Search

Parametric oscillations of an interface separating two fluid phases create nonlinear surface waves, called Faraday waves, which organise into simple patterns, like squares and hexagons, as well as complex structures, such as double hexagonal and superlattice patterns. In this work, we study the influence of surfactant-induced Marangoni stresses on the formation and transition of Faraday wave patterns. We use a quantity $B$, that assesses the relative importance of Marangoni stresses as compared to the the surface wave dynamics. Our results show that the threshold acceleration required to destabilise a surfactant-covered interface through vibration increases with increasing $B$. For a surfactant-free interface, a square wave pattern is observed. As $B$ is incremented, we report transitions from squares to asymmetric squares, weakly wavy stripes, and ultimately to ridges and hills. These hills are a consequence of the bi-directional Marangoni stresses at the neck of the ridges. The mechanisms underlying the pattern transitions and the formation of exotic ridges and hills are discussed., Comment: 10 pages, 4 figures

Tokenization is a crucial step that bridges human-readable text with model-readable discrete tokens. However, recent studies have revealed that tokenizers can be exploited to elicit unwanted model behaviors. In this work, we investigate incomplete tokens, i.e., undecodable tokens with stray bytes resulting from byte-level byte-pair encoding (BPE) tokenization. We hypothesize that such tokens are heavily reliant on their adjacent tokens and are fragile when paired with unfamiliar tokens. To demonstrate this vulnerability, we introduce improbable bigrams: out-of-distribution combinations of incomplete tokens designed to exploit their dependency. Our experiments show that improbable bigrams are significantly prone to hallucinatory behaviors. Surprisingly, alternative tokenizations of the same phrases result in drastically lower rates of hallucination (93% reduction in Llama3.1). We caution against the potential vulnerabilities introduced by byte-level BPE tokenizers, which may impede the development of trustworthy language models.

Recent advancements in Large Language Models (LLMs) have established them as agentic systems capable of planning and interacting with various tools. These LLM agents are often paired with web-based tools, enabling access to diverse sources and real-time information. Although these advancements offer significant benefits across various applications, they also increase the risk of malicious use, particularly in cyberattacks involving personal information. In this work, we investigate the risks associated with misuse of LLM agents in cyberattacks involving personal data. Specifically, we aim to understand: 1) how potent LLM agents can be when directed to conduct cyberattacks, 2) how cyberattacks are enhanced by web-based tools, and 3) how affordable and easy it becomes to launch cyberattacks using LLM agents. We examine three attack scenarios: the collection of Personally Identifiable Information (PII), the generation of impersonation posts, and the creation of spear-phishing emails. Our experiments reveal the effectiveness of LLM agents in these attacks: LLM agents achieved a precision of up to 95.9% in collecting PII, generated impersonation posts where 93.9% of them were deemed authentic, and boosted click rate of phishing links in spear phishing emails by 46.67%. Additionally, our findings underscore the limitations of existing safeguards in contemporary commercial LLMs, emphasizing the urgent need for robust security measures to prevent the misuse of LLM agents., Comment: 20 pages, To appear in Usenix Security 2025

Recent advances in natural language processing and the increased use of large language models have exposed new security vulnerabilities, such as backdoor attacks. Previous backdoor attacks require input manipulation after model distribution to activate the backdoor, posing limitations in real-world applicability. Addressing this gap, we introduce a novel Claim-Guided Backdoor Attack (CGBA), which eliminates the need for such manipulations by utilizing inherent textual claims as triggers. CGBA leverages claim extraction, clustering, and targeted training to trick models to misbehave on targeted claims without affecting their performance on clean data. CGBA demonstrates its effectiveness and stealthiness across various datasets and models, significantly enhancing the feasibility of practical backdoor attacks. Our code and data will be available at https://github.com/PaperCGBA/CGBA., Comment: Under Review

