The evaluation of vision-language models (VLMs) has mainly relied on English-language benchmarks, leaving significant gaps in both multilingual and multicultural coverage. While multilingual benchmarks have expanded, both in size and languages, many rely on translations of English datasets, failing to capture cultural nuances. In this work, we propose Kaleidoscope, as the most comprehensive exam benchmark to date for the multilingual evaluation of vision-language models. Kaleidoscope is a large-scale, in-language multimodal benchmark designed to evaluate VLMs across diverse languages and visual inputs. Kaleidoscope covers 18 languages and 14 different subjects, amounting to a total of 20,911 multiple-choice questions. Built through an open science collaboration with a diverse group of researchers worldwide, Kaleidoscope ensures linguistic and cultural authenticity. We evaluate top-performing multilingual vision-language models and find that they perform poorly on low-resource languages and in complex multimodal scenarios. Our results highlight the need for progress on culturally inclusive multimodal evaluation frameworks.

We introduce Universal NER (UNER), an open, community-driven project to develop gold-standard NER benchmarks in many languages. The overarching goal of UNER is to provide high-quality, cross-lingually consistent annotations to facilitate and standardize multilingual NER research. UNER v1 contains 18 datasets annotated with named entities in a cross-lingual consistent schema across 12 diverse languages. In this paper, we detail the dataset creation and composition of UNER; we also provide initial modeling baselines on both in-language and cross-lingual learning settings. We release the data, code, and fitted models to the public.

We analyze a specific class of forward-scattering diagrams with imaginary kinematics, which, via the optical theorem, describe processes involving two identical fermions occupying the same state. What initially seems to be a contradiction turns out to be a key element in how the Pauli exclusion principle manifests itself in scattering theory. The discussion is entirely basic and could easily fit into any quantum field theory textbook. To the best of our knowledge, however, this point has not been addressed in the literature, and we aim to fill this gap.

A $d$-dimensional nowhere-zero $r$-flow on a graph $G$, an $(r,d)$-NZF from now on, is a flow where the value on each edge is an element of $\mathbb{R}^d$ whose (Euclidean) norm lies in the interval $[1, r-1]$. Such a notion is a natural generalization of the well-known concept of a circular nowhere-zero $r$-flow (i.e.\ $d = 1$). The minimum of the real numbers $r$ such that a graph $G$ admits an $(r, d)$-NZF is called the $d$-dimensional flow number of $G$ and is denoted by $\phi_d(G)$. In this paper we provide a geometric description of some $d$-dimensional flows on a graph $G$, and we prove that the existence of a suitable cycle double cover of $G$ is equivalent, for $G$, to admit such a geometrically constructed $(r,d)$-NZF. This geometric approach allows us to provide upper bounds for $\phi_{d-2}(G)$ and $\phi_{d-1}(G)$, assuming that $G$ admits an (oriented) $d$-cycle double cover.

Ontologies are a standard tool for creating semantic schemata in many knowledge intensive domains of human interest. They are becoming increasingly important also in the areas that have been until very recently dominated by subsymbolic knowledge representation and machine-learning (ML) based data processing. One such area is information security, and specifically, malware detection. We thus propose PE Malware Ontology that offers a reusable semantic schema for Portable Executable (PE - the Windows binary format) malware files. This ontology is inspired by the structure of the EMBER dataset, which focuses on the static malware analysis of PE files. With this proposal, we hope to provide a unified semantic representation for the existing and future PE-malware datasets and facilitate the application of symbolic, neuro-symbolic, or otherwise explainable approaches in the PE-malware-detection domain, which may produce interpretable results described by the terms defined in our ontology. In addition, we also publish semantically treated EMBER data, including fractional datasets, to support the reproducibility of experiments on EMBER. We supplement our work with a preliminary case study, conducted using concept learning, to show the general feasibility of our approach. While we were not able to match the precision of the state-of-the-art ML tools, the learned malware discriminators were interesting and highly interpretable.

