Search

While Large Language Models (LLMs) have shown remarkable capabilities in natural language understanding and generation, their performance often lags in lower-resource, non-English languages due to biases in the training data. In this work, we explore strategies for adapting the primarily English LLMs (Llama-2 and Llama-3) to Dutch, a language spoken by 30 million people worldwide yet often underrepresented in LLM development. We collect 104GB of Dutch text ($32$B tokens) from various sources to first apply continued pretraining using low-rank adaptation (LoRA), complemented with Dutch posttraining strategies provided by prior work. For Llama-2, we consider using (i) the tokenizer of the original model, and (ii) training a new, Dutch-specific tokenizer combined with embedding reinitialization. We evaluate our adapted models, ChocoLlama-2, both on standard benchmarks and a novel Dutch benchmark, ChocoLlama-Bench. Our results demonstrate that LoRA can effectively scale for language adaptation, and that tokenizer modification with careful weight reinitialization can improve performance. Notably, Llama-3 was released during the course of this project and, upon evaluation, demonstrated superior Dutch capabilities compared to our Dutch-adapted versions of Llama-2. We hence apply the same adaptation technique to Llama-3, using its original tokenizer. While our adaptation methods enhanced Llama-2's Dutch capabilities, we found limited gains when applying the same techniques to Llama-3. This suggests that for ever improving, multilingual foundation models, language adaptation techniques may benefit more from focusing on language-specific posttraining rather than on continued pretraining. We hope this work contributes to the broader understanding of adapting LLMs to lower-resource languages, and to the development of Dutch LLMs in particular.

We present near- and mid-infrared (IR) broadband imaging observations of the edge-on protoplanetary disk around HH 30 with the James Webb Space Telescope/Near Infrared Camera (NIRCam) and the Mid-Infrared Instrument (MIRI). We combine these observations with archival optical/near-IR scattered light images obtained with the Hubble Space Telescope (HST) and a millimeter-wavelength dust continuum image obtained with the Atacama Large Millimeter/submillimeter Array (ALMA) with the highest spatial resolution ever obtained for this target. Our multiwavelength images clearly reveal the vertical and radial segregation of micron-sized and sub-mm-sized grains in the disk. In the near- and mid-IR, the images capture not only bi-reflection nebulae separated by a dark lane but also diverse dynamical processes occurring in the HH 30 disk, such as spiral- and tail-like structures, a conical outflow, and a collimated jet. In contrast, the ALMA image reveals a flat dust disk in the disk midplane. By performing radiative transfer simulations, we show that grains of about 3 $\mu$m in radius or larger are fully vertically mixed to explain the observed mid-IR scattered light flux and its morphology, whereas millimeter-sized grains are settled into a layer with a scale height of $\gtrsim1$ au at $100$ au from the central star. We also find a tension in the disk inclination angle inferred from optical/near-IR and mm observations with the latter being closer to an exactly edge-on. Finally, we report the first detection of the proper motion of an emission knot associated with the mid-IR collimated jet detected by combining two epochs of our MIRI 12.8-$\mu$m observations., Comment: 29 pages, 23 figures; Accepted for publication in ApJ

Early Solar System (SS) planetesimals constitute the parent bodies of most meteorites investigated today. Nucleosynthetic isotope anomalies of bulk meteorites have revealed a dichotomy between non-carbonaceous (NC) and carbonaceous (CC) groups. Planetesimals sampling NC and CC isotopic signatures are conventionally thought to originate from the "dry" inner disk, and volatile-rich outer disk, respectively, with their segregation enforced by a pressure bump close to the water-ice sublimation line, possibly tied to Jupiter's formation. This framework is challenged by emerging evidence that the oldest NC planetesimals (i.e., the iron meteorites parent bodies; IMPBs) were characterized by far higher oxidation states than previously imagined, suggesting abundant ($\gtrsim$ few wt.%) liquid water in their interiors prior to core differentiation. In this paper, we employ a model for a degassing icy planetesimal (heated by $^{26}$Al decay) to map the conditions for liquid water production therein. Our work culminates in threshold characteristic sizes for pebbles composing the said planetesimal, under which water-ice melting occurs. Adopting a model for a disk evolving under both turbulence and magnetohydrodynamic disk winds, and assuming pebble growth is fragmentation-limited, we self-consistently translate the threshold pebble size to lower limits on early SS turbulence. We find that if NC IMPBs were "wet," their constituent pebbles must have been smaller than a few centimeters, corresponding to typical values of the Shakura-Sunyaev $\alpha_{\nu}$ turbulence parameter in excess of $10^{-3}$. These findings argue against a quiescent SS disk (for <10 AU), are concordant with astronomical constraints on protoplanetary disk turbulence, and suggest pebble accretion played a secondary role in building our rocky planets.

In this paper we study a quantum version of the multiparty simultaneous message-passing (SMP) model, and we show that in some cases, quantum communication can replace public randomness, even with no entanglement between the parties. This was already known for two players, but not for more than two players, and indeed, so far all that was known was a negative result. Our main technical contribution is a compiler that takes any classical public-coin simultaneous protocol based on "modified equality queries," and converts it into a quantum simultaneous protocol without public coins with roughly the same communication complexity. We then use our compiler to derive protocols for several problems, including frequency moments, neighborhood diversity, enumeration of isolated cliques, and more., Comment: 20 pages

Although Large Language Models (LLMs) have shown promising performance in healthcare-related applications, their deployment in the medical domain poses unique challenges of ethical, regulatory, and technical nature. In this study, we employ a systematic participatory approach to investigate the needs and expectations regarding clinical applications of LLMs at Lausanne University Hospital, an academic medical center in Switzerland. Having identified potential LLM use-cases in collaboration with thirty stakeholders, including clinical staff across 11 departments as well nursing and patient representatives, we assess the current feasibility of these use-cases taking into account the regulatory frameworks, data protection regulation, bias, hallucinations, and deployment constraints. This study provides a framework for a participatory approach to identifying institutional needs with respect to introducing advanced technologies into healthcare practice, and a realistic analysis of the technology readiness level of LLMs for medical applications, highlighting the issues that would need to be overcome LLMs in healthcare to be ethical, and regulatory compliant., Comment: MeurIPS GenAI for Health Workshop