Parameter-efficient fine-tuning (PEFT) has become a key training strategy for large language models. However, its reliance on fewer trainable parameters poses security risks, such as task-agnostic backdoors. Despite their severe impact on a wide range of tasks, there is no practical defense solution available that effectively counters task-agnostic backdoors within the context of PEFT. In this study, we introduce Obliviate, a PEFT-integrable backdoor defense. We develop two techniques aimed at amplifying benign neurons within PEFT layers and penalizing the influence of trigger tokens. Our evaluations across three major PEFT architectures show that our method can significantly reduce the attack success rate of the state-of-the-art task-agnostic backdoors (83.6%$\downarrow$). Furthermore, our method exhibits robust defense capabilities against both task-specific backdoors and adaptive attacks. Source code will be obtained at https://github.com/obliviateARR/Obliviate., Comment: Under Review

Twitter is recognized as a crucial platform for the dissemination and gathering of Cyber Threat Intelligence (CTI). Its capability to provide real-time, actionable intelligence makes it an indispensable tool for detecting security events, helping security professionals cope with ever-growing threats. However, the large volume of tweets and inherent noises of human-crafted tweets pose significant challenges in accurately identifying security events. While many studies tried to filter out event-related tweets based on keywords, they are not effective due to their limitation in understanding the semantics of tweets. Another challenge in security event detection from Twitter is the comprehensive coverage of security events. Previous studies emphasized the importance of early detection of security events, but they overlooked the importance of event coverage. To cope with these challenges, in our study, we introduce a novel event attribution-centric tweet embedding method to enable the high precision and coverage of events. Our experiment result shows that the proposed method outperforms existing text and graph-based tweet embedding methods in identifying security events. Leveraging this novel embedding approach, we have developed and implemented a framework, Tweezers, that is applicable to security event detection from Twitter for CTI gathering. This framework has demonstrated its effectiveness, detecting twice as many events compared to established baselines. Additionally, we have showcased two applications, built on Tweezers for the integration and inspection of security events, i.e., security event trend analysis and informative security user identification.

We present a numerical study of the main sub-stages preceding aerosol formation via bursting bubbles: capillary wave propagation along the bubble, convergence at the bubble's apex, the ascent of a Worthington jet and its break-up to release liquid drops. We focus on two crucial yet overlooked aspects of the system: the presence of surface-active agents and dynamics driven by non-negligible gravitational effects, quantified by the Bond number. Our results propose, for the first time, a mechanism explaining capillary wave retardation in the presence of surfactants, involving the transition from bi- to uni-directional Marangoni stresses, which pull the interface upwards, countering the motion of the waves. We also quantitatively elucidate the variable nature of the waves' velocity with various surfactant parameters, including surfactant solubility and elasticity, a departure from the constant behaviour well-documented in clean interfaces., Comment: 22 pages, 14 figures

Cybersecurity information is often technically complex and relayed through unstructured text, making automation of cyber threat intelligence highly challenging. For such text domains that involve high levels of expertise, pretraining on in-domain corpora has been a popular method for language models to obtain domain expertise. However, cybersecurity texts often contain non-linguistic elements (such as URLs and hash values) that could be unsuitable with the established pretraining methodologies. Previous work in other domains have removed or filtered such text as noise, but the effectiveness of these methods have not been investigated, especially in the cybersecurity domain. We propose different pretraining methodologies and evaluate their effectiveness through downstream tasks and probing tasks. Our proposed strategy (selective MLM and jointly training NLE token classification) outperforms the commonly taken approach of replacing non-linguistic elements (NLEs). We use our domain-customized methodology to train CyBERTuned, a cybersecurity domain language model that outperforms other cybersecurity PLMs on most tasks., Comment: To appear in NAACL Findings 2024

