRI. An ensemble of deep convolutional neural networks was developed to perform lung segmentation, with a subsequent reconstruction of the 3-dimensional lung and computation of MRI volumetric measurements and compared to the standard manual segmentation. Results: The DL model successfully annotates lung segments with a volumetric dice score of 0.908 (Site 1) and 0.880 (Site 2), thereby reaching expert-level performance while demonstrating high transferability between study sites and robustness towards technical (low spatial resolution, movement artifacts) and disease conditions. Estimated lung volumes correlated with infant lung function testing measures and enabled the separation of neonates with and without BPD. Conclusion: Our work demonstrates the potential of AI-supported MRI measures to perform monitoring of neonatal lung development and characterization of respiratory diseases in this high-risk patient cohort.

No abstract

No abstract

Adversarial attacks have gained popularity recently due to their simplicity and impact. Their applicability to diverse security scenarios is however less understood. In particular, in some scenarios, attackers may come up naturally with ad-hoc black-box attack techniques inspired directly on characteristics of the problem space rather than using generic adversarial techniques. In this paper we explore an intuitive attack technique for Mouse-based Behavioral Biometrics and compare its effectiveness against adversarial machine learning attacks. We show that attacks leveraging on domain knowledge have higher transferability when applied to various machine-learning techniques and are also more difficult to defend against. We also propose countermeasures against such attacks and discuss their effectiveness.

No abstract