We study the volume of the critical clusters for the percolation of the level sets of the Gaussian free field on metric graphs. On $\mathbb{Z}^d$ below the upper-critical dimension $d=6$, we show that the largest such cluster in a box of side length $r$ has volume of order $r^{\frac{d+2}{2}}$, as conjectured by Werner in arXiv:2002.11487. This is in contrast to the mean-field regime $d>6$, where this volume is of order $r^4$. We further obtain precise asymptotic tails for the volume of the critical cluster of the origin, and a lower bound on the tail of the volume of the near-critical cluster of the origin below the upper-critical dimension. Our proof extends to any graph with polynomial volume growth and polynomial decay of the Green's function as long as the critical one-arm probability decays as the square root of the Green's function, which is satisfied in low enough dimension., Comment: 35 pages, 1 figure

The creep behavior of an amorphous poly(etherimide) (PEI) polymer is investigated in the vicinity of its glass transition in a weakly non linear regime where the acceleration of the creep response is driven by local configurational rearrangements. From the time shifts of the creep compliance curves under stresses from 1 to 15~\si{\mega\pascal} and in the temperature range between $T_g -10K$ and $T_g$, where $T_g$ is the glass transition, we determine a macroscopic acceleration factor. The macroscopic acceleration is shown to vary as $e^{-(\Sigma/Y)^n} $ with $n=2 \pm 0.2$, where $\Sigma$ is the macroscopic stress and $Y$ is a decreasing function of compliance. Because at the beginning of creep, the stress is homogeneous, the macroscopic acceleration is thus similar to the local one, in agreement with the recent theory of Long \textit{et al.} (\textit{Phys. Rev. Mat.} (2018) \textbf{2}, 105601 ) which predicts $n=2$. For larger compliances, the decrease of the of $Y$ is interpreted as a signature of the development of stress disorder during creep., Comment: arXiv admin note: substantial text overlap with arXiv:2403.00355

We report on the successful design and fabrication of optical modules using a 50 micron pitch polymer waveguide interface, integrated for low loss, high density optical data transfer with very low space requirements on a Si photonics die. This prototype module meets JEDEC reliability standards and promises to increase the number of optical fibers that can be connected at the edge of a chip, a measure known as beachfront density, by six times compared to state of the art technology. Scalability of the polymer waveguide to less than 20 micron pitch stands to improve the bandwidth density upwards of 10 Tbps/mm.

It is essential that mesoscopic simulations of reactive systems reproduce the correct statistical distributions at thermodynamic equilibrium. By considering a compressible fluctuating hydrodynamics (FHD) simulation method of ideal gas mixtures undergoing reversible reactions described by the chemical Langevin equations, we show that thermodynamic consistency in reaction rates and the use of instantaneous temperatures for the evaluation of reaction rates is required for fluctuations for the overall system to be correct. We then formulate the required properties of a thermodynamically-consistent reaction (TCR) model. As noted in the literature, while reactions are often discussed in terms of forward and reverse rates, these rates should not be modeled independently because they must be compatible with thermodynamic equilibrium for the system. Using a simple TCR model where each chemical species has constant heat capacity, we derive the explicit condition that the forward and reverse reaction rate constants must satisfy in order for the system to be thermodynamically consistent. We perform equilibrium and non-equilibrium simulations of ideal gas mixtures undergoing a reversible dimerization reaction to measure the fluctuational behavior of the system numerically. We confirm that FHD simulations with the TCR model give the correct static structure factor of equilibrium fluctuations. For the statistically steady simulation of a gas mixture between two isothermal walls with different temperatures, we show using the TCR model that the temperature variance agrees with the corresponding thermodynamic-equilibrium temperature variance in the interior of the system, whereas noticeable deviations are present in regions near walls, where chemistry is far from equilibrium.

The modeling of the beating of cilia and flagella in fluids is a particularly active field of study, given the biological relevance of these organelles. Various mathematical models have been proposed to represent the nonlinear dynamics of flagella, whose motion is powered by the work of molecular motors attached to filaments composing the axoneme. Here, we formulate and solve a nonlinear model of activation based on the sliding feedback mechanism, capturing the chemical and configurational changes of molecular motors driving axonemal motion. This multiscale model bridges microscopic motor dynamics with macroscopic flagellar motion, providing insight into the emergence of oscillatory beating. We validate the framework through linear stability analysis and fully nonlinear numerical simulations, showing the onset of spontaneous oscillations. To make the analysis more comprehensive, we compare our approach with two established sliding feedback models.

Entangled matter displays unusual and attractive properties and mechanisms: tensile strength, capabilities for assembly and disassembly, damage tolerance. While some of the attributes and mechanisms share some traits with traditional granular materials, fewer studies have focused on entanglement and strength and there are large gaps in our understanding of the mechanics of these materials. In this report we focus on the tensile properties and mechanics of bundles made of staple-like particles, and particularly on the effect of adjusting the angle between the legs and the crown in individual staples. Our experiments, combined with discrete element models, show competing mechanisms between entanglement strength and geometric engagement between particles, giving rise to an optimum crown-leg angle that maximizes strength. We also show that tensile forces are transmitted by a small fraction of the staples, which is organized in only 1-3 force chains. The formation and breakage of these chains is highly dynamic: as force chains break, they are replaced by fresh ones which were previously mechanically invisible. Entangled matter as "granular metamaterials" offer interesting perspectives in terms of materials design, and a vast design space for individual particles. Since their properties can be tuned with the shape of the staple, we interpret these entangled materials are "granular metamaterials" with unusual combination of properties: simultaneous strength and toughness, controlled assembly and disassembly, re-conformability, recyclability., Comment: 26 pages, 11 figures, submitted to Journal of the Mechanics and Physics of Solids