Recent research has suggested that there are clear differences in the language used in the Dark Web compared to that of the Surface Web. As studies on the Dark Web commonly require textual analysis of the domain, language models specific to the Dark Web may provide valuable insights to researchers. In this work, we introduce DarkBERT, a language model pretrained on Dark Web data. We describe the steps taken to filter and compile the text data used to train DarkBERT to combat the extreme lexical and structural diversity of the Dark Web that may be detrimental to building a proper representation of the domain. We evaluate DarkBERT and its vanilla counterpart along with other widely used language models to validate the benefits that a Dark Web domain specific model offers in various use cases. Our evaluations show that DarkBERT outperforms current language models and may serve as a valuable resource for future research on the Dark Web., Comment: 9 pages (main paper), 17 pages (including bibliography and appendix), to appear at the ACL 2023 Main Conference

Modern x86 processors support an AVX instruction set to boost performance. However, this extension may cause security issues. We discovered that there are vulnerable properties in implementing masked load/store instructions. Based on this, we present a novel AVX timing side-channel attack that can defeat address space layout randomization. We demonstrate the significance of our attack by showing User and Kernel ASLR breaks on the recent Intel and AMD processors in various environments, including cloud computing systems, an SGX enclave (a fine-grained ASLR break), and major operating systems. We further demonstrate that our attack can be used to infer user behavior, such as Bluetooth events and mouse movements. We highlight that stronger isolation or more fine-grained randomization should be adopted to successfully mitigate our presented attacks., Comment: Accepted to Design Automation Conference (DAC) 2023

NFT (Non-fungible Token) has drastically increased in its size, accounting for over \$16.9B of total market capitalization. Despite the rapid growth of NFTs, this market has not been examined thoroughly from a financial perspective. In this paper, we conduct methodical analyses to identify NFT market movers who play a significant role in potentially manipulating and oscillating NFT values. We collect over 3.8M NFT transaction data from the Ethereum Blockchain from January 2021 to February 2022 to extract trading information in line with the NFT lifecycle: (i) mint, (ii) transfer/sale, and (iii) burn. Based on the size of held NFT values, we classify NFT traders into three groups (whales, dolphins, and minnows). In total, we analyze 430K traders from 91 different NFT collection sources. We find that the top 0.1\% of NFT traders (i.e., whales) drive the NFT market with consistent, high returns. We then identify and characterize the NFT whales' unique investment strategies (e.g., mint/sale patterns, wash trading) to empirically understand the whales in the NFT market for the first time.

As Non-Fungible Tokens (NFTs) continue to grow in popularity, NFT users have become targets of phishing attacks by cybercriminals, called \textit{NFT drainers}. Over the last year, \$100 million worth of NFTs were stolen by drainers, and their presence remains a serious threat to the NFT trading space. However, no work has yet comprehensively investigated the behaviors of drainers in the NFT ecosystem. In this paper, we present the first study on the trading behavior of NFT drainers and introduce the first dedicated NFT drainer detection system. We collect 127M NFT transaction data from the Ethereum blockchain and 1,135 drainer accounts from five sources for the year 2022. We find that drainers exhibit significantly different transactional and social contexts from those of regular users. With these insights, we design \textit{DRAINCLoG}, an automatic drainer detection system utilizing Graph Neural Networks. This system effectively captures the multifaceted web of interactions within the NFT space through two distinct graphs: the NFT-User graph for transaction contexts and the User graph for social contexts. Evaluations using real-world NFT transaction data underscore the robustness and precision of our model. Additionally, we analyze the security of \textit{DRAINCLoG} under a wide variety of evasion attacks., Comment: To appear in NDSS 2024

