Search

We present the first extraction of transverse-momentum-dependent distributions of unpolarised quarks from experimental Drell-Yan data using neural networks to parametrise their nonperturbative part. We show that neural networks outperform traditional parametrisations providing a more accurate description of data. This work establishes the feasibility of using neural networks to explore the multi-dimensional partonic structure of hadrons and paves the way for more accurate determinations based on machine-learning techniques.

In ecology, the description of species composition and biodiversity calls for statistical methods that involve estimating features of interest in unobserved samples based on an observed one. In the last decade, the Bayesian nonparametrics literature has thoroughly investigated the case where data arise from a homogeneous population. In this work, we propose a novel framework to address heterogeneous populations, specifically dealing with scenarios where data arise from two areas. This setting significantly increases the mathematical complexity of the problem and, as a consequence, it received limited attention in the literature. While early approaches leverage on computational methods, we provide a distributional theory for the in-sample analysis of any observed sample and we enable out-of-sample prediction for the number of unseen distinct and shared species in additional samples of arbitrary sizes. The latter also extends the frequentist estimators which solely deal with the one-step ahead prediction. Furthermore, our results can be applied to address the sample size determination in sampling problems aimed at detecting shared species. Our results are illustrated in a real-world dataset concerning a population of ants in the city of Trieste.

Persistent excitation (PE) is a necessary and sufficient condition for uniform exponential parameter convergence in several adaptive, identification, and learning schemes. In this article, we consider, in the context of multi-input linear time-invariant (LTI) systems, the problem of guaranteeing PE of commonly-used regressors by applying a sufficiently rich (SR) input signal. Exploiting the analogies between time shifts and time derivatives, we state simple necessary and sufficient PE conditions for the discrete- and continuous-time frameworks. Moreover, we characterize the shape of the set of SR input signals for both single-input and multi-input systems. Finally, we show with a numerical example that the derived conditions are tight and cannot be improved without including additional knowledge of the considered LTI system.

Full QCD+QED simulations allow to evaluate isospin breaking corrections to hadron masses. With the openQxD code, we are able to perform these simulations employing C-periodic boundary conditions, implemented through a doubling of the physical lattice along one spatial direction. The use of these boundary conditions introduces non-zero Wick contractions between two quark or two antiquark fields, that, in the case of the computation of baryon masses, lead to partially connected additional contributions that we expect to vanish in the infinite volume limit. These contributions are challenging because they involve an all-to-all propagator connecting one point in the physical lattice and one in the mirror lattice. We present a way to compute these corrections to the $\Omega^-$ baryon mass using a combination of point and stochastic source inversions. This work is part of the program of the RC* collaboration.

Federated Learning (FL) emerged as a decentralized paradigm to train models while preserving privacy. However, conventional FL struggles with data heterogeneity and class imbalance, which degrade model performance. Clustered FL balances personalization and decentralized training by grouping clients with analogous data distributions, enabling improved accuracy while adhering to privacy constraints. This approach effectively mitigates the adverse impact of heterogeneity in FL. In this work, we propose a novel clustered FL method, FedGWC (Federated Gaussian Weighting Clustering), which groups clients based on their data distribution, allowing training of a more robust and personalized model on the identified clusters. FedGWC identifies homogeneous clusters by transforming individual empirical losses to model client interactions with a Gaussian reward mechanism. Additionally, we introduce the Wasserstein Adjusted Score, a new clustering metric for FL to evaluate cluster cohesion with respect to the individual class distribution. Our experiments on benchmark datasets show that FedGWC outperforms existing FL algorithms in cluster quality and classification accuracy, validating the efficacy of our approach.

We simulate the electronic and transport properties of metal/two-dimensional material/metal vertical heterostructures, with a focus on graphene, hexagonal boron nitride and two phases of molybdenum diselenide. Using density functional theory and non-equilibrium Green's function, we assess how stacking configurations and material thickness impact important properties, such as density of states, potential barriers and conductivity. For monolayers, strong orbital hybridization with the metallic electrodes significantly alters the electronic characteristics, with the formation of states within the gap of the semiconducting 2D materials. Trilayers reveal the critical role of interlayer coupling, where the middle layer retains its intrinsic properties, thus influencing the overall conductivity. Our findings highlight the potential for customized multilayer designs to optimize electronic device performance based on two-dimensional materials., Comment: 16 pages, 6 figures

Temporal metamaterials are artificial materials whose electromagnetic properties change over time. In analogy with spatial media and metamaterials, where their properties change smoothly or abruptly over space, temporal metamaterials can exhibit a smooth variation over time, realizing a temporal non-homogeneous medium, or a stepwise transition, realizing the temporal version of dielectric slabs or multilayer structures. In this Letter, we focus our attention on temporal multilayer structures, and we propose the synthesis of higher-order transfer functions by modeling the wave propagation through a generalized temporal multilayer structure, consisting of a cascade over time of different media. The tailoring of the scattering response of temporal structure as a function of frequency is presented, deriving the corresponding scattering coefficients for a properly designed set of medium properties, i.e., permittivity and permeability, and application time, in analogy with what is typically done in optical and electromagnetic spatial multilayered structures. This allows us to design novel electromagnetic and optical devices with higher-order transfer functions by exploiting the temporal dimension instead of the spatial one.

