Search

The widespread adoption of artificial intelligence (AI) in next-generation communication systems is challenged by the heterogeneity of traffic and network conditions, which call for the use of highly contextual, site-specific, data. A promising solution is to rely not only on real-world data, but also on synthetic pseudo-data generated by a network digital twin (NDT). However, the effectiveness of this approach hinges on the accuracy of the NDT, which can vary widely across different contexts. To address this problem, this paper introduces context-aware doubly-robust (CDR) learning, a novel semi-supervised scheme that adapts its reliance on the pseudo-data to the different levels of fidelity of the NDT across contexts. CDR is evaluated on the task of downlink beamforming, showing superior performance compared to previous state-of-the-art semi-supervised approaches., Comment: This work has been submitted to the IEEE for possible publication

Astrophysical foreground substraction is crucial to retrieve the cosmic microwave background (CMB) polarization out of the observed data. Recent efforts have been carried out towards the development of a minimally informed component separation method to handle a priori unknown foreground spectral energy distributions (SEDs), while being able to estimate both cosmological, foreground, and potentially instrumental parameters, jointly. In this paper, we develop a semi-analytical performance forecasting framework for the minimally informed method and we validate it by comparing its results against direct sampling of the harmonic-based likelihood and the pixel domain implementation MICMAC. We then use the forecasting tool to demonstrate the robustness of the bias correction procedure introduced in the minimally informed approach. We find that a data-driven approach based on the currently available observational data is enough to efficiently regularize the bias of the method.

The Hubble tension between direct measurements of the expansion rate and the prediction of the $\Lambda$CDM cosmological model calibrated on the Cosmic Microwave Background (CMB), is a strong motivation to explore alternative cosmological models. A popular class of such models includes an additional component of dark energy relevant in the early Universe, which solves the Hubble tension. These Early Dark Energy (EDE) models however typically worsen the $S_8$ tension between measurements of weak gravitational lensing and the $\Lambda$CDM expectation. Here, we show how combining EDE with scale-dependent primordial non-Gaussianities (sPNG) can in principle solve both tensions at once. Such sPNG are related to non-trivial inflationary physics, and observationally affect the non-linear regime of structure formation while leaving the linear regime mostly untouched. Depending on the amplitude of the sPNG, such models can either yield back the $\Lambda$CDM expectation for the PS in the non-linear regime, and hence an $S_8$ parameter compatible with $\Lambda$CDM, or can even reconcile the value of $S_8$ from weak-lensing measurements with the CMB, while solving the Hubble tension in all cases. In such models, both tensions would then be entirely related to pre-CMB physics of the early Universe., Comment: 17 pages, 3 figures, 2 tables, comments welcome :)

Materials which exhibit the Pockels effect are notable for their strong electro-optic interaction and rapid response times and are therefore used extensively in classical electro-optic components for data and telecommunication applications. Yet many materials optimized for room-temperature operation see their Pockels coefficients at cryogenic temperatures significantly reduced - a major hurdle for emerging quantum technologies which have even more rigorous demands than their classical counterpart. A noted example is $\mathrm{BaTiO_3}$, which features the strongest effective Pockels coefficient at room temperature, only to see it reduced to a third (i.e. $\mathrm{r_{eff}} \approx$ 170 pm/V) at a few Kelvin. Here, we show that this behaviour is not inherent and can even be reversed: Strontium titanate ($\mathrm{SrTiO_3}$), a material normally not featuring a Pockels coefficient, can be engineered to exhibit an $\mathrm{r_{eff}}$ of 345 pm/V at cryogenic temperatures - a record value in any thin-film electro-optic material. By adjusting the stoichiometry, we can increase the Curie temperature and realise a ferroelectric phase that yields a high Pockels coefficient, yet with limited optical losses - on the order of decibels per centimetre. Our findings position $\mathrm{SrTiO_3}$ as one of the most promising materials for cryogenic quantum photonics applications., Comment: Main text 7 pages, including supplement information 31 pages