We present a parametric study of the unsteady phenomena associated with the flow of elongated gas bubbles travelling through liquid-filled square capillaries under high Weber number conditions. These conditions consistently induce the formation of a re-entrant jet at the back of the bubble that commonly gives way to a deep liquid cavity. Subsequent steps include pinch-off events in the cavity to generate one or multiple encapsulated drops which may coalesce, in conjunction with the bursting of the bubble-liquid interface by either the cavity or the drops. Some of these interfacial instabilities have previously been reported experimentally (Olbricht 1996) and numerically (Izbassarov & Muradoglu 2016) for liquid-liquid flow in microchannels. We carry out three-dimensional direct numerical simulations based on a hybrid interface-tracking/level-set method capable of accounting for the presence and dynamic exchange of surfactants between the liquid bulk phase and the liquid-gas interface. Our results indicate that the delicate interplay amongst inertia, capillarity, viscosity, surfactant adsorption/desorption kinetics, and Marangoni stresses has a dramatic influence over the non-axisymmetric morphological structures of the encapsulated drops-elongated bubble. This strong coupling also influences the pinch-off time, penetration depth of the cavity, and number, size, and velocity of the encapsulated drops across the bubble. The observed phenomena are summarised in three main morphological regimes based on surfactant-related parameters and dimensionless groups. A discussion of the flow regime maps is also provided., Comment: 32 pages, 11 figures

The hidden nature and the limited accessibility of the Dark Web, combined with the lack of public datasets in this domain, make it difficult to study its inherent characteristics such as linguistic properties. Previous works on text classification of Dark Web domain have suggested that the use of deep neural models may be ineffective, potentially due to the linguistic differences between the Dark and Surface Webs. However, not much work has been done to uncover the linguistic characteristics of the Dark Web. This paper introduces CoDA, a publicly available Dark Web dataset consisting of 10000 web documents tailored towards text-based Dark Web analysis. By leveraging CoDA, we conduct a thorough linguistic analysis of the Dark Web and examine the textual differences between the Dark Web and the Surface Web. We also assess the performance of various methods of Dark Web page classification. Finally, we compare CoDA with an existing public Dark Web dataset and evaluate their suitability for various use cases., Comment: To appear at NAACL 2022 (main conference)

Fake news, false or misleading information presented as news, has a significant impact on many aspects of society, such as in politics or healthcare domains. Due to the deceiving nature of fake news, applying Natural Language Processing (NLP) techniques to the news content alone is insufficient. The multi-level social context information (news publishers and engaged users in social media) and temporal information of user engagement are important information in fake news detection. The proper usage of this information, however, introduces three chronic difficulties: 1) multi-level social context information is hard to be used without information loss, 2) temporal information is hard to be used along with multi-level social context information, 3) news representation with multi-level social context and temporal information is hard to be learned in an end-to-end manner. To overcome all three difficulties, we propose a novel fake news detection framework, Hetero-SCAN. We use Meta-Path to extract meaningful multi-level social context information without loss. Meta-Path, a composite relation connecting two node types, is proposed to capture the semantics in the heterogeneous graph. We then propose Meta-Path instance encoding and aggregation methods to capture the temporal information of user engagement and produce news representation end-to-end. According to our experiment, Hetero-SCAN yields significant performance improvement over state-of-the-art fake news detection methods.

The breakup of an interface into a cascade of droplets and their subsequent coalescence is a generic problem of central importance to a large number of industrial settings such as mixing, separations, and combustion. We study the breakup of a liquid jet introduced through a cylindrical nozzle into a stagnant viscous phase via a hybrid interface-tracking/level-set method to account for the surface tension forces in a three-dimensional Cartesian domain. Numerical solutions are obtained for a range of Reynolds (Re) and Weber (We) numbers. We find that the interplay between the azimuthal and streamwise vorticity components leads to different interfacial features and flow regimes in Re-We space. We show that the streamwise vorticity plays a critical role in the development of the three-dimensional instabilities on the jet surface. In the inertia-controlled regime at high Re and We, we expose the details of the spatio-temporal development of the vortical structures affecting the interfacial dynamics. A mushroom-like structure is formed at the leading edge of the jet inducing the generation of a liquid sheet in its interior that undergoes rupture to form droplets. These droplets rotate inside the mushroom structure due to their interaction with the prevailing vortical structures. Additionally, Kelvin-Helmholtz vortices that form near the injection point deform in the streamwise direction to form hairpin vortices, which, in turn, trigger the formation of interfacial lobes in the jet core. The thinning of the lobes induces the creation of holes which expand to form liquid threads that undergo capillary breakup to form droplets., Comment: 29 pages, 16 figures

