Biblio

Recent research has shown that machine learning models are susceptible to adversarial examples, allowing attackers to trick a machine learning model into making a mistake and producing an incorrect output. Adversarial examples are commonly constructed or discovered by using gradient-based methods that require white-box access to the model. In most real-world AI system deployments, having complete access to the machine learning model is an unrealistic threat model. However, it is possible for an attacker to construct adversarial examples even in the black-box case - where we assume solely a query capability to the model - with a variety of approaches each with its advantages and shortcomings. We introduce AutoAttacker, a novel reinforcement learning framework where agents learn how to operate around the black-box model by querying it, to effectively extract the underlying decision behaviour, and to undermine it successfully. AutoAttacker is a first of kind framework that uses reinforcement learning and assumes nothing about the differentiability or structure of the underlying function and is thus robust towards common defenses like gradient obfuscation or adversarial training. Finally, without differentiable output, as in binary classification, most methods cease to operate and require either an approximation of the gradient, or another approach altogether. Our approach, however, maintains the capability to function when the output descriptiveness diminishes.

Internet of Things (IoT) devices offer new sources of contextual information, which can be leveraged by applications to make smart decisions. However, due to the decentralized and heterogeneous nature of such devices - each only having a partial view of their surroundings - there is an inherent risk of uncertain, unreliable and inconsistent observations. This is a serious concern for applications making security related decisions, such as context-aware authentication. We propose and evaluate a middleware for IoT that provides trustworthy context for a collaborative authentication use case. It abstracts a dynamic and distributed fusion scheme that extends the Chair-Varshney (CV) optimal decision fusion rule such that it can be used in a highly dynamic IoT environment. We compare performance and cost trade-offs against regular CV. Experimental evaluation demonstrates that our solution outperforms CV with 10% in a highly dynamic IoT environments, with the ability to detect and mitigate unreliable sensors.

Recent computing paradigms like cloud computing and big data have become very appealing to outsource computation and storage, making it easier to realize personalized and patient centric healthcare through real-time analytics on user data. Although these technologies can significantly complement resource constrained mobile and wearable devices to store and process personal health information, privacy concerns are keeping patients from reaping the full benefits. In this paper, we present and evaluate a practical smart-watch based lifelog application for diabetics that leverages the cloud and homomorphic encryption for caregivers to analyze blood glucose, insulin values, and other parameters in a privacy friendly manner to ensure confidentiality such that even a curious cloud service provider remains oblivious of sensitive health data.