A few members of the recently-discovered class of ultra-long period objects have been identified as binaries with white-dwarf primaries. In most cases however, electromagnetic data are inconclusive and isolated magnetars or compact binaries remain viable. If the pulsation period matches that of the orbit though -- as is the case for ILT J1101+5521 and GLEAM-X J0704-37 -- some of these elusive radio transients could be gravitational-wave bright in the mHz band. Space-based interferometers could thus be used to provide independent constraints on their nature. We quantify the signal-to-noise ratio for the known systems, under a variety of scenarios, and show that a few could be detectable for sufficiently large chirp masses. Astrophysical implications for (non-)detections are discussed.

Light QCD axions, introduced to solve the strong CP problem, may form condensates inside neutron stars, giving rise to a novel ground state of dense matter. We investigate how such axion condensates modify the equilibrium structure and radial oscillation spectrum of NSs. Using a realistic NS model with the BSk26 equation of state, and solving the coupled Tolman-Oppenheimer-Volkoff and Klein-Gordon equations together with a linear perturbation analysis, we find two distinct families of quasi-normal modes: weakly damped fluid-dominated oscillations and highly damped axion modes. The coupling between the fluid and the axion field introduces axion-induced damping of radial oscillations, with decay timescales of order seconds for kHz axion masses. Modes with frequencies above the axion mass are strongly damped, while those below remain unaffected. These results suggest that stellar oscillations could provide a novel probe of axion properties, opening prospects for axion asteroseismology in neutron stars.

Argument mining algorithms analyze the argumentative structure of essays, making them a valuable tool for enhancing education by providing targeted feedback on the students' argumentation skills. While current methods often use encoder or encoder-decoder deep learning architectures, decoder-only models remain largely unexplored, offering a promising research direction. This paper proposes leveraging open-source, small Large Language Models (LLMs) for argument mining through few-shot prompting and fine-tuning. These models' small size and open-source nature ensure accessibility, privacy, and computational efficiency, enabling schools and educators to adopt and deploy them locally. Specifically, we perform three tasks: segmentation of student essays into arguments, classification of the arguments by type, and assessment of their quality. We empirically evaluate the models on the Feedback Prize - Predicting Effective Arguments dataset of grade 6-12 students essays and demonstrate how fine-tuned small LLMs outperform baseline methods in segmenting the essays and determining the argument types while few-shot prompting yields comparable performance to that of the baselines in assessing quality. This work highlights the educational potential of small, open-source LLMs to provide real-time, personalized feedback, enhancing independent learning and writing skills while ensuring low computational cost and privacy.

The widespread adoption of chat interfaces based on Large Language Models (LLMs) raises concerns about promoting superficial learning and undermining the development of critical thinking skills. Instead of relying on LLMs purely for retrieving factual information, this work explores their potential to foster deeper reasoning by generating critical questions that challenge unsupported or vague claims in debate interventions. This study is part of a shared task of the 12th Workshop on Argument Mining, co-located with ACL 2025, focused on automatic critical question generation. We propose a two-step framework involving two small-scale open source language models: a Questioner that generates multiple candidate questions and a Judge that selects the most relevant ones. Our system ranked first in the shared task competition, demonstrating the potential of the proposed LLM-based approach to encourage critical engagement with argumentative texts.

This study enhances the application of Physics-Informed Neural Networks (PINNs) for modeling discontinuous solutions in both hydrodynamics and relativistic hydrodynamics. Conventional PINNs, trained with partial differential equation residuals, frequently face convergence issues and lower accuracy near discontinuities. To address these issues, we build on the recently proposed locally linearized PINNs (LLPINNs), which improve shock detection by modifying the Jacobian matrix resulting from the linearization of the equations, only in regions where the velocity field exhibits strong compression. However, the original LLPINN framework required a priori knowledge of shock velocities, limiting its practical utility. We present a generalized LLPINN method that dynamically computes shock speeds using neighboring states and applies jump conditions through entropy constraints. Additionally, we introduce locally Roe PINNs (LRPINNs), which incorporate an approximate Roe Riemann solver to improve shock resolution and conservation properties across discontinuities. These methods are adapted to two-dimensional Riemann problems by using a divergence-based shock detection combined with dimensional splitting, delivering precise solutions. Compared to a high-order weighted essentially non-oscillatory solver, our method produces sharper shock transitions but smoother solutions in areas with small-scale vortex structures. Future research will aim to improve the resolution of these small-scale features without compromising the model's ability to accurately capture shocks.