3D-printed bellow soft pneumatic arms are widely adopted for their flexible design, ease of fabrication, and large deformation capabilities. However, their low stiffness limits their real-world applications. Although several methods exist to enhance the stiffness of soft actuators, many involve complex manufacturing processes not in line with modern goals of monolithic and automated additive manufacturing. With its simplicity, bead-jamming represents a simple and effective solution to these challenges. This work introduces a method for monolithic printing of a bellow soft pneumatic arm, integrating a tendon-driven central spine of bowl-shaped beads. We experimentally characterized the arm's range of motion in both unjammed and jammed states, as well as its stiffness under various actuation and jamming conditions. As a result, we provide an optimal jamming policy as a trade-off between preserving the range of motion and maximizing stiffness. The proposed design was further demonstrated in a switch-toggling task, showing its potential for practical applications., Comment: 6 pages, 8 figures, accepted by the 8th IEEE-RAS International Conference on Soft Robotics, RoboSoft 2025

Low Mode Average (LMA) is a technique to improve the quality of the signal-to-noise ratio in the long time separation of Euclidean correlation functions. We report on its beneficial impact in computing the vector-vector light connected two-point correlation functions and derived physical quantities in the mixed action lattice setup adopted by ETM collaboration. We focus on preliminary results of the computation within isospin symmetric QCD (isoQCD) of the $R$-ratio smeared with Gaussian kernels of widths down to $\sigma\sim250$ MeV, which is enough to appreciate the $\rho$ resonance around 770 MeV, using the Hansen-Lupo-Tantatlo (HLT) spectral-density reconstruction method., Comment: 12 pages, 5 figures

Sea-quark isospin-breaking effects (IBE) are difficult to compute since they require the evaluation of all-to-all propagators. However, the quest for high-precision calculations motivates a detailed study of these contributions. There are strong arguments that the stochastic error associated with these quantities should diverge in the continuum and infinite-volume limit, resulting in a possible bottleneck for the method. In this work, we present the study of the error scaling for these quantities using $N_f=3$ $O(a)$-improved Wilson fermions QCD with C-periodic boundary conditions in space, a pion mass $M_{\pi}=400$ MeV, a range of lattice spacings $a=0.05, 0.075, 0.1$ fm, and volumes $L=1.6, 2.4, 3.2$ fm. The analysis of the error as a function of the number of stochastic sources shows that we reach the gauge error for the dominant contributions. The errors do not show the leading order divergence $1/a$ for strong-IBE and $1/a^2$ for electromagnetic IBE, in the considered range of lattice spacings. On the other hand, our data are consistent with the predicted leading divergence $\sqrt{V}$., Comment: 9 pages, 8 figures, RC* collaboration, Contribution to the 41st International Symposium on Lattice Field Theory, 28th July-3rd August 2024, University of Liverpool, United Kingdom. Corrected the bibliography, the abstract and equation 2

The use of 5G New Radio (NR) spectrum around 26 GHz is currently raising the quest on its compatibility with the well-established Earth Exploration-Satellite Service (EESS), which may be blinded by the spurious radiation emitted Above-the-Horizon (AtH) by Base Station (BS) antennas. Indeed, AtH grating lobes are often present during cell scanning due to the large inter-element spacing in BS array antennas for achieving higher gains with a reduced number of RF chains. In this letter, we propose an approach based on an electrically thin metasurface-based dome for the reduction of AtH grating lobes in 5G-NR BS antennas. The proposed scanning range shifting approach exploits the natural lower amplitude of the grating lobes when the antenna array scans in an angular region closer to the broadside direction. The grating lobe reduction is here demonstrated considering a 1x4 phased linear antenna array operating under dual-liner 45deg-slant polarization. A simple design procedure for designing the metasurface dome is reported, together with the antenna performances, evaluated through a proper set of numerical experiments. It is shown that the grating lobe radiation towards the satellite region is significantly reduced, whereas the overall insertion loss is moderate.

The experimental evidence of purely reactive loads impedance matching is here provided by exploiting the special scattering response under complex excitations. The study starts with a theoretical analysis of the reflection properties of an arbitrary reactive load and identifies the proper excitation able to transform the purely reactive load into a virtual resistive load during the time the signal is applied. To minimize reflections between the load and the transmission line, the excitation must have a complex frequency, leading to a propagating signal with a tailored temporal envelope. The aim of this work is to design and, for the first time,experimentally demonstrate this anomalous scattering behavior in microwave circuits, showing that the time-modulated signals can be exploited as a new degree of freedom for achieving impedance matching without introducing neither a matching network nor resistive elements, that are typically used for ensuring power dissipation and, thus, zero reflection. The proposed matching strategy does not alter the reactive load that is still lossless, enabling an anomalous termination condition where the energy is not dissipated nor reflected, but indefinitely accumulated in the reactive load. The stored energy leaks out the load as soon as the applied signal changes or stops.

Quantum communication rates in terrestrial quantum networks are fundamentally limited by fibre loss, even in the presence of quantum repeaters. Satellites offer a solution for long-distance communication, with the most commonly explored scenario involving prepare-and-measure protocols connecting from orbit to a trusted-node ground station via free-space down-links. In contrast, up-link scenarios allow for entanglement to be distributed between a satellite and remote end users in terrestrial networks, eliminating any trust requirement on the ground station. Here we demonstrate an ultra-bright source of far-non-degenerate entangled photons and perform quantum key distribution in emulated high-loss satellite scenarios. With a loss profile corresponding to that of one of the pioneering Micius up-link experiments, and a terrestrial end user separated by 10~km of telecom fibre we achieve secure key bit accumulation of 5.2~kbit in a single emulated overpass in the asymptotic limit. Our results confirm the viability of upcoming low-Earth orbit receiver satellite missions.

The rapid evolution of large language models (LLMs) has transformed conversational agents, enabling complex human-machine interactions. However, evaluation frameworks often focus on single tasks, failing to capture the dynamic nature of multi-turn dialogues. This paper introduces a dynamic benchmarking framework to assess LLM-based conversational agents through interactions with synthetic users. The framework integrates generative agent simulation to evaluate performance on key dimensions: information extraction, context awareness, and adaptive engagement. By simulating various aspects of user behavior, our work provides a scalable, automated, and flexible benchmarking approach. Experimental evaluation - within a loan application use case - demonstrates the framework's effectiveness under one-shot and few-shot extraction conditions. Results show that adaptive strategies improve data extraction accuracy, especially when handling ambiguous responses. Future work will extend its applicability to broader domains and incorporate additional metrics (e.g., conversational coherence, user engagement). This study contributes a structured, scalable approach to evaluating LLM-based conversational agents, facilitating real-world deployment.