The characterization and analysis of light curves are vital for understanding the physical and rotational properties of artificial space objects such as satellites, rocket stages, and space debris. This paper introduces the Light Curve Dataset Creator (LCDC), a Python-based toolkit designed to facilitate the preprocessing, analysis, and machine learning applications of light curve data. LCDC enables seamless integration with publicly available datasets, such as the newly introduced Mini Mega Tortora (MMT) database. Moreover, it offers data filtering, transformation, as well as feature extraction tooling. To demonstrate the toolkit's capabilities, we created the first standardized dataset for rocket body classification, RoBo6, which was used to train and evaluate several benchmark machine learning models, addressing the lack of reproducibility and comparability in recent studies. Furthermore, the toolkit enables advanced scientific analyses, such as surface characterization of the Atlas 2AS Centaur and the rotational dynamics of the Delta 4 rocket body, by streamlining data preprocessing, feature extraction, and visualization. These use cases highlight LCDC's potential to advance space debris characterization and promote sustainable space exploration. Additionally, they highlight the toolkit's ability to enable AI-focused research within the space debris community.

Raman spectroscopy is a powerful experimental technique for characterizing molecules and materials that is used in many laboratories. First-principles theoretical calculations of Raman spectra are important because they elucidate the microscopic effects underlying Raman activity in these systems. These calculations are often performed using the canonical harmonic approximation which cannot capture certain thermal changes in the Raman response. Anharmonic vibrational effects were recently found to play crucial roles in several materials, which motivates theoretical treatments of the Raman effect beyond harmonic phonons. While Raman spectroscopy from molecular dynamics (MD-Raman) is a well-established approach that includes anharmonic vibrations and further relevant thermal effects, MD-Raman computations were long considered to be computationally too expensive for practical materials computations. In this perspective article, we highlight that recent advances in the context of machine learning have now dramatically accelerated the involved computational tasks without sacrificing accuracy or predictive power. These recent developments highlight the increasing importance of MD-Raman and related methods as versatile tools for theoretical prediction and characterization of molecules and materials.

The capabilities of recent large language models (LLMs) to generate high-quality content indistinguishable by humans from human-written texts raises many concerns regarding their misuse. Previous research has shown that LLMs can be effectively misused for generating disinformation news articles following predefined narratives. Their capabilities to generate personalized (in various aspects) content have also been evaluated and mostly found usable. However, a combination of personalization and disinformation abilities of LLMs has not been comprehensively studied yet. Such a dangerous combination should trigger integrated safety filters of the LLMs, if there are some. This study fills this gap by evaluating vulnerabilities of recent open and closed LLMs, and their willingness to generate personalized disinformation news articles in English. We further explore whether the LLMs can reliably meta-evaluate the personalization quality and whether the personalization affects the generated-texts detectability. Our results demonstrate the need for stronger safety-filters and disclaimers, as those are not properly functioning in most of the evaluated LLMs. Additionally, our study revealed that the personalization actually reduces the safety-filter activations; thus effectively functioning as a jailbreak. Such behavior must be urgently addressed by LLM developers and service providers.

We present a novel approach to constructing gravitational instantons based on the observation that the gravitational action of general relativity in its teleparallel formulation can be expressed as a product of the torsion and excitation forms. We introduce a new class of solutions where these two forms are equal, which we term the self-excited instantons, and advocate for their use over the self-dual instantons of Eguchi and Hanson. These new self-excited instantons exhibit striking similarities to BPST instantons in Yang-Mills theory, as their action reduces to a topological Nieh-Yan term, which allows us to identify the axial torsion as a topological current and show that the gravitational action is given by a topological charge.

We propose a Quantum Field Theory (QFT) approach to neutrino oscillations in vacuum. The neutrino emission and detection are identified with the charged-current vertices of a single second-order Feynman diagram for the underlying process, enclosing neutrino propagation between these two points. The key point of our approach is the definition of the space-time setup typical for neutrino oscillation experiments, implying macroscopically large but finite volumes of the source and detector separated by a sufficiently large distance $L$. We derive an $L$-dependent master formula for the charged lepton production rate, which provides the QFT basis for the analysis of neutrino oscillations. Our formula depends on the underlying process and is not reducible to the conventional approach resorting to the concept of neutrino oscillation probability, which originates from non-relativistic quantum mechanics (QM). We demonstrate that for some particular choice of the underlying process our QFT formula approximately coincides with the conventional one under some assumptions.