Light QCD axions, introduced to solve the strong CP problem, may form condensates inside neutron stars, giving rise to a novel ground state of dense matter. We investigate how such axion condensates modify the equilibrium structure and radial oscillation spectrum of NSs. Using a realistic NS model with the BSk26 equation of state, and solving the coupled Tolman-Oppenheimer-Volkoff and Klein-Gordon equations together with a linear perturbation analysis, we find two distinct families of quasi-normal modes: weakly damped fluid-dominated oscillations and highly damped axion modes. The coupling between the fluid and the axion field introduces axion-induced damping of radial oscillations, with decay timescales of order seconds for kHz axion masses. Modes with frequencies above the axion mass are strongly damped, while those below remain unaffected. These results suggest that stellar oscillations could provide a novel probe of axion properties, opening prospects for axion asteroseismology in neutron stars.

This paper describes the submissions to the efficiency track for GPUs at the Workshop for Neural Machine Translation and Generation by members of the University of Edinburgh, Adam Mickiewicz University, Tilde and University of Alicante. We focus on efficient implementation of the recurrent deep-learning model as implemented in Amun, the fast inference engine for neural machine translation. We improve the performance with an efficient mini-batching algorithm, and by fusing the softmax operation with the k-best extraction algorithm. Submissions using Amun were first, second and third fastest in the GPU efficiency track.

Modern machine translation (MT) systems depend on large parallel corpora, often collected from the Internet. However, recent evidence indicates that (i) a substantial portion of these texts are machine-generated translations, and (ii) an overreliance on such synthetic content in training data can significantly degrade translation quality. As a result, filtering out non-human translations is becoming an essential pre-processing step in building high-quality MT systems. In this work, we propose a novel approach that directly exploits the internal representations of a surrogate multilingual MT model to distinguish between human and machine-translated sentences. Experimental results show that our method outperforms current state-of-the-art techniques, particularly for non-English language pairs, achieving gains of at least 5 percentage points of accuracy.

This paper describes the submissions to the efficiency track for GPUs at the Workshop for Neural Machine Translation and Generation by members of the University of Edinburgh, Adam Mickiewicz University, Tilde and University of Alicante. We focus on efficient implementation of the recurrent deep-learning model as implemented in Amun, the fast inference engine for neural machine translation. We improve the performance with an efficient mini-batching algorithm, and by fusing the softmax operation with the k-best extraction algorithm. Submissions using Amun were first, second and third fastest in the GPU efficiency track.

Astrophysical black holes are likely to be surrounded by various forms of matter in the form of disks or halos. While a number of studies have examined the impact of an environment on the lensing of light or gravitational waves from cosmological sources, these have, thus far, been carried out in either a Newtonian or post-Newtonian framework where the environment is superimposed on the black-hole spacetime. By using an exact solution in general relativity describing a black hole embedded within a realistic halo of Hernquist matter distribution, we study deflection angles and image amplification in a fully relativistic setup. It is shown that large ``bumps'' corresponding to the peak of the mass distribution can significantly adjust the inferences made for either the source or lens in various contexts. As an application, we consider ``echoes'' of gravitational waves, sourced by astrophysical lenses rather than being intrinsic to the compact object that produces the signal.

The existence of light QCD axions, whose mass depends on an additional free parameter, can lead to a new ground state of matter, where the sourced axion field reduces the nucleon effective mass. The presence of the axion field has structural consequences, in particular, it results in a thinner (or even prevents its existence) heat-blanketing envelope, significantly altering the cooling patterns of neutron stars. We exploit the anomalous cooling behavior to constrain previously uncharted regions of the axion parameter space by comparing model predictions with existing data from isolated neutron stars. Notably, this analysis does not require the light QCD axion to be the dark matter candidate.

