Helmholtz Center for Information Security
CISPA – Helmholtz-Zentrum für InformationssicherheitNot a member yet
3406 research outputs found
Sort by
The Largest Connected Subgraph Game
Get PDFWe introduce the largest connected subgraph game played on an undirected graph G. Each round, Alice colours an uncoloured vertex of G red, and then, Bob colours one blue. Once every vertex is coloured, Alice (Bob, resp.) wins if there is a red (blue, resp.) connected subgraph whose order is greater than that of any blue (red, resp.) connected subgraph. If neither player wins, it is a draw. We prove that Alice can ensure Bob never wins, and define a class of graphs (reflection graphs) in which the game is a draw. We show that the game is PSPACE-complete in bipartite graphs of diameter 5, and that recognising reflection graphs is GI-hard. We prove that the game is a draw in paths if and only if the path has even order or at least 11 vertices, and that Alice wins in cycles if and only if the cycle is of odd order. We also give an algorithm computing the outcome of the game in cographs in linear time
The Peril of Popular Deep Learning Uncertainty Estimation Methods
Get PDFUncertainty estimation (UE) techniques -- such as the Gaussian process (GP), Bayesian neural networks (BNN), Monte Carlo dropout (MCDropout) -- aim to improve the interpretability of machine learning models by assigning an estimated uncertainty value to each of their prediction outputs. However, since too high uncertainty estimates can have fatal consequences in practice, this paper analyzes the above techniques.
Firstly, we show that GP methods always yield high uncertainty estimates on out of distribution (OOD) data. Secondly, we show on a 2D toy example that both BNNs and MCDropout do not give high uncertainty estimates on OOD samples. Finally, we show empirically that this pitfall of BNNs and MCDropout holds on real world datasets as well. Our insights (i) raise awareness for the more cautious use of currently popular UE methods in Deep Learning, (ii) encourage the development of UE methods that approximate GP-based methods -- instead of BNNs and MCDropout, and (iii) our empirical setups can be used for verifying the OOD performances of any other UE method. The source code is available at https://github.com/epfml/uncertainity-estimation
Exploring Authentication for Security-Sensitive Tasks on Smart Home Voice Assistants
Get PDFSmart home assistants such as Amazon Alexa and Google Home are primarily used for day-to-day tasks like checking the weather or controlling other IoT devices. Security-sensitive use cases such as online banking and voice-controlled door locks are already available and are expected to become more popular in the future.
However, the current state-of-the-art authentication for smart home assistants consists of users saying low-security PINs aloud, which does not meet the security requirements of security-sensitive tasks. Therefore, we explore the design space for future authentication mechanisms.
We conducted semi-structured interviews with N = 16 Alexa-users incorporating four high-risk scenarios. Using these scenarios, we explored perceived risks, mitigation strategies, and design-aspects to create secure experiences. Among other things, we found that participants are primarily concerned about eavesdropping bystanders, do not trust voice-based PINs, and would prefer trustworthy voice recognition. Our results also suggest that they have context-dependent (location and bystanders) requirements for smart home assistant authentication. Based on our findings, we construct design recommendations to inform the design of future authentication mechanisms
With a Little Help from My Friends: Constructing Practical Anonymous Credentials
Get PDFAnonymous credentials (ACs) are a powerful cryptographic tool for the secure use of digital services, when simultaneously aiming for strong privacy guarantees of users combined with strong authentication guarantees for providers of services. They allow users to selectively prove possession of attributes encoded in a credential without revealing any other meaningful information about themselves. While there is a significant body of research on AC systems, modern use-cases of ACs such as mobile applications come with various requirements not sufficiently considered so far. These include preventing the sharing of credentials and coping with resource constraints of the platforms (e.g., smart cards such as SIM cards in smartphones). Such aspects are typically out of scope of AC constructions, and, thus AC systems that can be considered entirely practical have been elusive so far.
In this paper we address this problem by introducing and formalizing the notion of core/helper anonymous credentials (CHAC). The model considers a constrained core device (e.g., a SIM card) and a powerful helper device (e.g., a smartphone). The key idea is that the core device performs operations that do not depend on the size of the credential or the number of attributes, but at the same time the helper device is unable to use the credential without its help. We present a provably secure generic construction of CHACs using a combination of signatures with flexible public keys (SFPK) and the novel notion of aggregatable attribute-based equivalence class signatures (AAEQ) along with a concrete instantiation. The key characteristics of our scheme are that the size of showing tokens is independent of the number of attributes in the credential(s) and that the core device only needs to compute a single elliptic curve scalar multiplication, regardless of the number of attributes. We confirm the practical efficiency of our CHACs with an implementation of our scheme on a Multos smart card as the core and an Android smartphone as the helper device. A credential showing requires less than 500 ms on the smart card and around 200 ms on the smartphone (even for a credential with 1000 attributes)
InfoScrub: Towards Attribute Privacy by Targeted Obfuscation
No full textPersonal photos of individuals when shared online, apart from exhibiting a myriad of memorable details, also reveals a wide range of private information and potentially entails privacy risks (e.g., online harassment, tracking). To mitigate such risks, it is crucial to study techniques that allow individuals to limit the private information leaked in visual data. We tackle this problem in a novel image obfuscation framework: to maximize entropy on inferences over targeted privacy attributes, while retaining image fidelity. We approach the problem based on an encoder-decoder style architecture, with two key novelties: (a) introducing a discriminator to perform bi-directional translation simultaneously from multiple unpaired domains; (b) predicting an image interpolation that maximizes uncertainty over a target set of attributes. We find our approach generates obfuscated images faithful to the original input images and additionally increases uncertainty by 6.2x (or up to 0.85 bits) over the non-obfuscated counterparts