We present the analysis of OGLE-2014-BLG-1760, a planetary system in the galactic bulge. We combine Keck Adaptive Optics follow-up observations in $K$-band with re-reduced light curve data to confirm the source and lens star identifications and stellar types. The re-reduced MOA dataset had an important impact on the light curve model. We find the Einstein ring crossing time of the event to be $\sim$ 2.5 days shorter than previous fits, which increases the planetary mass-ratio and decreases the source angular size by a factor of 0.25. Our OSIRIS images obtained 6 years after the peak of the event show a source-lens separation of 54.20 $\pm$ 0.23 mas, which leads to a relative proper motion of $\mu_{\rm rel}$ = 9.14 $\pm$ 0.05 mas/yr, larger than the previous light curve-only models. Our analysis shows that the event consists of a Jupiter-mass planet of $M_{\rm p}$ = 0.931 $\pm$ 0.117 $M_{\rm Jup}$ orbiting a K-dwarf star of $M_*$ = 0.803 $\pm$ 0.097 $M_{\odot}$ with a $K$-magnitude of $K_{\rm L}$ = 18.30 $\pm$ 0.05, located in the galactic bulge or bar. We also attempt to constrain the source properties using the source angular size $\theta_*$ and $K$-magnitude. Our results favor the scenario of the source being a younger star in the galactic disk, behind the galactic center, but future multicolor observations are needed to constrain the source and thus the lens properties., Comment: 20 pages, 8 figures. In review in AJ

The Athena mission entered a redefinition phase in July 2022, driven by the imperative to reduce the mission cost at completion for the European Space Agency below an acceptable target, while maintaining the flagship nature of its science return. This notably called for a complete redesign of the X-ray Integral Field Unit (X-IFU) cryogenic architecture towards a simpler active cooling chain. Passive cooling via successive radiative panels at spacecraft level is now used to provide a 50 K thermal environment to an X-IFU owned cryostat. 4.5 K cooling is achieved via a single remote active cryocooler unit, while a multi-stage Adiabatic Demagnetization Refrigerator ensures heat lift down to the 50 mK required by the detectors. Amidst these changes, the core concept of the readout chain remains robust, employing Transition Edge Sensor microcalorimeters and a SQUID-based Time-Division Multiplexing scheme. Noteworthy is the introduction of a slower pixel. This enables an increase in the multiplexing factor (from 34 to 48) without compromising the instrument energy resolution, hence keeping significant system margins to the new 4 eV resolution requirement. This allows reducing the number of channels by more than a factor two, and thus the resource demands on the system, while keeping a 4' field of view (compared to 5' before). In this article, we will give an overview of this new architecture, before detailing its anticipated performances. Finally, we will present the new X-IFU schedule, with its short term focus on demonstration activities towards a mission adoption in early 2027., Comment: 44 pages, 14 figures, accepted for publication in Experimental Astronomy

Microfabricated alkali vapor cells are key to enable miniature devices such as atomic clocks and optically pumped magnetometers with reduced size, weight and power. Yet, more versatile fabrication methods are still needed to further expand their use cases. Here, we demonstrate a novel approach to collectively fill and seal microfabricated cesium cells using locally-sealed microchannels patterned within one of the glass substrates comprising the cells. Unlike current methods that rely on wafer-level anodic bonding as the last sealing step, an approach based on local sealing opens the path to features so far limited to traditional glass-blown cells, including the ability to deposit temperature-sensitive antirelaxation coatings, reaching lower background gas pressure without an additional gettering material or filling with diverse atomic or molecular species., Comment: 5 pages, 5 figures, submitted to Applied Physics Letters