We consider the mixing of a viscous fluid by the rotation of a pitched blade turbine inside an open, cylindrical tank, with air as the lighter fluid above. To examine the flow and interfacial dynamics, we utilise a highly-parallelised implementation of a hybrid front-tracking/level-set method that employs a domain-decomposition parallelisation strategy. Our numerical technique is designed to capture faithfully complex interfacial deformation, and changes of topology, including interface rupture and dispersed phase coalescence. As shown via transient, three-dimensional direct numerical simulations, the impeller induces the formation of primary vortices that arise in many idealised rotating flows as well as several secondary vortical structures resembling Kelvin-Helmholtz, vortex breakdown, blade tip vortices, and end-wall corner vortices. As the rotation rate increases, a transition to `aeration' is observed when the interface reaches the rotating blades leading to the entrainment of air bubbles into the viscous fluid and the creation of a bubbly, rotating, free surface flow. The mechanisms underlying the aeration transition are probed as are the routes leading to it, which are shown to exhibit a strong dependence on flow history., Comment: rejected version

We study the effect of insoluble surfactants on the wave dynamics of vertically-falling liquid films. We use three-dimensional numerical simulations and employ a hybrid interface-tracking/level-set method, taking into account Marangoni stresses induced by gradients of interfacial surfactant concentration. Our numerical predictions for the evolution of the surfactant-free, three-dimensional wave topology are validated against the experimental work of Park & Nosoko (2003). The addition of surfactants is found to influence significantly the development of horseshoe-shaped waves. At low Marangoni numbers, we show that the wave fronts exhibit spanwise oscillations before eventually acquiring a quasi two-dimensional shape. In addition, the presence of Marangoni stresses are found to suppress the peaks of the travelling waves and preceding capillary wave structures. At high Marangoni numbers, a near complete rigidification of the interface is observed., Comment: 10 pages, 5 figures

The effect of surfactants on the tail and film dynamics of elongated gas bubbles propagating through circular capillary tubes is investigated by means of an extensive three-dimensional numerical study using a hybrid front-tracking/level-set method. The focus is on the visco-inertial regime, which occurs when the Reynolds number of the flow is much larger than unity. Under these conditions, `clean' bubbles exhibit interface undulations in the proximity of the tail \cite{Magnini_prf_2017}, with an amplitude that increases with the Reynolds number. We perform a systematic analysis of the impact of a wide range of surfactant properties, including elasticity, bulk surfactant concentration, solubility, and diffusivity, on the bubble and flow dynamics in the presence of inertial effects. The results show that the introduction of surfactants is effective in suppressing the tail undulations as they tend to accumulate near the bubble tail. Here, large Marangoni stresses are generated, which lead to a local `rigidification' of the bubble. This effect becomes more pronounced for larger surfactant elasticities and adsorption depths. At reduced surfactant solubility, a thicker rigid film region forms at the bubble rear, where a Couette film flow is established, while undulations still appear at the trailing edge of the downstream `clean' film region. In such conditions, the bubble length becomes an influential parameter, with short bubbles becoming completely rigid., Comment: 20 pages and 12 figures

We consider a spherical variant of the Faraday problem, in which a spherical drop is subjected to a time-periodic body force, as well as surface tension. We use a full three-dimensional parallel front-tracking code to calculate the interface motion of the parametrically forced oscillating viscous drop, as well as the velocity field inside and outside the drop. Forcing frequencies are chosen so as to excite spherical harmonic wavenumbers ranging from 1 to 6. We excite gravity waves for wavenumbers 1 and 2 and observe translational and oblate-prolate oscillation, respectively. For wavenumbers 3 to 6, we excite capillary waves and observe patterns analogous to the Platonic solids. For low viscosity, both subharmonic and harmonic responses are accessible. The patterns arising in each case are interpreted in the context of the theory of pattern formation with spherical symmetry.