This work presents the computation of real corrections to the impact factor for forward Higgs boson production, preserving the full dependence on the top-quark mass. The results are shown to align with the BFKL factorization framework, particularly in reproducing the expected rapidity divergence. Additionally, the subtraction of this divergence has been demonstrated using the appropriate counterterm within the BFKL scheme. In the infinite-top-mass limit, our findings reproduce the previously established result., Comment: 6 pages, 2 figures. Presented by Gabriele Gatto at "Diffraction and Low-$x$ 2024'', Trabia (Palermo, Italy), September 8-14, 2024

Recent advances in machine learning have opened new avenues for optimizing detector designs in high-energy physics, where the complex interplay of geometry, materials, and physics processes has traditionally posed a significant challenge. In this work, we introduce the $\textit{end-to-end}$ optimization framework AIDO that leverages a diffusion model as a surrogate for the full simulation and reconstruction chain, enabling gradient-based design exploration in both continuous and discrete parameter spaces. Although this framework is applicable to a broad range of detectors, we illustrate its power using the specific example of a sampling calorimeter, focusing on charged pions and photons as representative incident particles. Our results demonstrate that the diffusion model effectively captures critical performance metrics for calorimeter design, guiding the automatic search for layer arrangement and material composition that aligns with known calorimeter principles. The success of this proof-of-concept study provides a foundation for future applications of end-to-end optimization to more complex detector systems, offering a promising path toward systematically exploring the vast design space in next-generation experiments., Comment: 14 pages, 9 figures, submitted to MDPI particles

The increasing demand for electronics is driving shorter development cycles for application-specific integrated circuits (ASICs). To meet these constraints, hardware designers emphasize reusability and modularity of IP blocks, leveraging standard system-on-chip (SoC) architectures with integrated processors and common interconnects. While these architectures reduce design and verification efforts, they also introduce complexity, as verification must encompass both hardware and software execution. To enhance reusability, hardware IP blocks are often described in higher-abstraction-level languages such as Chisel and SystemRDL, relying on compilers to generate Verilog for RTL simulation and implementation. At the system level, SoC modeling and verification leverage C++ and SystemC, underscoring the need for software compilation. Consequently, an effective build system must support both hardware design flows and software compilation, including cross-compilation for C++, C, and assembly. Existing hardware build systems lack sufficient support for software compilation, necessitating the development of a new solution. In response, the Microelectronics section of CERN initiated SoCMake, initially as part of the System-on-Chip Radiation Tolerant Ecosystem (SOCRATES). Designed to automate the generation of fault-tolerant RISC-V SoCs for high-energy physics environments, SoCMake has since evolved into a generic open-source build tool for SoC generation., Comment: Presented at the 2024 Workshop on Open-Source EDA Technology (https://woset-workshop.github.io/)

We discuss minimal covariant quantum space-time ${\cal M}^{1,3}_0$, which is defined through the minimal doubleton representation of $\mathfrak{so}(4,2)$. An elementary definition in terms of generators and relations is given. This space is shown to admit a semi-classical interpretation as quantized twistor space ${\mathbb C} P^{1,2}$, viewed as a quantized $S^2$-bundle over a 3+1-dimensional $k=-1$ FLRW space-time. In particular we find an over-complete set of (quasi-) coherent states, with a large hierarchy between the uncertainty scale and the geometric curvature scale. This provides an interesting background for the IKKT model, leading to a $\mathfrak{hs}$-extended gravitational gauge theory, which is free of ghosts due to the constraints on phase space arising from the doubleton representation., Comment: 29 pages, 5 figures

Markov Chain Monte Carlo simulations of lattice Quantum Chromodynamics (QCD) are the only known tool to investigate non-perturbatively the theory of the strong interaction and are required to perform precision tests of the Standard Model of Particle Physics. As the Markov Chain is a serial process, the sole option for improving the sampling rate is accelerating each individual update step. Heterogeneous clusters of GPU-accelerated nodes offer large total memory bandwidth which can be used to speed-up our application, openQxD-1.1, which is dominated by inversions of the Dirac operator, a large sparse matrix. In this work we investigate offloading the inversion to GPU using the lattice-QCD library QUDA, and our early results demonstrate a significant potential speed-up in the time-to-solution for state-of-the-art problem sizes. Minimal extensions to the existing QUDA library are required for our specific physics programme while greatly enhancing the performance portability of our code and retaining the reliability and robustness of existing applications in openQxD-1.1. Our new interface will enable us to utilize pre-exascale infrastructure and reduce the systematic uncertainty in our physics predictions by incorporating the effects of quantum electromagnetism (QED) in our simulations.

An approach to improve neural network interpretability is via clusterability, i.e., splitting a model into disjoint clusters that can be studied independently. We define a measure for clusterability and show that pre-trained models form highly enmeshed clusters via spectral graph clustering. We thus train models to be more modular using a "clusterability loss" function that encourages the formation of non-interacting clusters. Using automated interpretability techniques, we show that our method can help train models that are more modular and learn different, disjoint, and smaller circuits. We investigate CNNs trained on MNIST and CIFAR, small transformers trained on modular addition, and language models. Our approach provides a promising direction for training neural networks that learn simpler functions and are easier to interpret., Comment: 8 pages, under review. arXiv admin note: text overlap with arXiv:2409.15747 (author note: this is an extension of that workshop paper but has different authors)