Reed conjectured that the chromatic number of any graph is closer to its clique number than to its maximum degree plus one. We consider a recolouring version of this conjecture, with respect to Kempe changes. Namely, we investigate the largest $\varepsilon$ such that all graphs $G$ are $k$-recolourable for all $k \ge \lceil \varepsilon \omega(G) + (1 -\varepsilon)(\Delta(G)+1) \rceil$. For general graphs, an existing construction of a frozen colouring shows that $\varepsilon \le 1/3$. We show that this construction is optimal in the sense that there are no frozen colourings below that threshold. For this reason, we conjecture that $\varepsilon = 1/3$. For triangle-free graphs, we give a construction of frozen colourings that shows that $\varepsilon \le 4/9$, and prove that it is also optimal. In the special case of odd-hole-free graphs, we show that $\varepsilon = 1/2$, and that this is tight up to one colour., Comment: 17 pages, 8 figures

Flexible fibers at the microscopic scale, such as flagella and cilia, play essential roles in biological and synthetic systems. The dynamics of these slender filaments in viscous flows involve intricate interactions between their mechanical properties and hydrodynamic drag. In this paper, considering a 1D, planar, inextensible Euler-Bernoulli rod in a viscous fluid modeled by Resistive Force Theory, we establish the existence and uniqueness of solutions for the $N$-link model, a mechanical model, designed to approximate the continuous filament with rigid segments. Then, we prove the convergence of the $N$-link model's solutions towards the solutions to classical elastohydrodynamics equations of a flexible slender rod. This provides an existence result for the limit model, comparable to those by Mori and Ohm [Nonlinearity, 2023], in a different functional context and with different methods. Due to its mechanical foundation, the discrete system satisfies an energy dissipation law, which serves as one of the main ingredients in our proofs. Our results provide mathematical validation for the discretization strategy that consists in approximating a continuous filament by the mechanical $N$-link model, which does not correspond to a classical approximation of the underlying PDE.

The Early Release Observations (ERO) from Euclid have detected several new galaxy-galaxy strong gravitational lenses, with the all-sky survey expected to find 170,000 new systems, greatly enhancing studies of dark matter, dark energy, and constraints on the cosmological parameters. As a first step, visual inspection of all galaxies in one of the ERO fields (Perseus) was carried out to identify candidate strong lensing systems and compared to the predictions from Convolutional Neural Networks (CNNs). However, the entire ERO data set is too large for expert visual inspection. In this paper, we therefore extend the CNN analysis to the whole ERO data set, using different CNN architectures and methodologies. Using five CNN architectures, we identified 8,469 strong gravitational lens candidates from IE-band cutouts of 13 Euclid ERO fields, narrowing them to 97 through visual inspection, including 14 grade A and 31 grade B candidates. We present the spectroscopic confirmation of a strong gravitational lensing candidate, EUCLJ081705.61+702348.8. The foreground lensing galaxy, an early-type system at redshift z = 0.335, and the background source, a star-forming galaxy at redshift z = 1.475 with [O II] emission, are both identified. Lens modeling using the Euclid strong lens modeling pipeline reveals two distinct arcs in a lensing configuration, with an Einstein radius of 1.18 \pm 0.03 arcseconds, confirming the lensing nature of the system. These findings highlight the importance of a broad CNN search to efficiently reduce candidates, followed by visual inspection to eliminate false positives and achieve a high-purity sample of strong lenses in Euclid.

The Redheffer matrix $A_n \in \mathbb{R}^{n \times n}$ is defined by setting $A_{ij} = 1$ if $j=1$ or $i$ divides $j$ and 0 otherwise. One of its many interesting properties is that $\det(A_n) = O(n^{1/2 + \varepsilon})$ is equivalent to the Riemann hypothesis. The singular vector $v \in \mathbb{R}^n$ corresponding to the largest singular value carries a lot of information: $v_k$ is small if $k$ is prime and large if $k$ has many divisors. We prove that the vector $w$ whose $k-$th entry is the sum of the inverse divisors of $k$, $w_k = \sum_{d|k} 1/d$, is close to a singular vector in a precise quantitative sense.