Entangled matter provides intriguing perspectives in terms of deformation mechanisms, mechanical properties, assembly and disassembly. However, collective entanglement mechanisms are complex, occur over multiple length scales, and they are not fully understood to this day. In this report, we propose a simple pick-up test to measure the entanglement in staple-like particles with various leg lengths, crown-leg angles, and backbone thickness. We also present a new "throw-bounce-tangle" model based on a 3D geometrical entanglement criterion between two staples, and a Monte Carlo approach to predict the probabilities of entanglement in a bundle of staples. This relatively simple model is computationally efficient and it predicts an average density of entanglement which is consistent with the entanglement strength measured experimentally. Entanglement is very sensitive to the thickness of the backbone of the staples, even in regimes where that thickness is a small fraction (<0.04) of the other dimensions. We demonstrate an interesting use for this model to optimize staple-like particles for maximum entanglement. New designs of tunable "entangled granular metamaterials" can produce attractive combinations of strength, extensibility, and toughness that may soon outperform lightweight engineering materials such as solid foams and lattices., Comment: 27 pages, 13 figures, submitted to Granular Matter

In the last few decades, ultrafast demagnetization elicited by ultrashort laser pulses has been the subject of a large body of work that aims to better understand and control this phenomenon. Although specific magnetic materials' properties play a key role in defining ultrafast demagnetization dynamics, features of the driving laser pulse such as its duration and photon energy might also contribute. Here, we report ultrafast demagnetization of a cobalt/platinum multilayer in a broad spectral range spanning from the near-infrared (near-IR) to the mid-infrared (mid-IR), with wavelengths between 0.8 and 8.7 $\mu$m. The ultrafast dynamics of the macroscopic magnetization is tracked via the time-resolved magneto-optical Kerr effect. We show that the ultrafast demagnetization of the sample can be efficiently induced over that entire excitation spectrum with minimal dependence on the excitation wavelength. Instead, we confirm that the temporal profile of the pump excitation pulse is an important factor influencing ultrafast demagnetization dynamics.

Multimodal AI has the potential to significantly enhance document-understanding tasks, such as processing receipts, understanding workflows, extracting data from documents, and summarizing reports. Code generation tasks that require long-structured outputs can also be enhanced by multimodality. Despite this, their use in commercial applications is often limited due to limited access to training data and restrictive licensing, which hinders open access. To address these limitations, we introduce BigDocs-7.5M, a high-quality, open-access dataset comprising 7.5 million multimodal documents across 30 tasks. We use an efficient data curation process to ensure our data is high-quality and license-permissive. Our process emphasizes accountability, responsibility, and transparency through filtering rules, traceable metadata, and careful content analysis. Additionally, we introduce BigDocs-Bench, a benchmark suite with 10 novel tasks where we create datasets that reflect real-world use cases involving reasoning over Graphical User Interfaces (GUI) and code generation from images. Our experiments show that training with BigDocs-Bench improves average performance up to 25.8% over closed-source GPT-4o in document reasoning and structured output tasks such as Screenshot2HTML or Image2Latex generation. Finally, human evaluations showed a preference for outputs from models trained on BigDocs over GPT-4o. This suggests that BigDocs can help both academics and the open-source community utilize and improve AI tools to enhance multimodal capabilities and document reasoning. The project is hosted at https://bigdocs.github.io ., Comment: The project is hosted at https://bigdocs.github.io

We consider the global minimization of a particular type of minimum structured optimization problems wherein the variables must belong to some basic set, the feasible domain is described by the intersection of a large number of functional constraints and the objective stems as the pointwise minimum of a collection of functional pieces. Among others, this setting includes all non-convex piecewise linear problems. The global minimum of a minimum structured problem can be computed using a simple enumeration scheme by solving a sequence of individual problems involving each a single piece and then taking the smallest individual minimum. We propose a new algorithm, called Upper-Lower Optimization (ULO), tackling problems from the aforementioned class. Our method does not require the solution of every individual problem listed in the baseline enumeration scheme, yielding potential substantial computational gains. It alternates between (a) local-search steps, which minimize upper models, and (b) suggestion steps, which optimize lower models. We prove that ULO can return a solution of any prescribed global optimality accuracy. According to a computational complexity model based on the cost of solving the individual problems, we analyze in which practical scenarios ULO is expected to outperform the baseline enumeration scheme. Finally, we empirically validate our approach on a set of piecewise linear programs (PL) of incremental difficulty.

As of December 2024, the ARC-AGI benchmark is five years old and remains unbeaten. We believe it is currently the most important unsolved AI benchmark in the world because it seeks to measure generalization on novel tasks -- the essence of intelligence -- as opposed to skill at tasks that can be prepared for in advance. This year, we launched ARC Prize, a global competition to inspire new ideas and drive open progress towards AGI by reaching a target benchmark score of 85\%. As a result, the state-of-the-art score on the ARC-AGI private evaluation set increased from 33\% to 55.5\%, propelled by several frontier AGI reasoning techniques including deep learning-guided program synthesis and test-time training. In this paper, we survey top approaches, review new open-source implementations, discuss the limitations of the ARC-AGI-1 dataset, and share key insights gained from the competition.

Given an initial point $x_0 \in \mathbb{R}^d$ and a sequence of vectors $v_1, v_2, \dots$ in $\mathbb{R}^d$, we define a greedy sequence by setting $x_{n} = x_{n-1} \pm v_n$ where the sign is chosen so as to minimize $\|x_n\|$. We prove that if the vectors $v_i$ are chosen uniformly at random from $\mathbb{S}^{d-1}$ then elements of the sequence are, on average, approximately at distance $\|x_n\| \sim \sqrt{\pi d/8}$ from the origin. We show that the sequence $(\|x_n\|)_{n=1}^{\infty}$ has an invariant measure $\pi_d$ depending only on $d$ and we determine its mean and study its decay for all $d$. We also investigate a completely deterministic example in $d=2$ where the $v_n$ are derived from the van der Corput sequence. Several additional examples are considered.

In this work, we extend the concept of Robinson spaces to asymmetric dissimilarities, enhancing their applicability in representing and analyzing complex data. Within this generalized framework, we introduce two different problems that extend the classical seriation problem: an optimization problem and a decision problem. We establish that these problems are NP-hard and NP-complete, respectively. Despite this complexity results, we identify several non-trivial instances where these problems can be solved in polynomial time, providing valuable insights into their tractability.