Linear recurrent neural networks enable powerful long-range sequence modeling with constant memory usage and time-per-token during inference. These architectures hold promise for streaming applications at the edge, but deployment in resource-constrained environments requires hardware-aware optimizations to minimize latency and energy consumption. Unstructured sparsity offers a compelling solution, enabling substantial reductions in compute and memory requirements--when accelerated by compatible hardware platforms. In this paper, we conduct a scaling study to investigate the Pareto front of performance and efficiency across inference compute budgets. We find that highly sparse linear RNNs consistently achieve better efficiency-performance trade-offs than dense baselines, with 2x less compute and 36% less memory at iso-accuracy. Our models achieve state-of-the-art results on a real-time streaming task for audio denoising. By quantizing our sparse models to fixed-point arithmetic and deploying them on the Intel Loihi 2 neuromorphic chip for real-time processing, we translate model compression into tangible gains of 42x lower latency and 149x lower energy consumption compared to a dense model on an edge GPU. Our findings showcase the transformative potential of unstructured sparsity, paving the way for highly efficient recurrent neural networks in real-world, resource-constrained environments., Comment: Under review

Thanks to their compliance and adaptability, soft robots can be deployed to perform tasks in constrained or complex environments. In these scenarios, spatial awareness of the surroundings and the ability to localize the robot within the environment represent key aspects. While state-of-the-art localization techniques are well-explored in autonomous vehicles and walking robots, they rely on data retrieved with lidar or depth sensors which are bulky and thus difficult to integrate into small soft robots. Recent developments in miniaturized Time of Flight (ToF) sensors show promise as a small and lightweight alternative to bulky sensors. These sensors can be potentially distributed on the soft robot body, providing multi-point depth data of the surroundings. However, the small spatial resolution and the noisy measurements pose a challenge to the success of state-of-the-art localization algorithms, which are generally applied to much denser and more reliable measurements. In this paper, we enforce distributed VL53L5CX ToF sensors, mount them on the tip of a soft robot, and investigate their usage for self-localization tasks. Experimental results show that the soft robot can effectively be localized with respect to a known map, with an error comparable to the uncertainty on the measures provided by the miniaturized ToF sensors., Comment: The manuscript has been accepted as a contributed paper for the 8th IEEE-RAS International Conference on Soft Robotics (RoboSoft 2025)

Bipartite graphs, representing two-mode networks, arise in many research fields. These networks have two disjoint node sets representing distinct entity types, for example persons and groups, with edges representing associations between the two entity types. In bipartite graphs, the smallest possible cycle is a cycle of length four, and hence four-cycles are the smallest structure to model closure in such networks. Exponential-family random graph models (ERGMs) are a widely used model for social, and other, networks, including specifically bipartite networks. Existing ERGM terms to model four-cycles in bipartite networks, however, are relatively rarely used. In this work we demonstrate some problems with these existing terms to model four-cycles, and define new ERGM terms to help overcome these problems. The position of the new terms in the ERGM dependence hierarchy, and their interpretation, is discussed. The new terms are demonstrated in simulation experiments, and their application illustrated with ERGM models of empirical networks ranging in size from hundreds of nodes to hundreds of thousands of nodes., Comment: 54 pages including appendices

We extend the recently introduced Clifford dressed Time-Dependent Variational Principle (TDVP) to efficiently compute many-body wavefunction amplitudes in the computational basis. This advancement enhances the study of Loschmidt echoes, which generally require accurate calculations of the overlap between the evolved state and the initial wavefunction. By incorporating Clifford disentangling gates during TDVP evolution, our method effectively controls entanglement growth while keeping the computation of these amplitudes accessible. Specifically, it reduces the problem to evaluating the overlap between a Matrix Product State (MPS) and a stabilizer state, a task that remains computationally feasible within the proposed framework. To demonstrate the effectiveness of this approach, we first benchmark it on the one-dimensional transverse-field Ising model. We then apply it to more challenging scenarios, including a non-integrable next-to-nearest-neighbor Ising chain and a two-dimensional Ising model. Our results highlight the versatility and efficiency of the Clifford-augmented MPS, showcasing its capability to go beyond the evaluation of simple expectation values. This makes it a powerful tool for exploring various aspects of many-body quantum dynamics., Comment: 7 pages, 5 figures

Polar materials with optical phonons in the meV range are excellent candidates for both dark matter direct detection (via dark photon-mediated scattering) and light dark matter absorption. In this study, we propose, for the first time, the metal halide perovskites MAPbI$_3$, MAPbCl$_3$, and CsPbI$_3$ for these purposes. Our findings reveal that CsPbI$_3$ is the best material, significantly improving exclusion limits compared to other polar materials. For scattering, CsPbI$_3$ can probe dark matter masses down to the keV range. For absorption, it enhances sensitivity to detect dark photon masses below $\sim 10~{\rm meV}$. The only material which has so far been investigated and that could provide competitive bounds is CsI, which, however, is challenging to grow in kilogram-scale sizes due to its considerably lower stability compared to CsPbI$_3$. Moreover, CsI is isotropic while the anisotropic structure of CsPbI$_3$ enables daily modulation analysis, showing that a significant percentage of daily modulation exceeding 1% is achievable for dark matter masses below $40~{\rm keV}$.

We derive the mathematical condition under which the physical solution of the many-electron problem, obtained by self-consistent approaches, becomes unstable upon increasing interaction strength. The validity of our criterion is explicitly verified by performing self-consistent calculations of basic interacting models. In this context, we eventually unveiled the precise connection linking the misleading convergence of self-consistent schemes to the multivaluedness of the Luttinger-Ward functional as well as to the divergences of the irreducible vertex function. Our analysis also explains how a misleading convergence of self-consistent approximations can occur even in parameter regions without vertex divergences. Even more importantly, it allows us to define a general procedure for stabilizing the physical solution, when it is unstable in conventional self-consistent schemes.

