Search

Intent detection is a text classification task whose aim is to recognize and label the semantics behind a users query. It plays a critical role in various business applications. The output of the intent detection module strongly conditions the behavior of the whole system. This sequence analysis task is mainly tackled using deep learning techniques. Despite the widespread use of these techniques, the internal mechanisms used by networks to solve the problem are poorly understood. Recent lines of work have analyzed the computational mechanisms learned by RNNs from a dynamical systems perspective. In this work, we investigate how different RNN architectures solve the SNIPS intent detection problem. Sentences injected into trained networks can be interpreted as trajectories traversing a hidden state space. This space is constrained to a low-dimensional manifold whose dimensionality is related to the embedding and hidden layer sizes. To generate predictions, RNN steers the trajectories towards concrete regions, spatially aligned with the output layer matrix rows directions. Underlying the system dynamics, an unexpected fixed point topology has been identified with a limited number of attractors. Our results provide new insights into the inner workings of networks that solve the intent detection task., Comment: Camera-Ready version. Accepted paper at 27th European Conference on Artificial Intelligence (ECAI-2024)

Web browsers, particularly Google Chrome and other Chromium-based browsers, have grown in popularity over the past decade, with browser extensions becoming an integral part of their ecosystem. These extensions can customize and enhance the user experience, providing functionality that ranges from ad blockers to, more recently, AI assistants. Given the ever-increasing importance of web browsers, distribution marketplaces for extensions play a key role in keeping users safe by vetting submissions that display abusive or malicious behavior. In this paper, we characterize the prevalence of malware and other infringing extensions in the Chrome Web Store (CWS), the largest distribution platform for this type of software. To do so, we introduce SimExt, a novel methodology for detecting similarly behaving extensions that leverages static and dynamic analysis, Natural Language Processing (NLP) and vector embeddings. Our study reveals significant gaps in the CWS vetting process, as 86% of infringing extensions are extremely similar to previously vetted items, and these extensions take months or even years to be removed. By characterizing the top kinds of infringing extension, we find that 83% are New Tab Extensions (NTEs) and raise some concerns about the consistency of the vetting labels assigned by CWS analysts. Our study also reveals that only 1% of malware extensions flagged by the CWS are detected as malicious by anti-malware engines, indicating a concerning gap between the threat landscape seen by CWS moderators and the detection capabilities of the threat intelligence community.

This paper introduces an innovative dataset specifically crafted for challenging agricultural settings (a greenhouse), where achieving precise localization is of paramount importance. The dataset was gathered using a mobile platform equipped with a set of sensors typically used in mobile robots, as it was moved through all the corridors of a typical Mediterranean greenhouse featuring tomato crop. This dataset presents a unique opportunity for constructing detailed 3D models of plants in such indoor-like space, with potential applications such as robotized spraying. For the first time to the best knowledge of authors, a dataset suitable to put at test Simultaneous Localization and Mapping (SLAM) methods is presented in a greenhouse environment, which poses unique challenges. The suitability of the dataset for such goal is assessed by presenting SLAM results with state-of-the-art algorithms. The dataset is available online in \url{https://arm.ual.es/arm-group/dataset-greenhouse-2024/}., Comment: 29 pages, 15 figures

We study the entanglement entropy associated with a holographic RG flow from $\textrm{AdS}_7$ to $\textrm{AdS}_{4} \times \mathbb{H}_3$, where $\mathbb{H}_3$ is a $3$-dimensional hyperbolic manifold with curvature $\kappa$. The dual six-dimensional RG flow is disconnected from Lorentz-invariant flows. In this context we address various notions of central charges and identify a monotonic candidate $c$-function that captures IR aspects of the flow. The UV behavior of the holographic entanglement entropy and, in particular its universal term, display an interesting dependence on the curvature, $\kappa$. We then contrast our holographic results with existing field theory computations in six dimensions and find a series of new corrections in curvature to the universal term in the entanglement entropy., Comment: 31 pages, 9 figures. v2: References added, matches JHEP version

Since Pretrained Language Models (PLMs) are the cornerstone of the most recent Information Retrieval (IR) models, the way they encode semantic knowledge is particularly important. However, little attention has been given to studying the PLMs' capability to capture hierarchical semantic knowledge. Traditionally, evaluating such knowledge encoded in PLMs relies on their performance on a task-dependent evaluation approach based on proxy tasks, such as hypernymy detection. Unfortunately, this approach potentially ignores other implicit and complex taxonomic relations. In this work, we propose a task-agnostic evaluation method able to evaluate to what extent PLMs can capture complex taxonomy relations, such as ancestors and siblings. The evaluation is based on intrinsic properties that capture the hierarchical nature of taxonomies. Our experimental evaluation shows that the lexico-semantic knowledge implicitly encoded in PLMs does not always capture hierarchical relations. We further demonstrate that the proposed properties can be injected into PLMs to improve their understanding of hierarchy. Through evaluations on taxonomy reconstruction, hypernym discovery and reading comprehension tasks, we show that the knowledge about hierarchy is moderately but not systematically transferable across tasks., Comment: Accepted at ECIR 2024