We study a Hermitian matrix model with a quartic potential, modified by a curvature term $\mathrm{tr}(R\Phi^2)$, where $R$ is a fixed external matrix. Motivated by the truncated Heisenberg algebra formulation of the Grosse-Wulkenhaar model, this term breaks unitary invariance and gives rise to an effective multitrace matrix model via perturbative expansion. We analyze the resulting action analytically and numerically, focusing on the shift of the triple point and suppression of the noncommutative stripe phase -- features linked to renormalizability. Our findings, supported by Hamiltonian Monte Carlo simulations, indicate that the curvature term drives the phase structure toward renormalizable behavior by eliminating the stripe phase.

We introduce the novel method for estimation of mean and Gaussian curvature and several related quantities for polygonal meshes. The algebraic quadric fitting curvature (AQFC) is based on local approximation of the mesh vertices and associated normals by a quadratic surface. The quadric is computed as an implicit surface, so it minimizes algebraic distances and normal deviations from the approximated point-normal neighbourhood of the processed vertex. Its mean and Gaussian curvature estimate is then obtained as the respective curvature of its orthogonal projection onto the fitted quadratic surface. Experimental results for both sampled parametric surfaces and arbitrary meshes are provided. The proposed method AQFC approaches the true curvatures of the reference smooth surfaces with increasing density of sampling, regardless of its regularity. It is resilient to irregular sampling of the mesh, compared to the contemporary curvature estimators. In the case of arbitrary meshes, obtained from scanning, AQFC provides robust curvature estimation.

We consider a $2$-dimensional autonomous system subject to a $1$-periodic perturbation, i.e. $$ \dot{\vec{x}}=\vec{f}(\vec{x})+\epsilon\vec{g}(t,\vec{x},\epsilon),\quad \vec{x}\in\Omega .$$ We assume that for $\epsilon=0$ there is a trajectory $\vec{\gamma}(t)$ homoclinic to the origin which is a critical point: in this context Melnikov theory provides a sufficient condition for the insurgence of a chaotic pattern when $\epsilon \ne 0$. In this paper we show that for any line $\Xi$ transversal to $\{\vec{\gamma}(t) \mid t \in \mathbb{R} \}$ and any $\tau \in [0,1]$ we can find a set $\Sigma^+(\Xi,\tau)$ of initial conditions giving rise to a pattern chaotic just in the future, located in $\Xi$ at $t=\tau$. Further diam$(\Sigma^+(\Xi,\tau)) \le \epsilon^{(1+\nu)/ \underline{\sigma}}$ where $\underline{\sigma}>0$ is a constant and $\nu>0$ is a parameter that can be chosen as large as we wish. The same result holds true for the set $\Sigma^-(\Xi,\tau)$ of initial conditions giving rise to a pattern chaotic just in the past. In fact all the results are developed in a piecewise-smooth context, assuming that $\vec{0}$ lies on the discontinuity curve $\Omega^0$: we recall that in this setting chaos is not possible if we have sliding phenomena close to the origin. This paper can also be considered as the first part of the project to show the existence of classical chaotic phenomena when sliding close to the origin is not present.