A longstanding open question in sub-Riemannian geometry is the smoothness of (the arc-length parameterization of) length-minimizing curves. In [6], this question is negative answered, with an example of a $C^2$ but not $C^3$ length-minimizer of a real-analytic (even polynomial) sub-Riemannian structure. In this paper, we study a class of examples of sub-Riemannian structures that generalizes that presented in [6], and we prove that length-minimizing curves must be at least of class $C^2$ within these examples. In particular, we prove that Theorem 1.1 in [6] is sharp.

We characterise the scale-by-scale transfers of energy, enstrophy and helicity in homogeneous and isotropic polymeric turbulence using direct numerical simulations. The microscale Reynolds number is set to $Re_\lambda \approx 460$, and the Deborah number $De = \tau_p/\tau_f$ is varied between $1/9 \le De \le 9$; $\tau_p$ is the polymeric relaxation time and $\tau_f$ is the turnover time of the largest scales of the flow. The study relies on the exact scale-by-scale budget equations (derived from the the governing model equations) for energy, enstrophy and helicity, which account for the back-reaction of the polymers on the flow. Polymers act as a sink/source in the flow, and provide alternative routes for the scale-by-scale transfers of the three quantities, whose relevance changes with $De$. We find that polymers deplete the nonlinear energy cascade mainly at smaller scales, by weakening both the extreme forward as well as reverse local events. The new polymer-driven energy flux dominates at small scales for $De \ge 1$, and on average transfers energy from larger to smaller scales with localised backscatter events. Polymers weaken the stretching of vorticity with the enstrophy being mainly generated by the fluid-polymer interaction, especially when $De \ge 1$. Accordingly, an inspection of the small-scale flow topology shows that polymers favour events with two-dimensional state of straining, and promote/inhibit extreme extension/rotation events: in polymeric turbulence shear and planar extensional flows are more probable. The helicity injected at the largest scales shows a similar transfer process to as energy, being mainly driven by the nonlinear cascade at large scales and by the polymer-driven flux at small scales. Polymers are found to favour events that break the small-scale mirror symmetry, with the relative helicity monotonically increasing with $De$ at all scales.

At the core of the popular Transformer architecture is the self-attention mechanism, which dynamically assigns softmax weights to each input token so that the model can focus on the most salient information. However, the softmax structure slows down the attention computation due to its row-wise nature, and inherently introduces competition among tokens: as the weight assigned to one token increases, the weights of others decrease. This competitive dynamic may narrow the focus of self-attention to a limited set of features, potentially overlooking other informative characteristics. Recent experimental studies have shown that using the element-wise sigmoid function helps eliminate token competition and reduce the computational overhead. Despite these promising empirical results, a rigorous comparison between sigmoid and softmax self-attention mechanisms remains absent in the literature. This paper closes this gap by theoretically demonstrating that sigmoid self-attention is more sample-efficient than its softmax counterpart. Toward that goal, we illustrate that each row of the self-attention matrix can be represented as a mixture of experts. Our analysis shows that ''experts'' in sigmoid self-attention require significantly less data to achieve the same approximation error as those in softmax self-attention. We corroborate our theoretical findings through extensive experiments on both synthetic and real-world datasets., Comment: Fanqi Yan, Huy Nguyen contributed equally to this work. 51 pages, 2 figures, 3 tables

The improved sensitivity of interferometric facilities to the 21-cm line of atomic hydrogen (HI) enables studies of its properties in galaxies beyond the local Universe. In this work, we perform a 21 cm line spectral stacking analysis combining the MIGHTEE and CHILES surveys in the COSMOS field to derive a robust HI-stellar mass relation at z=0.36. In particular, by stacking thousands of star-forming galaxies subdivided into stellar mass bins, we optimize the signal-to-noise ratio of targets and derive mean HI masses in the different stellar mass intervals for the investigated galaxy population. We combine spectra from the two surveys, estimate HI masses, and derive the scaling relation log10(MHI) = (0.32 +- 0.04)log10(M*) + (6.65 +- 0.36). Our findings indicate that galaxies at z=0.36 are HI richer than those at z=0, but HI poorer than those at z=1, with a slope consistent across redshift, suggesting that stellar mass does not significantly affect HI exchange mechanisms. We also observe a slower growth rate HI relative to the molecular gas, supporting the idea that the accretion of cold gas is slower than the rate of consumption of molecular gas to form stars. This study contributes to understanding the role of atomic gas in galaxy evolution and sets the stage for future development of the field in the upcoming SKA era., Comment: 23 pages, 16 figures, accepted for publication in ApJ

Noninterference theory supports the analysis of secure computations in multi-level security systems. Classical equivalence-based approaches to noninterference mainly rely on bisimilarity. In a nondeterministic setting, assessing noninterference through weak bisimilarity is adequate for irreversible systems, whereas for reversible ones branching bisimilarity has been recently proven to be more appropriate. In this paper we address the same two families of systems, with the difference that probabilities come into play in addition to nondeterminism. For irreversible systems we extend the results of Aldini, Bravetti, and Gorrieri developed in a generative-reactive probabilistic setting, while for reversible systems we extend the results of Esposito, Aldini, Bernardo, and Rossi developed in a purely nondeterministic setting. We recast noninterference properties by adopting probabilistic variants of weak and branching bisimilarities for irreversible and reversible systems respectively. Then we investigate a taxonomy of those properties as well as their preservation and compositionality aspects, along with a comparison with the nondeterministic taxonomy. The adequacy of the extended noninterference theory is illustrated via a probabilistic smart contract example., Comment: arXiv admin note: text overlap with arXiv:2311.15670