Input Algebras
Get PDFGrammar-based test generators are highly efficient in producing syntactically valid test inputs, and give their user precise control over which test inputs should be generated. Adapting a grammar or a test generator towards a particular testing goal can be tedious, though. We introduce the concept of a grammar transformer, specializing a grammar towards inclusion or exclusion of specific patterns: “The phone number must not start with 011 or +1”. To the best of our knowledge, ours is the first approach to allow for arbitrary Boolean combinations of patterns, giving testers unprecedented flexibility in creating targeted software tests. The resulting specialized grammars can be used with any grammar-based fuzzer for targeted test generation, but also as validators to check whether the given specialization is met or not, opening up additional usage scenarios. In our evaluation on real-world bugs, we show that specialized grammars are accurate both in producing and validating targeted inputs
Euro-PVI: Pedestrian Vehicle Interactions in Dense Urban Centers
No full textAccurate prediction of pedestrian and bicyclist paths is integral to the development of reliable autonomous vehicles in dense urban environments. The interactions between vehicle and pedestrian or bicyclist have a significant impact on the trajectories of traffic participants e.g. stopping or turning to avoid collisions. Although recent datasets and trajectory prediction approaches have fostered the development of autonomous vehicles yet the amount of vehicle-pedestrian (bicyclist) interactions modeled are sparse. In this work, we propose Euro-PVI, a dataset of pedestrian and bicyclist trajectories. In particular, our dataset caters more diverse and complex interactions in dense urban scenarios compared to the existing datasets. To address the challenges in predicting future trajectories with dense interactions, we develop a joint inference model that learns an expressive multi-modal shared latent space across agents in the urban scene. This enables our Joint-b-cVAE approach to better model the distribution of future trajectories. We achieve state of the art results on the nuScenes and Euro-PVI datasets demonstrating the importance of capturing interactions between ego-vehicle and pedestrians (bicyclists) for accurate predictions
Beyond the Spectrum: Detecting Deepfakes via Re-Synthesis
Get PDFThe rapid advances in deep generative models over the past years have led to highly realistic media, known as deepfakes, that are commonly indistinguishable from real to human eyes. These advances make assessing the authenticity of visual data increasingly difficult and pose a misinformation threat to the trustworthiness of visual content in general. Although recent work has shown strong detection accuracy of such deepfakes, the success largely relies on identifying frequency artifacts in the generated images, which will not yield a sustainable detection approach as generative models continue evolving and closing the gap to real images. In order to overcome this issue, we propose a novel fake detection that is designed to re-synthesize testing images and extract visual cues for detection. The re-synthesis procedure is flexible, allowing us to incorporate a series of visual tasks - we adopt super-resolution, denoising and colorization as the re-synthesis. We demonstrate the improved effectiveness, cross-GAN generalization, and robustness against perturbations of our approach in a variety of detection scenarios involving multiple generators over CelebA-HQ, FFHQ, and LSUN datasets. Source code is available at https://github.com/SSAW14/BeyondtheSpectrum
BUFFing signature schemes beyond unforgeability and the case of post-quantum signatures
Get PDFModern digital signature schemes can provide more guarantees than the standard notion of(strong) unforgeability, such as offering security even in the presence of maliciously generated keys, or requiring to know a message to produce a signature for it. The use of signature schemes that lack these properties has previously enabled attacks on real-world protocols. In this work we revisit several of these notions beyond unforgeability, establish relations among them, provide the first formal definition of nonre-signability, and a transformation that can provide these properties for a given signature scheme in a provable and efficient way.Our results are not only relevant for established schemes: for example, the ongoing NIST PQC competition towards standardizing post-quantum signature schemes has six finalists in its third round. We perform an in-depth analysis of the candidates with respect to their security properties beyond unforgeability. We show that many of them do not yet offer these stronger guarantees, which implies that the security guarantees of these post-quantum schemes are not strictly stronger than, but instead incomparable to, classical signature schemes. We show how applying our transformation would efficiently solve this,paving the way for the standardized schemes to provide these additional guarantees and thereby making them harder to misuse
On the Impossibility of Purely Algebraic Signatures
No full textThe existence of one-way functions implies secure digital signatures, but not public-key encryption (at least in a black-box setting). Somewhat surprisingly, though, efficient public-key encryption schemes appear to be much easier to construct from concrete algebraic assumptions (such as the factoring of Diffie-Hellman-like assumptions) than efficient digital signature schemes. In this work, we provide one reason for this apparent difficulty to construct efficient signature schemes.
Specifically, we prove that a wide range of algebraic signature schemes (in which verification essentially checks a number of linear equations over a group) fall to conceptually surprisingly simple linear algebra attacks. In fact, we prove that in an algebraic signature scheme, sufficiently many signatures can be linearly combined to a signature of a fresh message. We present attacks both in known-order and hidden-order groups (although in hidden-order settings, we have to restrict our definition of algebraic signatures a little). More explicitly, we show:
the insecurity of all algebraic signature schemes in Maurer’s generic group model (in pairing-free groups), as long as these schemes do not rely on other cryptographic assumptions, such as hash functions.
the insecurity of a natural class of signatures in hidden-order groups, where verification consists of linear equations over group elements.
We believe that this highlights the crucial role of public verifiability in digital signature schemes. Namely, while public-key encryption schemes do not require any publicly verifiable structure on ciphertexts, it is exactly this structure on signatures that invites attacks like ours and makes it hard to construct efficient signatures