A vertex-girth-regular $vgr(v,k,g,\lambda)$-graph is a $k$-regular graph of girth $g$ and order $v$ in which every vertex belongs to exactly $\lambda$ cycles of length $g$. While all vertex-transitive graphs are necessarily vertex-girth-regular, the majority of vertex-girth-regular graphs are not vertex-transitive. Similarly, while many of the smallest $k$-regular graphs of girth $g$, the so-called $(k,g)$-cages, are vertex-girth-regular, infinitely many vertex-girth-regular graphs of degree $k$ and girth $g$ exist for many pairs $k,g$. Due to these connections, the study of vertex-girth-regular graphs promises insights into the relations between the classes of extremal, highly symmetric, and locally regular graphs of given degree and girth. This paper lays the foundation to such study by investigating the fundamental properties of $vgr(v,k,g,\lambda)$-graphs, specifically the relations necessarily satisfied by the parameters $v,k,g$ and $\lambda$ to admit the existence of a corresponding vertex-girth-regular graph, by presenting constructions of infinite families of $vgr(v,k,g,\lambda)$-graphs, and by establishing lower bounds on the number $v$ of vertices in a $vgr(v,k,g,\lambda)$-graph. It also includes computational results determining the orders of smallest cubic and quartic graphs of small girths.

The problem of self-calibration of two cameras from a given fundamental matrix is one of the basic problems in geometric computer vision. Under the assumption of known principal points and square pixels, the well-known Bougnoux formula offers a means to compute the two unknown focal lengths. However, in many practical situations, the formula yields inaccurate results due to commonly occurring singularities. Moreover, the estimates are sensitive to noise in the computed fundamental matrix and to the assumed positions of the principal points. In this paper, we therefore propose an efficient and robust iterative method to estimate the focal lengths along with the principal points of the cameras given a fundamental matrix and priors for the estimated camera parameters. In addition, we study a computationally efficient check of models generated within RANSAC that improves the accuracy of the estimated models while reducing the total computational time. Extensive experiments on real and synthetic data show that our iterative method brings significant improvements in terms of the accuracy of the estimated focal lengths over the Bougnoux formula and other state-of-the-art methods, even when relying on inaccurate priors.

A $d$-dimensional nowhere-zero $r$-flow on a graph $G$, an $(r,d)$-NZF from now on, is a flow where the value on each edge is an element of $\mathbb{R}^d$ whose (Euclidean) norm lies in the interval $[1,r-1]$. Such a notion is a natural generalization of the well-known concept of circular nowhere-zero $r$-flow (i.e.\ $d=1$). For every bridgeless graph $G$, the $5$-flow Conjecture claims that $\phi_1(G)\leq 5$, while a conjecture by Jain suggests that $\phi_d(G)=1$, for all $d \geq 3$. Here, we address the problem of finding a possible upper-bound also for the remaining case $d=2$. We show that, for all bridgeless graphs, $\phi_2(G) \le 1 + \sqrt{5}$ and that the oriented $5$-cycle double cover Conjecture implies $\phi_2(G)\leq \tau^2$, where $\tau$ is the Golden Ratio. Moreover, we propose a geometric method to describe an $(r,2)$-NZF of a cubic graph in a compact way, and we apply it in some instances. Our results and some computational evidence suggest that $\tau^2$ could be a promising upper bound for the parameter $\phi_2(G)$ for an arbitrary bridgeless graph $G$. We leave that as a relevant open problem which represents an analogous of the $5$-flow Conjecture in the $2$-dimensional case (i.e. complex case).

We analyze multitrace random matrix models with the help of the saddle point approximation and we introduce a multitrace term of type $-c_1c_3$ to the action. We obtain the numerical phase diagram of the model, with a stable asymmetric phase and the triple point. Furthermore, we examine response functions in this model.

In this paper, we propose a novel approach for recovering focal lengths from three-view homographies. By examining the consistency of normal vectors between two homographies, we derive new explicit constraints between the focal lengths and homographies using an elimination technique. We demonstrate that three-view homographies provide two additional constraints, enabling the recovery of one or two focal lengths. We discuss four possible cases, including three cameras having an unknown equal focal length, three cameras having two different unknown focal lengths, three cameras where one focal length is known, and the other two cameras have equal or different unknown focal lengths. All the problems can be converted into solving polynomials in one or two unknowns, which can be efficiently solved using Sturm sequence or hidden variable technique. Evaluation using both synthetic and real data shows that the proposed solvers are both faster and more accurate than methods relying on existing two-view solvers. The code and data are available on this https URL