We present a pilot study on extracting the form factors of the semileptonic decay of a $ B_s $ meson to the $P$-wave $ D_s^{**} $ states from $B_s$ four-point correlators. With their inclusive nature, four-point correlators include contributions from all possible final states. From the extracted $ P $-wave form factors, we obtain numerical results for the corresponding Isgur-Wise form factors. The results suggest significant contributions from radial excitations to the Uraltsev sum rule at zero-recoil. In this pilot study, a coarse lattice of $ 24^3\times 64 $ with lattice spacing of $0.11\,\mathrm{fm}$ is used for the analysis., Comment: 10 pages, 3 figures, Contribution to the 41st International Symposium on Lattice Field Theory (LATTICE2024), 28 July - 3 August 2024, Liverpool, UK

The advent of Large Language Models (LLMs) has significantly advanced the field of automated code generation. LLMs rely on large and diverse datasets to learn syntax, semantics, and usage patterns of programming languages. For low-resource languages (i.e., niche programming languages characterized by the scarcity of training data), the limited availability of such data hampers the models' ability to generalize effectively, resulting in poorer code generation performance as compared to high-resource languages. For this reason, there is a quest for techniques able to close this performance gap. We present an empirical study investigating the effectiveness of several approaches for boosting LLMs' performance on low-resource languages, namely: (i) a classic fine-tuning, which is however capped in size by the scarcity of training data; (ii) three variants of in-context learning, with prompts crafted to provide the LLM with additional information about the low-resource language (e.g., few-shot examples showcasing features of the targeted language); and (iii) a pre-training objective teaching the model how to translate between high- and low-resource languages. The context of our study are two low-resource languages (R and Racket) and six LLMs having different architectures and sizes. Our findings reveal that a fine-tuning is usually the best choice for smaller LLMs, possibly due to the fact that even a small dataset is sufficient to train their limited number of parameters. With the increase in size of the models, in-context learning becomes more and more effective, representing a safe and cheap bet (i.e., it always helps, but with different magnitudes). Differently, very large LLMs may deteriorate their performance on low-resource languages when fine-tuning is performed, possibly due to the lack of enough data needed to effectively update their weights., Comment: Accepted at ICPC'25

A longstanding open question in sub-Riemannian geometry is the following: are sub-Riemannian length minimizers smooth? We give a negative answer to this question, exhibiting an example of a $C^2$ but not $C^3$ length-minimizer of a real-analytic (even polynomial) sub-Riemannian structure.

This study focuses on the application of deep geometric models to solve the 3x3x3 Rubik's Cube. We begin by discussing the cube's graph representation and defining distance as the model's optimization objective. The distance approximation task is reformulated as a node classification problem, effectively addressed using Graph Neural Networks (GNNs). After training the model on a random subgraph, the predicted classes are used to construct a heuristic for $A^*$ search. We conclude with experiments comparing our heuristic to that of the DeepCubeA model.

We consider a reproducing kernel Hilbert space of discrete entire functions on the square lattice $\mathbb Z^2$ inspired by the classical Paley-Wiener space of entire functions of exponential growth in the complex plane. For such space we provide a Paley-Wiener type characterization and a sampling result., Comment: 24 pages

The destruction of metallicity due to the mutual Coulomb interaction of quasiparticles gives rise to fascinating phenomena of solid state physics such as the Mott metal-insulator transition and the pseudogap. A key observable characterizing their occurrences is the single-particle spectral function, determined by the fermionic self-energy. In this paper we investigate in detail how real space fluctuations are responsible for a self-energy that drives the Mott-Hubbard metal-insulator transition. To this aim we first introduce a real space fluctuation diagnostics approach to the Hedin equation, which connects the fermion-boson coupling vertex $\lambda$ to the self-energy $\Sigma$. Second, by using cellular dynamical mean-field theory calculations for $\lambda$ we identify the leading physical processes responsible for the destruction of metallicity across the transition., Comment: 21 pages, 17 figures

Artificial intelligence has significantly advanced skin cancer diagnosis by enabling rapid and accurate detection of malignant lesions. In this domain, most publicly available image datasets consist of single, isolated skin lesions positioned at the center of the image. While these lesion-centric datasets have been fundamental for developing diagnostic algorithms, they lack the context of the surrounding skin, which is critical for improving lesion detection. The iToBoS dataset was created to address this challenge. It includes 16,954 images of skin regions from 100 participants, captured using 3D total body photography. Each image roughly corresponds to a $7 \times 9$ cm section of skin with all suspicious lesions annotated using bounding boxes. Additionally, the dataset provides metadata such as anatomical location, age group, and sun damage score for each image. This dataset aims to facilitate training and benchmarking of algorithms, with the goal of enabling early detection of skin cancer and deployment of this technology in non-clinical environments., Comment: Article Submitted to Scientific Data

We present an iterative approach for planning and controlling motions of underactuated robots with uncertain dynamics. At its core, there is a learning process which estimates the perturbations induced by the model uncertainty on the active and passive degrees of freedom. The generic iteration of the algorithm makes use of the learned data in both the planning phase, which is based on optimization, and the control phase, where partial feedback linearization of the active dofs is performed on the model updated on-line. The performance of the proposed approach is shown by comparative simulations and experiments on a Pendubot executing various types of swing-up maneuvers. Very few iterations are typically needed to generate dynamically feasible trajectories and the tracking control that guarantees their accurate execution, even in the presence of large model uncertainties.