Bypass rewiring improves connectivity and robustness of networks against removal of nodes including failures and attacks. A concept of bypass rewiring on directed networks is proposed, and random bypass rewiring on infinite directed random networks is analytically and numerically investigated with simulations. As a result, random bypass rewiring makes infinite directed (undirected) random networks extremely robust for arbitrary occupation probabilities if and only if in-degree of every node except a fixed number of nodes is equal to the out-degree (every node except a finite number of nodes has even degree); bypass rewiring makes the percolation threshold 0. Consequently, a finite network has a strongly connected spanning subnetwork which has an Eulerian path or cycle if and only if there exists an way of bypass rewiring to make the finite network extremely robust for every combination of removed nodes; Eulerian networks are extremely robust for every combination of removed nodes.

We adapt and extend a formulation for soluble surfactant transport in multiphase flows recently presented by Muradoglu & Tryggvason (JCP 274 (2014) 737-757) to the context of the Level Contour Reconstruction Method (Shin et al. IJNMF 60 (2009) 753-778) which is a hybrid method that combines the advantages of the Front-tracking and Level Set methods. Particularly close attention is paid to the formulation and numerical implementation of the surface gradients of surfactant concentration and surface tension. Various benchmark tests are performed to demonstrate the accuracy of different elements of the algorithm. To verify surfactant mass conservation, values for surfactant diffusion along the interface are compared with the exact solution for the problem of uniform expansion of a sphere. The numerical implementation of the discontinuous boundary condition for the source term in the bulk concentration is compared with the approximate solution. Surface tension forces are tested for Marangoni drop translation. Our numerical results for drop deformation in simple shear are compared with experiments and results from previous simulations. All benchmarking tests compare well with existing data thus providing confidence that our adapted LCRM formulation for surfactant advection and diffusion is accurate and effective in three-dimensional multiphase flows. We also demonstrate that this approach applies easily to massively parallel simulations.

Three dimensional simulations on the viscous folding in diverging microchannels reported by Cubaud and Mason are performed using the parallel code BLUE for multi-phase flows. The more viscous liquid L_1 is injected into the channel from the center inlet, and the less viscous liquid L_2 from two side inlets. Liquid L_1 takes the form of a thin filament due to hydrodynamic focusing in the long channel that leads to the diverging region. The thread then becomes unstable to a folding instability, due to the longitudinal compressive stress applied to it by the diverging flow of liquid L_2. Given the long computation time, we were limited to a parameter study comprising five simulations in which the flow rate ratio, the viscosity ratio, the Reynolds number, and the shape of the channel were varied relative to a reference model. In our simulations, the cross section of the thread produced by focusing is elliptical rather than circular. The initial folding axis can be either parallel or perpendicular to the narrow dimension of the chamber. In the former case, the folding slowly transforms via twisting to perpendicular folding, or it may remain parallel. The direction of folding onset is determined by the velocity profile and the elliptical shape of the thread cross section in the channel that feeds the diverging part of the cell. Due to the high viscosity contrast and very low Reynolds numbers, direct numerical simulations of this two-phase flow are very challenging and to our knowledge these are the first three-dimensional direct parallel numerical simulations of viscous threads in microchannels. Our simulations provide good qualitative comparison of the early time onset of the folding instability however since the computational time for these simulations is quite long, for such viscous threads, long-time comparisons with experiments for quantities such as folding frequency are limited., Comment: 11pages, 8 figures, this is the pre-print (pre-referring) version. Accepted by Microfluidics and Nanofluidics on Sept. 11th, 2016

We report on the numerical and theoretical study of the subcritical bifurcation of parametrically amplified waves appearing at the interface between two immiscible incompressible fluids when the layer of the lower fluid is very shallow. As a critical control parameter is surpassed, small amplitude surface waves bifurcate towards highly nonlinear ones, with twice their amplitude. We propose a simple phenomenological model which can describe the observed bifurcation. We relate this hysteresis with the change of shear stress using a simple stress balance, in agreement with numerical results., Comment: 5 pages, 4 figures. Submitted to Physical Review E