The coupling between axions and photons modifies Maxwell's equations, introducing a dynamo term in the magnetic induction equation. In neutron stars, for critical values of the axion decay constant and axion mass, the magnetic dynamo mechanism increases the total magnetic energy of the star. We show that this generates substantial internal heating due to enhanced dissipation of crustal electric currents. These mechanisms would lead magnetized neutron stars to increase their magnetic energy and thermal luminosity by several orders of magnitude, in contrast to observations of thermally-emitting neutron stars. To prevent the activation of the dynamo, bounds on the allowed axion parameter space can be derived.

We present a general procedure to solve numerically the general relativistic magnetohydrodynamics (GRMHD) equations within the framework of the 3+1 formalism. The work reported here extends our previous investigation in general relativistic hydrodynamics (Banyuls et al. 1997) where magnetic fields were not considered. The GRMHD equations are written in conservative form to exploit their hyperbolic character in the solution procedure. All theoretical ingredients necessary to build up high-resolution shock-capturing schemes based on the solution of local Riemann problems (i.e. Godunov-type schemes) are described. In particular, we use a renormalized set of regular eigenvectors of the flux Jacobians of the relativistic magnetohydrodynamics equations. In addition, the paper describes a procedure based on the equivalence principle of general relativity that allows the use of Riemann solvers designed for special relativistic magnetohydrodynamics in GRMHD. Our formulation and numerical methodology are assessed by performing various test simulations recently considered by different authors. These include magnetized shock tubes, spherical accretion onto a Schwarzschild black hole, equatorial accretion onto a Kerr black hole, and magnetized thick accretion disks around a black hole prone to the magnetorotational instability.

The performance of model-based decomposition approaches rooted in the Freeman-Durden concept is an active research line in PolSAR field according to the considerable attention it has deserved along the last twenty years. Certainly, most of subsequent proposals have been driven by the only objective of getting a better qualitative balance among scattering mechanisms according to theoretical expectations. This idea is not a negative aspect per se, as has led to a more rigorous understanding of orientation effects in both urban and natural areas and hence to improved land cover classifications. However, an in-depth quantitative analysis on the output parameters is usually lacking in this topic. The attention has been mostly paid to the power of dominant contributions, whereas the accuracy and interpretation of other parameters useful for practical applications have been almost systematically overlooked. The questions that remain to be answered are: What is the actual role of all parameters describing the models? Can we assign them a consistent physical interpretation or are some of them acting just as fitting parameters? The present work aims to promote the discussion on these open issues regarding the quantitative assessment of model-based PolSAR decomposition schemes. To proceed with, we have simulated the coherency matrix according to one of existing general models and different scenarios. The inversion performance has been analysed in terms of the histograms of output parameters, standard deviation and bias. The analysis reveals that even the backscattering powers associated with all three basic scattering mechanisms are estimated with a non-negligible error higher than 10% for some cases. Despite these conclusions are subject to a particular model and inversion approach they suggest that a careful consideration of physically-based decompositions outcomes should be taken.

This report revisits the role of the extinction coefficient in radar backscattering-based models for forest monitoring. A review of a number of works dealing with this issue has revealed a diversity of extinction values being unclear its dependence on the sensor frequency and the forest type. In addition, a backscattering model directly derived from the RVoG formulation is employed to analyse the saturation of backscattering level as a function of vegetation height and the presence of a decreasing trend of backscatter beyond the saturation point as suggested in previous works in the literature. According to this analysis it seems reasonable to think that further research specially focused on dedicated experimental measurements of the extinction coefficient should be carried out.

We investigate the chiral magnetic instability in the crust of a neutron star as a potential mechanism for amplifying magnetic fields. This instability may become active when small deviations from chemical equilibrium are sustained over decades, driven by the star's gradual spin-down or residual heat loss. Our findings suggest that this mechanism can produce strong, large-scale magnetic fields consistent with models that align with observational data. Additionally, this instability naturally generates magnetic helicity in the star's crust, which is crucial for forming and maintaining strong dipolar toroidal fields, often invoked to explain magnetar observational phenomena. Our results offer a microphysically-based alternative to classical hydrodynamical dynamos for the origin of magnetar magnetic fields, addressing a long-standing debate in the field.