Federated testbeds enable collaborative research by providing access to diverse resources, including computing power, storage, and specialized hardware like GPUs, programmable switches and smart Network Interface Cards (NICs). Efficiently sharing these resources across federated institutions is challenging, particularly when resources are scarce and costly. GPUs are crucial for AI and machine learning research, but their high demand and expense make efficient management essential. Similarly, advanced experimentation on programmable data plane requires very expensive programmable switches (e.g., based on P4) and smart NICs. This paper introduces SHARY (SHaring Any Resource made easY), a dynamic reservation system that simplifies resource booking and management in federated environments. We show that SHARY can be adopted for heterogenous resources, thanks to an adaptation layer tailored for the specific resource considered. Indeed, it can be integrated with FIGO (Federated Infrastructure for GPU Orchestration), which enhances GPU availability through a demand-driven sharing model. By enabling real-time resource sharing and a flexible booking system, FIGO improves access to GPUs, reduces costs, and accelerates research progress. SHARY can be also integrated with SUP4RNET platform to reserve the access of P4 switches., Comment: Extended version of a paper accepted to NOMS 2025 conference as a short paper

Supermassive black holes (SMBH) are thought to grow through accretion of matter and mergers. Models of SMBH mergers have long suffered the final parsec problem, where SMBH binaries may stall before energy loss from gravitational waves (GW) becomes significant, leaving the pair unmerged. Direct evidence of coalesced SMBH remains elusive. Theory predicts that GW recoiling black holes can occur following a black hole merger. Here we present decisive spectroscopic evidence that the gas bound to the SMBH in the spatially offset quasar 3C 186 is blue-shifted relative to the host galaxy. This is exclusively explained by the GW recoil super-kick scenario. This confirmation of the ejection process implies that the final parsec problem is resolved in nature, providing evidence that even the most massive black holes can merge., Comment: 31 pages, 10 figures, submitted

In this work we report Nernst effect measurements in single crystal bismuth samples, with special emphasis on the characterization of the Nernst coefficient when the magnetic field, heat current and generated voltage are aligned along specific directions relative to the crystal axes. We found significant differences between the different orientations, reflecting the highly anisotropic electronic structure of bismuth and compatible with the Nernst characteristics obtained from polycrystalline samples. These results not only complement the experimental works published in the past but also underline the role of crystalline orientation in the context of transverse thermoelectric effects, towards an efficient design of thermomagnetic devices like the ordinary-Nernst-effect-based energy harvesters., Comment: 11 pages, 5 figures

In safety-critical SoC applications such as automotive and aerospace, reliable transaction monitoring is crucial for maintaining system integrity. This paper introduces a drop-in Transaction Monitoring Unit (TMU) for AXI4 subordinate endpoints that detects transaction failures including protocol violations or timeouts and triggers recovery by resetting the affected subordinates. Two TMU variants address different constraints: a Tiny-Counter solution for tightly area-constrained systems and a Full-Counter solution for critical subordinates in mixed-criticality SoCs. The Tiny-Counter employs a single counter per outstanding transaction, while the Full-Counter uses multiple counters to track distinct transaction stages, offering finer-grained monitoring and reducing detection latencies by up to hundreds of cycles at roughly 2.5x the area cost. The Full-Counter also provides detailed error logs for performance and bottleneck analysis. Evaluations at both IP and system levels confirm the TMU's effectiveness and low overhead. In GF12 technology, monitoring 16-32 outstanding transactions occupies 1330-2616 um2 for the Tiny-Counter and 3452-6787 um2 for the Full-Counter; moderate prescaler steps reduce these figures by 18-39% and 19-32%, respectively, with no loss of functionality. Results from a full-system integration demonstrate the TMU's robust and precise monitoring capabilities in safety-critical SoC environments., Comment: 7 pages, 11 figures, accepted as a regular paper at DATE25

Quantum computing with superconducting circuits relies on high-fidelity driven nonlinear processes. An ideal quantum nonlinearity would selectively activate desired coherent processes at high strength, without activating parasitic mixing products or introducing additional decoherence. The wide bandwidth of the Josephson nonlinearity makes this difficult, with undesired drive-induced transitions and decoherence limiting qubit readout, gates, couplers and amplifiers. Significant strides have been recently made into building better `quantum mixers', with promise being shown by Kerr-free three-wave mixers that suppress driven frequency shifts, and balanced quantum mixers that explicitly forbid a significant fraction of parasitic processes. We propose a novel mixer that combines both these strengths, with engineered selection rules that make it essentially linear (not just Kerr-free) when idle, and activate clean parametric processes even when driven at high strength. Further, its ideal Hamiltonian is simple to analyze analytically, and we show that this ideal behavior is first-order insensitive to dominant experimental imperfections. We expect this mixer to allow significant advances in high-Q control, readout, and amplification., Comment: 23 pages (8 Main + 15 Appdx), 9 figures (3 Main + 6 Appdx)

Surgical guidance can be delivered in various ways. In neurosurgery, spatial guidance and orientation are predominantly achieved through neuronavigation systems that reference pre-operative MRI scans. Recently, there has been growing interest in providing live guidance by analyzing video feeds from tools such as endoscopes. Existing approaches, including anatomy detection, orientation feedback, phase recognition, and visual question-answering, primarily focus on aiding surgeons in assessing the current surgical scene. This work aims to provide guidance on a finer scale, aiming to provide guidance by forecasting the trajectory of the surgical instrument, essentially addressing the question of what to do next. To address this task, we propose a model that not only leverages the historical locations of surgical instruments but also integrates anatomical features. Importantly, our work does not rely on explicit ground truth labels for instrument trajectories. Instead, the ground truth is generated by a detection model trained to detect both anatomical structures and instruments within surgical videos of a comprehensive dataset containing pituitary surgery videos. By analyzing the interaction between anatomy and instrument movements in these videos and forecasting future instrument movements, we show that anatomical features are a valuable asset in addressing this challenging task. To the best of our knowledge, this work is the first attempt to address this task for manually operated surgeries.