The ultimate spatial limit to establish a Josephson coupling between two superconducting electrodes is an atomic-scale junction. The Josephson effect in such ultrasmall junctions has been used to unveil new switching dynamics, study coupling close to superconducting bound states or unveil non-reciprocal effects. However, coupling is weak and thermal smearing reduces the Cooper pair current magnitude. Here we show that a feedback element induces a time-dependent bistable regime which consists of periodic oscillations between two different Cooper pair tunneling states (corresponding to the DC and AC Josephson regimes respectively). The amplitude of the time-averaged current within the bistable regime is robust against thermal smearing. By tracing the periodic oscillations in the new bistable regime as a function of the position in a Scanning Tunneling Microscope (STM), we obtain atomic scale map of the critical current in 2H-NbSe$_2$ and find spatial modulations due to a pair density wave. Our results fundamentally improve our understanding of atomic size Josephson junctions and provide a promising new route to study superconducting materials through atomic scale maps of the Josephson coupling., Comment: 51 pages, 14 figures

This paper proposes a method for the deployment of a multi-agent system of unmanned aerial vehicles (UAVs) as a shield with potential applications in the protection of infrastructures. The shield shape is modeled as a quadric surface in the 3D space. To design the desired formation (target distances between agents and interconnections), an algorithm is proposed where the input parameters are just the parametrization of the quadric and the number of agents of the system. This algorithm guarantees that the agents are almost uniformly distributed over the virtual surface and that the topology is a Delaunay triangulation. Moreover, a new method is proposed to check if the resulting triangulation meets that condition and is executed locally. Because this topology ensures that the formation is rigid, a distributed control law based on the gradient of a potential function is proposed to acquire the desired shield shape and proofs of stability are provided. Finally, simulation and experimental results illustrate the effectiveness of the proposed approach.

Computer networks have been traditionally configured by humans using command-line interfaces. Some network abstractions have emerged in the last 10 years, but there is no easy way of comparing them to each other objectively. Therefore, there is no consensus in the industry of what direction modern network abstractions should take, and the adoption of these abstractions lags as a consequence. In this paper I propose a comparison framework using metrics derived from graph structures to evaluate the simplicity, efficiency, and effectiveness of different network abstraction models. The result of this comparison is that while some of the existing network abstractions are quite efficient to store network policy (such as the Kubernetes or the Cisco Application Centric Infrastructure models), others (notably public cloud) are still very infrastructure-centric and suffer from excessive complexity., Comment: 19 pages

Short Message Service (SMS) is a popular channel for online service providers to verify accounts and authenticate users registered to a particular service. Specialized applications, called Public SMS Gateways (PSGs), offer free Disposable Phone Numbers (DPNs) that can be used to receive SMS messages. DPNs allow users to protect their privacy when creating online accounts. However, they can also be abused for fraudulent activities and to bypass security mechanisms like Two-Factor Authentication (2FA). In this paper, we perform a large-scale and longitudinal study of the DPN ecosystem by monitoring 17,141 unique DPNs in 29 PSGs over the course of 12 months. Using a dataset of over 70M messages, we provide an overview of the ecosystem and study the different services that offer DPNs and their relationships. Next, we build a framework that (i) identifies and classifies the purpose of an SMS; and (ii) accurately attributes every message to more than 200 popular Internet services that require SMS for creating registered accounts. Our results indicate that the DPN ecosystem is globally used to support fraudulent account creation and access, and that this issue is ubiquitous and affects all major Internet platforms and specialized online services.

The Chromium open-source project has become a fundamental piece of the Web as we know it today, with multiple vendors offering browsers based on its codebase. One of its most popular features is the possibility of altering or enhancing the browser functionality through third-party programs known as browser extensions. Extensions have access to a wide range of capabilities through the use of APIs exposed by Chromium. The Debugger API -- arguably the most powerful of such APIs -- allows extensions to use the Chrome DevTools Protocol (CDP), a capability-rich tool for debugging and instrumenting the browser. In this paper, we describe several vulnerabilities present in the Debugger API and in the granting of capabilities to extensions that can be used by an attacker to take control of the browser, escalate privileges, and break context isolation. We demonstrate their impact by introducing six attacks that allow an attacker to steal user information, monitor network traffic, modify site permissions (\eg access to camera or microphone), bypass security interstitials without user intervention, and change the browser settings. Our attacks work in all major Chromium-based browsers as they are rooted at the core of the Chromium project. We reported our findings to the Chromium Development Team, who already fixed some of them and are currently working on fixing the remaining ones. We conclude by discussing how questionable design decisions, lack of public specifications, and an overpowered Debugger API have contributed to enabling these attacks, and propose mitigations.