The discovery of radio-quiet, X-ray thermally emitting isolated neutron stars (XINSs) in the ROSAT All-Sky Survey revealed a previously overlooked component of the neutron star population. Advancements in X-ray instrumentation and the availability of deep, wide-area optical surveys now enable us to explore XINSs at fainter X-ray fluxes and greater distances. In this study, we investigated candidates selected from the 4XMM-DR9 catalogue using XMM-Newton, focusing on long-term flux stability, spectral characterisation, and astrometry. By leveraging resources from the XMM2ATHENA project -- including updated catalogues, multiwavelength characterisation and machine learning classification -- we refined our understanding of this sample of soft X-ray emitters. Our findings enhance the characterisation of XINS candidates, laying the groundwork for more targeted investigations and future catalogue searches., Comment: 7 pages, 3 figures, proceedings of IWARA 2024, 11th International Workshop on Astronomy and Relativistic Astrophysics

We present the detection and characterization of TOI-4994b (TIC 277128619b), a warm Saturn-sized planet discovered by the NASA Transiting Exoplanet Survey Satellite (TESS). TOI-4994b transits a G-type star (V = 12.6 mag) with a mass, radius, and effective temperature of $M_{\star} =1.005^{+0.064}_{-0.061} M_{\odot}$, $R_{\star} = 1.055^{+0.040}_{-0.037} R_{\odot}$, and $T_{\rm eff} = 5640 \pm 110$ K. We obtained follow-up ground-based photometry from the Las Cumbres Observatory (LCO) and the Antarctic Search for Transiting ExoPlanets (ASTEP) telescopes, and we confirmed the planetary nature of TOI-4994b with multiple radial velocity observations from the PFS, CHIRON, HARPS, FEROS, and CORALIE instruments. From a global fit to the photometry and radial velocities, we determine that TOI-4994b is in a 21.5-day, eccentric orbit ($e = 0.32 \pm 0.04$) and has a mass of $M_{P}= 0.280^{+0.037}_{-0.034} M_{J}$, a radius of $R_{P}= 0.762^{+0.030}_{-0.027}R_{J}$, and a Saturn-like bulk density of $\rho_{p} = 0.78^{+0.16}_{-0.14}$ $\rm g/cm^3$. We find that TOI-4994 is a potentially viable candidate for follow-up stellar obliquity measurements. TOI-4994b joins the small sample of warm Saturn analogs and thus sheds light on our understanding of these rare and unique worlds., Comment: 17 pages, 10 figures. Accepted to AJ

We present the MULTIMODAL UNIVERSE, a large-scale multimodal dataset of scientific astronomical data, compiled specifically to facilitate machine learning research. Overall, the MULTIMODAL UNIVERSE contains hundreds of millions of astronomical observations, constituting 100\,TB of multi-channel and hyper-spectral images, spectra, multivariate time series, as well as a wide variety of associated scientific measurements and "metadata". In addition, we include a range of benchmark tasks representative of standard practices for machine learning methods in astrophysics. This massive dataset will enable the development of large multi-modal models specifically targeted towards scientific applications. All codes used to compile the MULTIMODAL UNIVERSE and a description of how to access the data is available at https://github.com/MultimodalUniverse/MultimodalUniverse, Comment: Accepted at NeurIPS Datasets and Benchmarks track

In this article we are interested for the numerical computation of spectra of non-self adjoint quadratic operators, in two and three spatial dimensions. Indeed, in the multidimensional case very few results are known on the location of the eignevalues. This leads to solve nonlinear eigenvalue problems. In introduction we begin with a review of theoretical results and numerical results obtained for the one dimensional case. Then we present the numerical methods developed to compute the spectra (finite difference discretization) for the two and three dimensional cases. The numerical results obtained are presented and analyzed. One difficulty here is that we have to compute eigenvalues of strongly non-self-adjoint operators which are unstable. This work is in continuity of a previous work in one spatial dimension.

OriStitch is a computational fabrication workflow to turn existing flat fabrics into self-folding 3D structures. Users turn fabrics into self-folding sheets by machine embroidering functional threads in specific patterns on fabrics, and then apply heat to deform the structure into a target 3D structure. OriStitch is compatible with a range of existing materials (e.g., leather, woven fabric, and denim). We present the design of specific embroidered hinges that fully close under exposure to heat. We discuss the stitch pattern design, thread and fabric selection, and heating conditions. To allow users to create 3D textiles using our hinges, we create a tool to convert 3D meshes to 2D stitch patterns automatically, as well as an end-to-end fabrication and actuation workflow. To validate this workflow, we designed and fabricated a cap (303 hinges), a handbag (338 hinges), and a cover for an organically shaped vase (140 hinges). In technical evaluation, we found that our tool successfully converted 23/28 models (textures and volumetric objects) found in related papers. We also demonstrate the folding performance across different materials (suede leather, cork, Neoprene, and felt).

We introduce a novel property of bounded Voronoi tessellations that enables cycle-free mesh sweeping algorithms. We prove that a topological sort of the dual graph of any Voronoi tessellation is feasible in any flow direction and dimension, allowing straightforward application to discontinuous Galerkin (DG) discretisations of first-order hyperbolic partial differential equations and the Boltzmann Transport Equation (BTE) without requiring flux-cycle corrections. We also present an efficient algorithm to perform the topological sort on the dual mesh nodes, ensuring a valid sweep ordering. This result expands the applicability of DG methods for transport problems on polytopal meshes by providing a robust framework for scalable, parallelised solutions. To illustrate its effectiveness, we conduct a series of computational experiments showcasing a DG scheme for BTE, demonstrating both computational efficiency and adaptability to complex geometries., Comment: 19 pages, 10 figures

In practice, low Earth orbit (LEO) and medium Earth orbit (MEO) satellite networks consist of multiple orbits, each populated with many satellites. A widely used spatial architecture for satellites is the Walker constellation, where the longitudes of orbits are equally spaced and the satellites are periodically distributed along the orbits. In this paper, we develop a stochastic geometry model for Walker constellations by constructing orbits that are invariant with respect to (w.r.t.) Earth's spin and satellites that are invariant w.r.t. orbit-wise rotation. This model enables an analysis based on dynamical systems, which allows one to address structural properties such as periodicity and ergodicity. It also enables a stochastic geometry analysis where, for a typical user at a given latitude, we derive the performance of downlink communications as a function of key constellation parameters, including orbit inclination and altitude, the number of orbits, the number of satellites, and the user latitude., Comment: submitted to IEEE Journal

We demonstrate the advantages of THz frequency combs for high-resolution spectroscopy. This benefits from wide spectral coverage and the exact knowledge of the frequency position of each comb component. Heterodyne detection combined with a fast Fourier spectrometer enables rapid and simultaneous measurement of more than 80 frequency comb modes covering a 7.5 GHz bandwidth. A spectrum is obtained in under 20 minutes yielding a uniform resolution of 70 kHz. This new setup has been validated by recording more than 150 lines of methanol around 723 GHz, and represents a new solution to exploit THz frequency combs for high-resolution spectroscopy.