Comets, asteroids, and other small bodies are thought to be remnants of the original planetesimal population of the Solar System. As such, their physical, chemical, and isotopic properties hold crucial details on how and where they formed and how they evolved. Yet, placing precise constraints on the formation region of these bodies has been challenging. Data from spacecraft missions have a particularly high potential of addressing the question of the origin of the visited bodies. ESA's Rosetta mission to comet 67P/Churyumov-Gerasimenko returned data from the comet for two years on its journey around the Sun. This extensive data set has revolutionised our view on comets and still holds unsolved problems. We use the Rosetta/ROSINA (Rosetta Orbiter Spectrometer for Ion and Neutral Analysis) measurement of the volatile/ice composition and the Rosetta/COSIMA (COmetary Secondary Ion Mass Analyzer) measurements of the refractory composition of comet 67P. These measurements are combined using a Monte Carlo method. The refractory-to-ice ratio is a free parameter that is constrained a posteriori. Using only the composition, we constrain the refractory-to-ice ratio to $0.5<\chi<1.7$, and derive the bulk elemental abundances for 67P of H, C, N, O, Na, Mg, Al, S, K, Ar, Ca, Cr, Mn, Fe, Kr, and Xe. We find the noble gas xenon in near solar elemental abundance in comet 67P. Krypton is slightly depleted, while argon is heavily depleted. Comet 67P is enriched in all three noble gases by up to 2.5 orders of magnitude compared to CI chondrites. We show this is consistent with a formation region between 25 and 35 au in a protoplanetary disk region with temperatures between 30 and 40 K and with the trapping of dust for a long time in rings of the protoplanetary disk., Comment: 11 pages, accepted for publication in Planetary and Space Science

The development of color centers in silicon enables scalable quantum technologies by combining telecom-wavelength emission and compatibility with mature silicon fabrication. However, large-scale integration requires precise control of each emitter's optical transition to generate indistinguishable photons for quantum networking. Here, we demonstrate a foundry-fabricated photonic integrated circuit (PIC) combining suspended silicon waveguides with a microelectromechanical (MEMS) cantilever to apply local strain and spectrally tune individual G-centers. Applying up to 35 V between the cantilever and the substrate induces a reversible wavelength shift of the zero-phonon line exceeding 100 pm, with no loss in brightness. Moreover, by modeling the strain-induced shifts with a `digital twin' physical model, we achieve vertical localization of color centers with sub-3 nm vertical resolution, directly correlating their spatial position, dipole orientation, and spectral behavior. This method enables on-demand, low-power control of emission spectrum and nanoscale localization of color centers, advancing quantum networks on a foundry-compatible platform., Comment: 30 pages, 28 figures

Color centers are among the most promising candidates for quantum information processing. Central requirements for their practical applications include controlled and efficient local activation in nanophotonic devices and identical spectral features. However, producing color centers in a controlled and reliable way is inherently challenging due to the lack of comprehensive theoretical insights into their formation and the difficulty of streamlining the generation process for rapid in-situ optimization. We address these challenges by developing an end-to-end first-principles model that captures the underlying formation process of color centers. Emitters are activated through laser annealing, which allows for in-situ creation and the possibility of model-based control. Notably, our model enables the estimation of the emitters' inhomogeneous broadening down to 16 GHz in bare silicon, which translates into the creation of emitters with highly similar spectral properties. Finally, we address the challenge of in-situ deterministic activation of color centers in nanophotonic devices by going beyond bare silicon and demonstrating successful laser writing in photonic crystal optical cavities. These results lay the foundation for deterministic and large-scale integration of color centers within quantum photonic platforms., Comment: 17 pages, 10 figures, 1 table

Combining multiple events into population analyses is a cornerstone of gravitational-wave astronomy. A critical component of such studies is the assumed population model, which can range from astrophysically motivated functional forms to non-parametric treatments that are flexible but difficult to interpret. In practice, the current approach is to fit the data multiple times with different population models to identify robust features. We propose an alternative strategy: assuming the data have already been fit with a flexible model, we present a practical recipe to reconstruct the population distribution of a different model. As our procedure postprocesses existing results, it avoids the need to access the underlying gravitational-wave data again and handle selection effects. Additionally, our reconstruction metric provides a goodness-of-fit measure to compare multiple models. We apply this method to the mass distribution of black-hole binaries detected by LIGO/Virgo/KAGRA. Our work paves the way for streamlined gravitational-wave population analyses by fitting the data once and for all with advanced non-parametric methods and careful handling of selection effects, while the astrophysical interpretation is then made accessible using our reconstruction procedure on targeted models. The key principle is that of conceptually separating data description from data interpretation., Comment: 9 pages, 4 figures

We present the third phase of the largest high-frequency, high-resolution imaging survey of 231 nearby, hard X-ray selected AGN, with a very high $98 \pm 1\%$ detection fraction. This survey presents VLA 22 GHz radio observations with 1" spatial resolution covering over $6$ orders of magnitude in radio luminosity in nearby AGN that span $\sim4$ orders of magnitude in black hole mass and X-ray luminosity. We identify three different radio morphologies: $44 \pm 3\%$ (102/231) are compact or unresolved, $46 \pm 3\%$ (106/231) show an extended structure (star formation, possible one-sided jets, etc.), and $8 \pm 2\%$ (19/231) have a biconical or two-sided jet-like morphology. The remaining $2 \pm 1\%$ (4/231) sources are non-detections. The radio-to-X-ray luminosity ratios of the Swift-BAT AGN ($\text{L}_R/\text{L}_{14-195 \text{keV}} \sim 10^{-5.5}$ and $\text{L}_R/\text{L}_{2-10 \text{keV}} \sim 10^{-5}$) with a scatter of $\sim0.5$ dex are similar to that of coronally active stars ($\text{L}_R/\text{L}_X \sim 10^{-5}$). For most targets, extended emission in radio-quiet objects is broadly consistent with the expectation for star formation from previous FIR observations, once the contribution from the radio core has been subtracted. Our sample represents nearby analogs of distant AGN at the peak of black hole growth, and thus the high detection fraction in our work has important implications for future high frequency AGN radio surveys with the next generation VLA (ngVLA) or Square Kilometre Array (SKA), both of which should detect large fractions of more distant AGN., Comment: 26 pages, 8 figures, 4tables. Accepted for publication in ApJ