Helmholtz Center for Information Security
CISPA – Helmholtz-Zentrum für InformationssicherheitNot a member yet
3406 research outputs found
Sort by
On the Evolution of (Hateful) Memes by Means of Multimodal Contrastive Learning
Get PDFThe dissemination of hateful memes online has adverse effects on social media platforms and the real world. Detecting hateful memes is challenging, one of the reasons being the evolutionary nature of memes; new hateful memes can emerge by fusing hateful connotations with other cultural ideas or symbols. In this paper, we propose a framework that leverages multimodal contrastive learning models, in particular OpenAI’s CLIP, to identify targets of hateful content and systematically investigate the evolution of hateful memes. We find that semantic regularities exist in CLIP-generated embeddings that describe semantic relationships within the same modality (images) or across modalities (images and text). Leveraging this property, we study how hateful memes are created by combining visual elements from multiple images or fusing textual information with a hateful image. We demonstrate the capabilities of our framework for analyzing the evolution of hateful memes by focusing on antisemitic memes, particularly the Happy Merchant meme. Using our framework on a dataset extracted from 4chan, we find 3.3K variants of the Happy Merchant meme, with some linked to specific countries, persons, or organizations. We envision that our framework can be used to aid human moderators by flagging new variants of hateful memes so that moderators can manually verify them and mitigate the problem of hateful content online
Counterfactuals Modulo Temporal Logics
Get PDFLewis' theory of counterfactuals is the foundation of many contemporary notions of causality. In this paper, we extend this theory in the temporal direction to enable symbolic counterfactual reasoning on infinite sequences, such as counterexamples found by a model checker and trajectories produced by a reinforcement learning agent. In particular, our extension considers a more relaxed notion of similarity between worlds and proposes two additional counterfactual operators that close a semantic gap between the previous two in this more general setting. Further, we consider versions of counterfactuals that minimize the distance to the witnessing counterfactual worlds, a common requirement in causal analysis. To automate counterfactual reasoning in the temporal domain, we introduce a logic that combines temporal and counterfactual operators, and outline decision procedures for the satisfiability and trace-checking problems of this logic
Learning Program Models from Generated Inputs
Get PDFRecent advances in Machine Learning (ML) show that Neural Machine Translation (NMT) models can mock the program behavior when trained on input-output pairs. Such models can mock the functionality of existing programs and serve as quick-to-deploy reverse engineering tools. Still, the problem of automatically learning such predictive and reversible models from programs needs to be solved. This work introduces a generic approach for automated and reversible program behavior modeling. It achieves 94% of overall accuracy in the conversion of Markdown-to-HTML and HTML-to-Markdown markups
Robust Routing Made Easy: Reinforcing Networks Against Non-Benign Faults
Get PDFWith the increasing scale of communication networks, the likelihood of failures grows as well. Since these networks form a critical backbone of our digital society, it is important that they rely on robust routing algorithms which ensure connectivity despite such failures. While most modern communication networks feature robust routing mechanisms, these mechanisms are often fairly complex to design and verify, as they need to account for the effects of failures and rerouting on communication.
This paper conceptualizes the design of robust routing mechanisms, with the aim to avoid such complexity. In particular, we showcase simple and generic blackbox transformations that increase resilience of routing against independently distributed failures, which allows to simulate the routing scheme on the original network, even in the presence of non-benign node failures (henceforth called faults). This is attractive as the system specification and routing policy can simply be preserved.