Epilepsy is known to drastically alter brain dynamics during seizures (ictal periods), but its effects on background (non-ictal) brain dynamics remain poorly understood. To investigate this, we analyzed an in-house dataset of brain activity recordings from epileptic zebrafish, focusing on two controlled genetic conditions across two fishlines. After using machine learning to segment and label recordings, we applied time-delay embedding and Persistent Homology -- a noise-robust method from Topological Data Analysis (TDA) -- to uncover topological patterns in brain activity. We find that ictal and non-ictal periods can be distinguished based on the topology of their dynamics, independent of genetic condition or fishline, which validates our approach. Remarkably, within a single wild-type fishline, we identified topological differences in non-ictal periods between seizure-prone and seizure-free individuals. These findings suggest that epilepsy leaves detectable topological signatures in brain dynamics even outside of ictal periods. Overall, this study demonstrates the utility of TDA as a quantitative framework to screen for topological markers of epileptic susceptibility, with potential applications across species.

Unconventional supersymmetry is a proposal, introduced by Alvarez-Valenzuela-Zanelli (AVZ), aiming to use the framework of supersymmetric field theory to describe fermionic matter fields and bosonic gauge fields in a unified way, as both parts of a single superconnection. It hinges upon the so-called matter ansatz. Unfortunately, the formal and conceptual status of the ansatz remained unclear, preventing unconventional supersymmetry to be used in a principled way as a general approach beyond the model in which it was first considered. In this letter, we lift this restriction by showing that the ansatz is a special case of the Dressing Field Method, a new systematic tool to exhibit the gauge-invariant content of general-relativistic gauge field theories., Comment: V2, 8 pages, revised version

Machine learning based surrogate models offer researchers powerful tools for accelerating simulation-based workflows. However, as standard datasets in this space often cover small classes of physical behavior, it can be difficult to evaluate the efficacy of new approaches. To address this gap, we introduce the Well: a large-scale collection of datasets containing numerical simulations of a wide variety of spatiotemporal physical systems. The Well draws from domain experts and numerical software developers to provide 15TB of data across 16 datasets covering diverse domains such as biological systems, fluid dynamics, acoustic scattering, as well as magneto-hydrodynamic simulations of extra-galactic fluids or supernova explosions. These datasets can be used individually or as part of a broader benchmark suite. To facilitate usage of the Well, we provide a unified PyTorch interface for training and evaluating models. We demonstrate the function of this library by introducing example baselines that highlight the new challenges posed by the complex dynamics of the Well. The code and data is available at https://github.com/PolymathicAI/the_well., Comment: 38th Conference on Neural Information Processing Systems (NeurIPS 2024) Track on Datasets and Benchmarks

Causal inference in observational studies has advanced significantly since Rosenbaum and Rubin introduced propensity score methods. Inverse probability of treatment weighting (IPTW) is widely used to handle confounding bias. However, newer methods, such as energy balancing (EB), kernel optimal matching (KOM), and covariate balancing propensity score by tailored loss function (TLF), offer model-free or non-parametric alternatives. Despite these developments, guidance remains limited in selecting the most suitable method for treatment effect estimation in practical applications. This study compares IPTW with EB, KOM, and TLF, focusing on their ability to estimate treatment effects since this is the primary objective in many applications. Monte Carlo simulations are used to assess the ability of these balancing methods combined with different estimators to evaluate average treatment effect. We compare these methods across a range of scenarios varying sample size, level of confusion, and proportion of treated. In our simulation, we observe no significant advantages in using EB, KOM, or TLF methods over IPTW. Moreover, these recent methods make obtaining confidence intervals with nominal coverage difficult. We also compare the methods on the PROBITsim dataset and get results similar to those of our simulations., Comment: 20 pages, 3 figures, 1 tables. Comments are welcome

Progress in our knowledge of tumor mechanisms and complexity led to the understanding of the physical parameters of cancer cells and their microenvironment, including the mechanical, thermal, and electrical properties, solid stress, and liquid pressure, as critical regulators of tumor progression and potential prognostic traits associated with clinical outcomes. The biological hallmarks of cancer and physical abnormalities of tumors are mutually reinforced, promoting a vicious cycle of tumor progression. A comprehensive analysis of the biological and physical tumor parameters is critical for developing more robust prognostic and diagnostic markers and improving treatment efficiency. Like the biological tumor traits, physical tumor features are characterized by inter-and intratumoral heterogeneity. The dynamic changes of physical tumor traits during tumor progression and as a result of tumor treatment highlight the necessity of their spatial and temporal analysis in clinical settings. This review focuses on the biological basis of the tumor-specific physical traits, the state-of-the-art methods of their analyses, and the perspective of clinical translation. The importance of tumor physical parameters for disease progression and therapy resistance, as well as current treatment strategies to monitor and target tumor physical traits in clinics, is highlighted., Comment: Cancer Heterogeneity and Plasticity, 2024

Common functional languages incentivize tail-recursive functions, as opposed to general recursive functions that consume stack space and may not scale to large inputs. This distinction occasionally requires writing functions in a tail-recursive style that may be more complex and slower than the natural, non-tail-recursive definition. This work describes our implementation of the *tail modulo constructor* (TMC) transformation in the OCaml compiler, an optimization that provides stack-efficiency for a larger class of functions -- tail-recursive *modulo constructors* -- which includes in particular the natural definition of `List.map` and many similar recursive data-constructing functions. We prove the correctness of this program transformation in a simplified setting -- a small untyped calculus -- that captures the salient aspects of the OCaml implementation. Our proof is mechanized in the Coq proof assistant, using the Iris base logic. An independent contribution of our work is an extension of the Simuliris approach to define simulation relations that support different calling conventions. To our knowledge, this is the first use of Simuliris to prove the correctness of a compiler transformation., Comment: Published at POPL 2025