Kappa-distributed velocities in plasmas are common in a wide variety of settings, from low-density to high-density plasmas, and appear in both space and laboratory plasmas. Despite being widely used as a model, the origin of the kappa distribution remains, to this day, unclear. For instance, deviations from the Maxwell-Boltzmann distribution are sometimes regarded as a signature of the non-additivity of the thermodynamic entropy, although there are alternative frameworks such as superstatistics where such an assumption is not needed. In this work we recover the kappa distribution for particle velocities in a non-equilibrium, steady-state plasma from the theory of superstatistics, assuming only a single requirement on the dependence between the kinetic energy of a test particle and that of its immediate environment. Most importantly, unlike previous derivations based on superstatistics, we do not make any assumption about temperature or its statistical distribution, instead obtaining them from the requirement on kinetic energies. Our results suggest that a description of kappa-distributed plasma is simpler in terms of features of the inverse temperature distribution rather than the traditional parameters $\kappa$ and the thermal velocity $v_{\text{th}}$.

Large language models (LLMs) have been leveraged for several years now, obtaining state-of-the-art performance in recognizing entities from modern documents. For the last few months, the conversational agent ChatGPT has "prompted" a lot of interest in the scientific community and public due to its capacity of generating plausible-sounding answers. In this paper, we explore this ability by probing it in the named entity recognition and classification (NERC) task in primary sources (e.g., historical newspapers and classical commentaries) in a zero-shot manner and by comparing it with state-of-the-art LM-based systems. Our findings indicate several shortcomings in identifying entities in historical text that range from the consistency of entity annotation guidelines, entity complexity, and code-switching, to the specificity of prompting. Moreover, as expected, the inaccessibility of historical archives to the public (and thus on the Internet) also impacts its performance., Comment: 5 pages, accepted to JCDL2023

Development of applications related to closed-loop control requires either testing on the field or on a realistic simulator, with the latter being more convenient, inexpensive, safe, and leading to shorter development cycles. To address that need, the present work introduces MVSim, a simulator for multiple vehicles or robots capable of running dozens of agents in simple scenarios, or a handful of them in complex scenarios. MVSim employs realistic physics-grounded friction models for tire-ground interaction, and aims at accurate and GPU-accelerated simulation of most common modern sensors employed in mobile robotics and autonomous vehicle research, such as depth and RGB cameras, or 2D and 3D LiDAR scanners. All depth-related sensors are able to accurately measure distances to 3D models provided by the user to define custom world elements. Efficient simulation is achieved by means of focusing on ground vehicles, which allows the use of a simplified 2D physics engine for body collisions while solving wheel-ground interaction forces separately. The core parts of the system are written in C++ for maximum efficiency, while Python, ROS 1, and ROS 2 wrappers are also offered for easy integration into user systems. A custom publish/subscribe protocol based on ZeroMQ (ZMQ) is defined to allow for multiprocess applications to access or modify a running simulation. This simulator enables and makes easier to do research and development on vehicular dynamics, autonomous navigation algorithms, and simultaneous localization and mapping (SLAM) methods., Comment: 6 pages, 6 figures, submitted

During the first days of the 2022 Russian invasion of Ukraine, Russia's media regulator blocked access to many global social media platforms and news sites, including Twitter, Facebook, and the BBC. To bypass the information controls set by Russian authorities, pro-Ukrainian groups explored unconventional ways to reach out to the Russian population, such as posting war-related content in the user reviews of Russian business available on Google Maps or Tripadvisor. This paper provides a first analysis of this new phenomenon by analyzing the creative strategies to avoid state censorship. Specifically, we analyze reviews posted on these platforms from the beginning of the conflict to September 2022. We measure the channeling of war messages through user reviews in Tripadvisor and Google Maps, as well as in VK, a popular Russian social network. Our analysis of the content posted on these services reveals that users leveraged these platforms to seek and exchange humanitarian and travel advice, but also to disseminate disinformation and polarized messages. Finally, we analyze the response of platforms in terms of content moderation and their impact.

This paper summarizes the joint participation of the Trading Central Labs and the L3i laboratory of the University of La Rochelle on both sub-tasks of the Shared Task FinSim-4 evaluation campaign. The first sub-task aims to enrich the 'Fortia ESG taxonomy' with new lexicon entries while the second one aims to classify sentences to either 'sustainable' or 'unsustainable' with respect to ESG (Environment, Social and Governance) related factors. For the first sub-task, we proposed a model based on pre-trained Sentence-BERT models to project sentences and concepts in a common space in order to better represent ESG concepts. The official task results show that our system yields a significant performance improvement compared to the baseline and outperforms all other submissions on the first sub-task. For the second sub-task, we combine the RoBERTa model with a feed-forward multi-layer perceptron in order to extract the context of sentences and classify them. Our model achieved high accuracy scores (over 92%) and was ranked among the top 5 systems.