Open-ended learning agents must efficiently prioritize goals in vast possibility spaces, focusing on those that maximize learning progress (LP). When such autotelic exploration is achieved by LLM agents trained with online RL in high-dimensional and evolving goal spaces, a key challenge for LP prediction is modeling one's own competence, a form of metacognitive monitoring. Traditional approaches either require extensive sampling or rely on brittle expert-defined goal groupings. We introduce MAGELLAN, a metacognitive framework that lets LLM agents learn to predict their competence and LP online. By capturing semantic relationships between goals, MAGELLAN enables sample-efficient LP estimation and dynamic adaptation to evolving goal spaces through generalization. In an interactive learning environment, we show that MAGELLAN improves LP prediction efficiency and goal prioritization, being the only method allowing the agent to fully master a large and evolving goal space. These results demonstrate how augmenting LLM agents with a metacognitive ability for LP predictions can effectively scale curriculum learning to open-ended goal spaces.

Human-In-The-Loop (HITL) frameworks are integral to many real-world computer vision systems, enabling human operators to make informed decisions with AI assistance. Conformal Predictions (CP), which provide label sets with rigorous guarantees on ground truth inclusion probabilities, have recently gained traction as a valuable tool in HITL settings. One key application area is video surveillance, closely associated with Human Action Recognition (HAR). This study explores the application of CP on top of state-of-the-art HAR methods that utilize extensively pre-trained Vision-Language Models (VLMs). Our findings reveal that CP can significantly reduce the average number of candidate classes without modifying the underlying VLM. However, these reductions often result in distributions with long tails. To address this, we introduce a method based on tuning the temperature parameter of the VLMs to minimize these tails without requiring additional calibration data. Our code is made available on GitHub at the address https://github.com/tbary/CP4VLM., Comment: 6 pages, 7 figures

We report the discovery of a complete Einstein ring around the elliptical galaxy NGC 6505, at $z=0.042$. This is the first strong gravitational lens discovered in Euclid and the first in an NGC object from any survey. The combination of the low redshift of the lens galaxy, the brightness of the source galaxy ($I_\mathrm{E}=18.1$ lensed, $I_\mathrm{E}=21.3$ unlensed), and the completeness of the ring make this an exceptionally rare strong lens, unidentified until its observation by Euclid. We present deep imaging data of the lens from the Euclid Visible Camera (VIS) and Near-Infrared Spectrometer and Photometer (NISP) instruments, as well as resolved spectroscopy from the Keck Cosmic Web Imager (KCWI). The Euclid imaging in particular presents one of the highest signal-to-noise ratio optical/near-infrared observations of a strong gravitational lens to date. From the KCWI data we measure a source redshift of $z=0.406$. Using data from the Dark Energy Spectroscopic Instrument (DESI) we measure a velocity dispersion for the lens galaxy of $\sigma_\star=303\pm15\,\mathrm{kms}^{-1}$. We model the lens galaxy light in detail, revealing angular structure that varies inside the Einstein ring. After subtracting this light model from the VIS observation, we model the strongly lensed images, finding an Einstein radius of 2.5 arcsec, corresponding to $2.1\,\mathrm{kpc}$ at the redshift of the lens. This is small compared to the effective radius of the galaxy, $R_\mathrm{eff}\sim 12.3\,\mathrm{arcsec}$. Combining the strong lensing measurements with analysis of the spectroscopic data we estimate a dark matter fraction inside the Einstein radius of $f_\mathrm{DM} = (11.1_{-3.5}^{+5.4})\%$ and a stellar initial mass-function (IMF) mismatch parameter of $\alpha_\mathrm{IMF} = 1.26_{-0.08}^{+0.05}$, indicating a heavier-than-Chabrier IMF in the centre of the galaxy., Comment: Accepted in A&A. Press release: https://www.esa.int/Science_Exploration/Space_Science/Euclid/Euclid_discovers_a_stunning_Einstein_ring