Recent work has shown that diffusion models can be used as powerful neural rendering engines that can be leveraged for inserting virtual objects into images. Unlike typical physics-based renderers, however, neural rendering engines are limited by the lack of manual control over the lighting setup, which is often essential for improving or personalizing the desired image outcome. In this paper, we show that precise lighting control can be achieved for object relighting simply by specifying the desired shadows of the object. Rather surprisingly, we show that injecting only the shadow of the object into a pre-trained diffusion-based neural renderer enables it to accurately shade the object according to the desired light position, while properly harmonizing the object (and its shadow) within the target background image. Our method, SpotLight, leverages existing neural rendering approaches and achieves controllable relighting results with no additional training. Specifically, we demonstrate its use with two neural renderers from the recent literature. We show that SpotLight achieves superior object compositing results, both quantitatively and perceptually, as confirmed by a user study, outperforming existing diffusion-based models specifically designed for relighting., Comment: Project page: https://lvsn.github.io/spotlight

Training speech separation models in the supervised setting raises a permutation problem: finding the best assignation between the model predictions and the ground truth separated signals. This inherently ambiguous task is customarily solved using Permutation Invariant Training (PIT). In this article, we instead consider using the Multiple Choice Learning (MCL) framework, which was originally introduced to tackle ambiguous tasks. We demonstrate experimentally on the popular WSJ0-mix and LibriMix benchmarks that MCL matches the performances of PIT, while being computationally advantageous. This opens the door to a promising research direction, as MCL can be naturally extended to handle a variable number of speakers, or to tackle speech separation in the unsupervised setting.

In this paper, we propose a new approach -- the Tempered Finite Element Method (TFEM) -- that extends the Finite Element Method (FEM) to classes of meshes that include zero-measure or nearly degenerate elements for which standard FEM approaches do not allow convergence. First, we review why the maximum angle condition [2] is not necessary for FEM convergence and what are the real limitations in terms of meshes. Next, we propose a simple modification of the classical FEM for elliptic problems that provably allows convergence for a wider class of meshes including bands of caps that cause locking of the solution in standard FEM formulations. The proposed method is trivial to implement in an existing FEM code and can be theoretically analyzed. We prove that in the case of exactly zero-measure elements it corresponds to mortaring. We show numerically and theoretically that what we propose is functional and sound. The remainder of the paper is devoted to extensions of the TFEM method to linear elasticity, mortaring of non-conforming meshes, high-order elements, and advection.

Adversarial perturbations aim to deceive neural networks into predicting inaccurate results. For visual object trackers, adversarial attacks have been developed to generate perturbations by manipulating the outputs. However, transformer trackers predict a specific bounding box instead of an object candidate list, which limits the applicability of many existing attack scenarios. To address this issue, we present a novel white-box approach to attack visual object trackers with transformer backbones using only one bounding box. From the tracker predicted bounding box, we generate a list of adversarial bounding boxes and compute the adversarial loss for those bounding boxes. Experimental results demonstrate that our simple yet effective attack outperforms existing attacks against several robust transformer trackers, including TransT-M, ROMTrack, and MixFormer, on popular benchmark tracking datasets such as GOT-10k, UAV123, and VOT2022STS., Comment: Accepted in The 3rd New Frontiers in Adversarial Machine Learning (AdvML Frontiers @NeurIPS2024)

In a cosmological context, the Einstein-Gauss-Bonnet theory contains, in $d+4$ dimensions, a dynamical compactification scenario in which the additional dimensions settle down to a configuration with a constant radion/scale factor. Sadly however this work demonstrates that such a quite appealing framework is plagued by instabilities, either from the background configuration's unsteadiness or the ghostly behaviors of the tensorial perturbations. New and stable solutions are found by relaxing one of the hypotheses defining the original compactification scenario. However, such configurations do not respect the current bounds on the speed of propagation of gravitational waves, and thus have to be discarded. Those results thus advocate for a comprehensive study of compactification scenarios in the Gauss-Bonnet framework, their stability, and the effects of matter inclusion., Comment: 19 pages, uses RevTeX

One of the challenge that face the effectiveness of Polyvinyl Chloride geomembranes (PVC GMs) as a hydraulic barrier is the capacity to withstand unexpected mechanical actions, particularly tensile forces, during installation and throughout their lifespan. These forces pose risks of premature failure and impermeability degradation. In this study, the characterization of the short and long-term mechanical response of PVC GMs to uniaxial tensile forces has been investigated. Uniaxial tensile test have been performed for tensile rates spanning several orders of magnitude. Analysis of the true stress-strain curves reveals a significant decrease in tensile modulus, strength, and strain at failure at low strain rates, which are relatively close to those applied in situ. Long-term investigations have been conducted as well, through relaxation tests. Our key results unveil two distinct characteristic times in stress relaxation, with the fast relaxation occurring over the first four hours. During this phase, the pre-relaxation loading rate affects the relaxation behavior. Beyond this phase, the relaxation behavior becomes independent from the pre-relaxation loading rate. Burger's rheological model is proposed to measure the stress relaxation at different rates. The model's results validate the existence of two characteristic times.

This paper introduces a scenario where a maneuverable satellite in geostationary orbit (GEO) conducts on-orbit attacks, targeting communication between a GEO satellite and a ground station, with the ability to switch between stationary and time-variant jamming modes. We propose a machine learning-based detection approach, employing the random forest algorithm with principal component analysis (PCA) to enhance detection accuracy in the stationary model. At the same time, an adaptive threshold-based technique is implemented for the time-variant model to detect dynamic jamming events effectively. Our methodology emphasizes the need for the use of orbital dynamics in integrating physical constraints from satellite dynamics to improve model robustness and detection accuracy. Simulation results highlight the effectiveness of PCA in enhancing the performance of the stationary model, while the adaptive thresholding method achieves high accuracy in detecting jamming in the time-variant scenario. This approach provides a robust solution for mitigating the evolving threats to satellite communication in GEO environments.