We present a scheme for constructing such a reinforced network, given an existing (synchronous) network and a routing scheme. We prove that this algorithm comes with small constant overheads, and only requires a minimal amount of additional node and edge resources; in fact, if the failure probability is smaller than 1/n, the algorithm can come without any overhead at all. At the same time, it allows to tolerate a large number of independent random (node) faults, asymptotically almost surely. We complement our analytical results with simulations on different real-world topologies
Semantic Debugging
Get PDFWhy does my program fail? We present a novel and general technique to automatically determine failure causes and conditions, using logical properties over input elements: "The program fails if and only if int(⟨length⟩) > len(⟨payload⟩) holds - that is, the given ⟨length⟩ is larger than the ⟨payload⟩ length." Our AVICENNA prototype uses modern techniques for inferring properties of passing and failing inputs and validating and refining hypotheses by having a constraint solver generate supporting test cases to obtain such diagnoses. As a result, AVICENNA produces crisp and expressive diagnoses even for complex failure conditions, considerably improving over the state of the art with diagnoses close to those of human experts
Second-Order Hyperproperties
Get PDFWe introduce Hyper^2LTL, a temporal logic for the specification of hyperproperties that allows for second-order quantification over sets of traces. Unlike first-order temporal logics for hyperproperties, such as HyperLTL, Hyper^2LTL can express complex epistemic properties like common knowledge, Mazurkiewicz trace theory, and asynchronous hyperproperties. The model checking problem of Hyper^2LTL is, in general, undecidable. For the expressive fragment where second-order quantification is restricted to smallest and largest sets, we present an approximate model-checking algorithm that computes increasingly precise under- and overapproximations of the quantified sets, based on fixpoint iteration and automata learning. We report on encouraging experimental results with our model-checking algorithm, which we implemented in the tool HySO
Automata-Based Software Model Checking of Hyperproperties
No full textWe develop model checking algorithms for Temporal Stream Logic (TSL) and Hyper Temporal Stream Logic (HyperTSL) modulo theories. TSL extends Linear Temporal Logic (LTL) with memory cells, functions and predicates, making it a convenient and expressive logic to reason over software and other systems with infinite data domains. HyperTSL further extends TSL to the specification of hyperproperties – properties that relate multiple system executions. As such, HyperTSL can express information flow policies like noninterference in software systems. We augment HyperTSL with theories, resulting in HyperTSL(T), and build on methods from LTL software verification to obtain model checking algorithms for TSL and HyperTSL(T). This results in a sound but necessarily incomplete algorithm for specifications contained in the ∀∗∃∗ fragment of HyperTSL(T). Our approach constitutes the first software model checking algorithm for temporal hyperproperties with quantifier alternations that does not rely on a finite-state abstraction
Compositional High-Quality Synthesis
Get PDFOver the last years, there has been growing interest in synthesizing reactive systems from quantitative specifications, with the goal of constructing correct and high-quality systems. Considering quantitative requirements in systems consisting of multiple components is challenging not only because of scalability limitations but also due to the intricate interplay between the different possibilities of satisfying a specification and the required cooperation between components. Compositional synthesis holds the promise of addressing these challenges.
We study the compositional synthesis of reactive systems consisting of multiple components, from requirements specified in a fragment of the logic LTL[F], which extends LTL with quality operators. We consider specifications that are combinations of local and shared quantitative requirements. We present a sound decomposition rule that allows for synthesizing one component at a time. The decomposition requires assume-guarantee contracts between the components, and we provide a method for iteratively refining the assumptions and guarantees. We evaluate our approach with a prototype implementation, demonstrating its advantages over monolithic synthesis and ability to generate decompositions
Checking and Sketching Causes on Temporal Sequences
No full textTemporal causality describes what concrete input behavior is responsible for some observed output behavior on a trace of a reactive system, and can be used to, e.g., generate explanations for counterexamples uncovered by a model checker. In this paper, we present CATS, the first tool that can automatically verify whether a given temporal property (specified in QPTL) is a cause for some observed ω-regular effect. In addition to checking whether a given property is a cause, CATS can search for potential causes by exhaustively exploring a cause sketch, i.e., a temporal formula in which some parts are left unspecified. Our experiments show that CATS can effectively check causes and search for causes in small reactive systems
Unsafe Diffusion: On the Generation of Unsafe Images and Hateful Memes From Text-To-Image Models
Get PDFState-of-the-art Text-to-Image models like Stable Diffusion and DALLE·2 are revolutionizing how people generate visual content. At the same time, society has serious concerns about how adversaries can exploit such models to generate problematic or unsafe images. In this work, we focus on demystifying the generation of unsafe images and hateful memes from Text-to-Image models. We first construct a typology of unsafe images consisting of five categories (sexually explicit, violent, disturbing, hateful, and political).
Then, we assess the proportion of unsafe images generated by four advanced Text-to-Image models using four prompt datasets. We find that Text-to-Image models can generate a substantial percentage of unsafe images; across four models and four prompt datasets, 14.56% of all generated images are unsafe. When comparing the four Text-to-Image models, we find different risk levels, with Stable Diffusion being the most prone to generating unsafe content (18.92% of all generated images are unsafe). Given Stable Diffusion’s tendency to generate more unsafe content, we evaluate its potential
to generate hateful meme variants if exploited by an adversary to attack a specific individual or community. We employ three image editing methods, DreamBooth, Textual Inversion, and SDEdit, which are supported by Stable Diffusion to generate variants. Our evaluation result shows that 24% of the generated images using DreamBooth are hateful meme variants that present the features of the original hateful meme and the target individual/community; these generated images are comparable to hateful meme variants collected from the real world. Overall, our results demonstrate that the danger of large-scale generation of unsafe images is imminent.
We discuss several mitigating measures, such as curating training data, regulating prompts, and implementing safety filters, and encourage better safeguard tools to be developed to prevent unsafe generation