This paper describes the problem of coordination of an autonomous Multi-Agent System which aims to solve the coverage planning problem in a complex environment. The considered applications are the detection and identification of objects of interest while covering an area. These tasks, which are highly relevant for space applications, are also of interest among various domains including the underwater context, which is the focus of this study. In this context, coverage planning is traditionally modelled as a Markov Decision Process where a coordinated MAS, a swarm of heterogeneous autonomous underwater vehicles, is required to survey an area and search for objects. This MDP is associated with several challenges: environment uncertainties, communication constraints, and an ensemble of hazards, including time-varying and unpredictable changes in the underwater environment. MARL algorithms can solve highly non-linear problems using deep neural networks and display great scalability against an increased number of agents. Nevertheless, most of the current results in the underwater domain are limited to simulation due to the high learning time of MARL algorithms. For this reason, a novel strategy is introduced to accelerate this convergence rate by incorporating biologically inspired heuristics to guide the policy during training. The PSO method, which is inspired by the behaviour of a group of animals, is selected as a heuristic. It allows the policy to explore the highest quality regions of the action and state spaces, from the beginning of the training, optimizing the exploration/exploitation trade-off. The resulting agent requires fewer interactions to reach optimal performance. The method is applied to the MSAC algorithm and evaluated for a 2D covering area mission in a continuous control environment., Comment: i-SAIRAS 2024 Conference

This article examines the organisational and geographical forces that shape the supply chains of artificial intelligence (AI) through outsourced and offshored data work. Bridging sociological theories of relational inequalities and embeddedness with critical approaches to Global Value Chains, we conduct a global case study of the digitally enabled organisation of data work in France, Madagascar, and Venezuela. The AI supply chains procure data work via a mix of arm's length contracts through marketplace-like platforms, and of embedded firm-like structures that offer greater stability but less flexibility, with multiple intermediate arrangements. Each solution suits specific types and purposes of data work in AI preparation, verification, and impersonation. While all forms reproduce well-known patterns of exclusion that harm externalised workers especially in the Global South, disadvantage manifests unevenly in different supply chain structures, with repercussions on remunerations, job security and working conditions. Unveiling these processes of contemporary technology development provides insights into possible policy implications., Comment: New Political Economy, In press

We compare multiple foreground-cleaning pipelines for estimating the tensor-to-scalar ratio, $r$, using simulated maps of the planned CMB-S4 experiment within the context of the South Pole Deep Patch. To evaluate robustness, we analyze bias and uncertainty on $r$ across various foreground suites using map-based simulations. The foreground-cleaning methods include: a parametric maximum likelihood approach applied to auto- and cross-power spectra between frequency maps; a map-based parametric maximum-likelihood method; and a harmonic-space internal linear combination using frequency maps. We summarize the conceptual basis of each method to highlight their similarities and differences. To better probe the impact of foreground residuals, we implement an iterative internal delensing step, leveraging a map-based pipeline to generate a lensing $B$-mode template from the Large Aperture Telescope frequency maps. Our results show that the performance of the three approaches is comparable for simple and intermediate-complexity foregrounds, with $\sigma(r)$ ranging from 3 to 5 $\times 10^{-4}$. However, biases at the $1-2\sigma$ level appear when analyzing more complex forms of foreground emission. By extending the baseline pipelines to marginalize over foreground residuals, we demonstrate that contamination can be reduced to within statistical uncertainties, albeit with a pipeline-dependent impact on $\sigma(r)$, which translates to a detection significance between 2 and 4$\sigma$ for an input value of $r = 0.003$. These findings suggest varying levels of maturity among the tested pipelines, with the auto- and cross-spectra-based approach demonstrating the best stability and overall performance. Moreover, given the extremely low noise levels, mutual validation of independent foreground-cleaning pipelines is essential to ensure the robustness of any potential detection., Comment: 25 pages, 14 figures