Objective: Direct cortical responses (DCR) and axono-cortical evoked potentials (ACEP) are generated by electrically stimulating the cortex either directly or indirectly through white matter pathways, potentially leading to different electrogenic processes. For ACEP, the slow conduction velocity of axons (median around 4 m.s$^{-1}$) is anticipated to induce a delay. For DCR, direct electrical stimulation (DES) of the cortex is expected to elicit additional cortical activity involving smaller and slower non-myelinated axons. We tried to validate these hypotheses. Methods: DES was administered either directly on the cortex or to white matter fascicles within the resection cavity, while recording DCR or ACEP at the cortical level in nine patients. Results: Short but significant delays (around 2 ms) were measurable for ACEP immediately following the initial component (around 7 ms). Subsequent activities (around 40 ms) exhibited notable differences between DCR and ACEP, suggesting the presence of additional cortical activities for DCR. Conclusion: Distinctions between ACEPs and DCRs can be made based on a delay at the onset of early components and the dissimilarity in the shape of the later components >40 ms after the DES artifact). Significance: The comparison of different types of evoked potentials allows to better understand the effects of DES., Comment: Clinical Neurophysiology, 2024

Context. The magneto-thermal instability (MTI) is one of many possible drivers of stratified turbulence in the intracluster medium (ICM) outskirts of galaxy clusters, where the background temperature gradient is aligned with the gravity. This instability occurs because of the fast anisotropic conduction of heat along magnetic field lines; but to what extent it impacts the ICM dynamics, energetics and overall equilibrium is still a matter of debate. Aims. This work aims at understanding MTI turbulence in an astrophysically stratified ICM atmosphere, its saturation mechanism, and its ability to carry energy and to provide non-thermal pressure support. Methods. We perform a series of 2D and 3D numerical simulations of the MTI in global spherical models of stratified ICM, thanks to the finite-volume code IDEFIX, using Braginskii-magnetohydrodynamics. We use volume-, shell-averaged and spectral diagnostics to study the saturation mechanism of the MTI, and its radial transport energy budget. Results. The MTI is found to saturate through a dominant balance between injection and dissipation of available potential energy, which amounts to marginalising the Braginskii heat flux but not the background temperature gradient itself. Accordingly, the strength and injection length of MTI-driven turbulence exhibit clear dependencies on the thermal diffusivity. The MTI drives cluster-size motions with Mach numbers up to $\mathcal{M} \sim 0.3$, even in presence of strong stable entropy stratification. We show that such mildly compressible flows can provide about $\sim 15\%$ of non-thermal pressure support in the outermost ICM regions, and that the convective transport itself is much less efficient than conduction at radially transporting energy. Finally, we show that the MTI saturation can be described by a diffusive mixing-length theory, shedding light on the diffusive buoyant nature of the instability., Comment: 20 pages, 8 figures, accepted for publication in A&A, arXiv abstract slightly shortened

Inspired by a paper by T. Chakradhar, K. Gittins, G. Habib and N. Peyerimhoff, we analyze their conjecture that the ground state energy of the magnetic Dirichlet-to-Neumann operator on the disk tends to $+\infty$ as the magnetic field tends to $+\infty$. This is an important step towards the analysis of the curvature effect in the case of general domains in $\mathbb R^2$., Comment: 25 pages

Task-aware robotic grasping is a challenging problem that requires the integration of semantic understanding and geometric reasoning. Traditional grasp planning approaches focus on stable or feasible grasps, often disregarding the specific tasks the robot needs to accomplish. This paper proposes a novel framework that leverages Large Language Models (LLMs) and Quality Diversity (QD) algorithms to enable zero-shot task-conditioned grasp selection. The framework segments objects into meaningful subparts and labels each subpart semantically, creating structured representations that can be used to prompt an LLM. By coupling semantic and geometric representations of an object's structure, the LLM's knowledge about tasks and which parts to grasp can be applied in the physical world. The QD-generated grasp archive provides a diverse set of grasps, allowing us to select the most suitable grasp based on the task. We evaluate the proposed method on a subset of the YCB dataset, where a Franka Emika robot is assigned to perform various actions based on object-specific task requirements. We created a ground truth by conducting a survey with six participants to determine the best grasp region for each task-object combination according to human intuition. The model was evaluated on 12 different objects across 4--7 object-specific tasks, achieving a weighted intersection over union (IoU) of 76.4% when compared to the survey data., Comment: 8 pages, 5 figures

Regular unsupervised domain adaptive person re-identification (ReID) focuses on adapting a model from a source domain to a fixed target domain. However, an adapted ReID model can hardly retain previously-acquired knowledge and generalize to unseen data. In this paper, we propose a Dual-level Joint Adaptation and Anti-forgetting (DJAA) framework, which incrementally adapts a model to new domains without forgetting source domain and each adapted target domain. We explore the possibility of using prototype and instance-level consistency to mitigate the forgetting during the adaptation. Specifically, we store a small number of representative image samples and corresponding cluster prototypes in a memory buffer, which is updated at each adaptation step. With the buffered images and prototypes, we regularize the image-to-image similarity and image-to-prototype similarity to rehearse old knowledge. After the multi-step adaptation, the model is tested on all seen domains and several unseen domains to validate the generalization ability of our method. Extensive experiments demonstrate that our proposed method significantly improves the anti-forgetting, generalization and backward-compatible ability of an unsupervised person ReID model., Comment: Accepted to TPAMI

Bandgaps in layered materials are critical for enabling functionalities such as tunable photodetection, efficient energy conversion, and nonlinear optical responses, which are essential for next-generation photonic and quantum devices. Gap engineering could form heterostructures with complementary materials like transition metal dichalcogenides or perovskites for multi-functional devices. Graphite, conventionally regarded as a gapless material, exhibits a bandgap of ~100 meV in nano-scale patterned highly oriented pyrolytic graphite (HOPG), as revealed by angle-resolved photoemission spectroscopy (ARPES) and Raman measurements. Our state-of-the-art calculations, incorporating photoemission matrix element effects, predict this bandgap with remarkable accuracy and attribute it to mechanical distortions introduced during patterning. This work bridges theory and experiment, providing the direct evidence of a tunable bandgap in HOPG. Beyond its fundamental significance, this finding opens new possibilities for designing materials with tailored electronic properties, enabling advancements in terahertz devices and optoelectronics., Comment: 33 pages, 9 figures

The discovery of in-plane anisotropic excitons in two-dimensional layered semiconductors enables state-of-the-art nanophotonic applications. A fundamental yet unknown parameter of these quasiparticles is the coherence time (T_2 ), which governs the quantum dephasing timescale, over which the coherent superposition of excitons can be maintained and manipulated. Here, we report the direct measurement of T_2 within the sub-picosecond range, along with multiple population decay timescales (T_1 ) at resonance for anisotropic excitons in pristine layered rhenium disulfide (ReS2). We observe a notable weak dependence on layer thickness for T_2 , and a quasi-independence for T_1 . The excitonic coherence in few-layer ReS2 exhibits exceptional robustness against optical density and temperature compared to other two-dimensional semiconductors, enabling quantum features even at room temperature. No photon echo fingerprints were observed in pristine ReS2, highlighting the homogeneous character of the anisotropic excitonic transitions and a particularly low level of disorder in exfoliated flakes. Lastly, our results for mono- to bulk-like ReS2 support a direct gap band structure regardless their layer thickness, addressing the ongoing discussion about its nature.

We expand the theoretical framework by Gottlieb el al. (2023), which connects binary merger populations with long and short binary gamma-ray bursts (lbGRBs and sbGRBs), incorporating kilonovae as a key diagnostic tool. We show that lbGRBs, powered by massive accretion disks around black holes (BHs), should be accompanied by bright, red kilonovae. In contrast, sbGRBs - if also powered by BHs - would produce fainter, red kilonovae, potentially biasing against their detection. However, magnetized hypermassive neutron star (HMNS) remnants that precede BH formation can produce jets with power ($P_{\rm NS} \approx 10^{51}\,{\rm erg\,s^{-1}}$) and Lorentz factor ($\Gamma>10$), likely compatible with sbGRB observations, and would result in distinctly bluer kilonovae, offering a pathway to identifying the sbGRB central engine. Recent modeling by Rastinejad et al. (2024) found luminous red kilonovae consistently accompany lbGRBs, supporting lbGRB originating from BH-massive disk systems, likely following a short-lived HMNS phase. The preferential association of sbGRBs with comparably luminous kilonovae argues against the BH engine hypothesis for sbGRBs, while the bluer hue of these KNe provides additional support for an HMNS-driven mechanism. Within this framework, BH-NS mergers likely contribute exclusively to the lbGRB population with red kilonovae. Our findings suggest that GW170817 may, in fact, have been an lbGRB to on-axis observers. Finally, we discuss major challenges faced by alternative lbGRB progenitor models, such as white dwarf-NS or white dwarf-BH mergers and accretion-induced collapse forming magnetars, which fail to align with observed GRB timescales, energies, and kilonova properties.

In this work, we propose a new approach based on positron annihilation spectroscopy to estimate the concentration of vacancy-type defects induced by self-ion irradiation in tungsten at room temperature, 500, and 700{\deg}C. Using experimental and Two-component density functional theory calculated annihilation characteristics of various vacancy clusters V$_{n}$ ($n$=1-65) and a positron trapping model associated with the simulated annealing algorithm, vacancy cluster concentration distribution could be extracted from experimental data. The method was validated against simulation results for room-temperature irradiation and transmission electron microscopy observations for higher temperatures. After irradiation at 500 and 700{\deg}C, small clusters (<20 vacancies, ~0.85 nm) undetectable by TEM were unveiled, with concentrations exceeding 10$^{25}$ m$^{-3}$, significantly higher than the concentration of TEM-visible defects (10$^{24}$ m$^{-3}$). Moreover, incorporating an oxygen-vacancy complex is deemed necessary to accurately replicate experimental data in samples subjected to high-temperature irradiation.

The universal dependence of hadron suppression, $R_{\rm{AA}}(p_\perp)$, observed at large-$p_\perp$ in heavy ion collisions at RHIC and LHC allows for a systematic determination of the average parton energy loss $\langle \epsilon \rangle$ in quark-gluon plasma (QGP). A simple relation between $\langle \epsilon \rangle$ and the soft particle multiplicity allows for probing the dependence of parton energy loss on the medium path-length. We find that all the available measurements are consistent with $\langle \epsilon \rangle \propto L^\beta$ with $\beta=1.02\pm^{0.09}_{0.06}$, consistent with the pQCD expectation of parton energy loss in a longitudinally expanding QGP. We then show, based on the model predictions, that the data on the azimuthal anisotropy coefficient divided by the collision eccentricity, $v_2/\rm{e}$, follows the same scaling property as $R_{\rm{AA}}$. Finally, a linear relationship between $v_2/\rm{e}$ and the logarithmic derivative of $R_{\rm{AA}}$ at large $p_\perp$ offers a purely data-driven access to the $L$ dependence of parton energy loss. Quite remarkably, both hadron and jet measurements obey this latter relationship, moreover with consistent values of $\beta$. This points to the same parametric path-length dependence of parton and jet energy loss in QGP., Comment: 6 pages, 4 figures. Talk given at the 42nd International Conference on High Energy Physics (ICHEP2024)

The Bethe-Salpeter equation has been extensively employed to compute the two-body electron-hole propagator and its poles which correspond to the neutral excitation energies of the system. Through a different time-ordering, the two-body Green's function can also describe the propagation of two electrons or two holes. The corresponding poles are the double ionization potentials and double electron affinities of the system. In this work, a Bethe-Salpeter equation for the two-body particle-particle propagator is derived within the linear-response formalism using a pairing field and anomalous propagators. This framework allows us to compute kernels corresponding to different self-energy approximations ($GW$, $T$-matrix, and second-Born) as in the usual electron-hole case. The performance of these various kernels is gauged for singlet and triplet valence double ionization potentials using a set of 23 small molecules. The description of double core hole states is also analyzed., Comment: 19 pages, 8 figures