Thin stellar streams, such as those resulting from the tidal disruption of globular clusters, have long been known and used as probes of the gravitational potential of our Galaxy, both its visible and dark contents. In particular, the presence of under-density regions, or gaps, along these streams is commonly interpreted as being due to the close passage of dark matter sub-halos. In this work, we investigate the perturbations induced on streams by the passage of dense stellar systems, such as globular clusters themselves, to test the possibility that they may cause the formation of gaps as well. In particular, we focus on the study of the stream of Palomar 5, a well-known globular cluster in the Galactic halo, which has particularly long tidal tails. For this purpose, we used a particle-test code to simulate Palomar 5's tidal tails when subjected to the Galaxy's gravitational field plus its whole system of globular clusters. Our study shows that the tails of Palomar 5 can be strongly perturbed by the close passage of other clusters, in particular of NGC 2808, NGC 7078, NGC 104, and that these perturbations induce the formation of gaps in the tails. These results show that globular clusters are capable of inducing gaps in streams--as other baryonic components such as giant molecular clouds and the galactic bar have been shown to do in other works. Therefore, when searching to construct the distribution function of dark matter sub halos within the Milky Way, the gap contribution from globular clusters must be included., Comment: 13 pages, 8 figures, 17 appendix pages

Nualart & Pecatti ([Nualart and Peccati, 2005, Thm 1]) established the first fourth-moment theorem for random variables in a fixed Wiener chaos, i.e. they showed that convergence of the sequence of fourth moments to the fourth moment of the standard Gaussian distribution is sufficient for weak convergence to the standard Gaussian. In this paper, we provide what we believe to be the first generalization to chaos expansions with more than a single term. Specifically, we show that a fourth-moment theorem holds for random variables consisting of sums of two multiple integrals of orders $p, q \in N$, where $p, q$ have different parities. Furthermore, we show that such random variables cannot themselves be Gaussian, again generalizing what is known for the fixed Wiener chaos setting. Finally, we show a fourth-moment theorem for variables with infinite Wiener chaos expansions when the terms in the expansions are independent and satisfy an additional regularity condition in terms of the Ornstein-Uhlenbeck operator., Comment: 8 pages

Scalable and secure data management is important in Internet of Things (IoT) applications such as smart water meters, where traditional blockchain storage can be restrictive due to high data volumes. This paper investigates a hybrid blockchain and InterPlanetary File System (IPFS) approach designed to optimise storage efficiency, enhance throughput, and reduce block time by offloading large data off-chain to IPFS while preserving on-chain integrity. A substrate-based private blockchain was developed to store smart water meter (SWM) data, and controlled experiments were conducted to evaluate blockchain performance with and without IPFS. Key metrics, including block size, block time, and transaction throughput, were analysed across varying data volumes and node counts. Results show that integrating IPFS significantly reduces on-chain storage demands, leading to smaller block sizes, increased throughput, and improved block times compared to blockchain-only storage. These findings highlight the potential of hybrid blockchain-IPFS models for efficiently and securely managing high-volume IoT data.

In Bayesian statistics, the choice of the prior can have an important influence on the posterior and the parameter estimation, especially when few data samples are available. To limit the added subjectivity from a priori information, one can use the framework of reference priors. However, computing such priors is a difficult task in general. We develop in this paper a flexible algorithm based on variational inference which computes approximations of reference priors from a set of parametric distributions using neural networks. We also show that our algorithm can retrieve reference priors when constraints are specified in the optimization problem to ensure the solution is proper. We propose a simple method to recover a relevant approximation of the parametric posterior distribution using Markov Chain Monte Carlo (MCMC) methods even if the density function of the parametric prior is not known in general. Numerical experiments on several statistical models of increasing complexity are presented. We show the usefulness of this approach by recovering the target distribution. The performance of the algorithm is evaluated on the prior distributions as well as the posterior distributions, jointly using variational inference and MCMC sampling.

Encryption-based attacks have introduced significant challenges for detection mechanisms that rely on predefined signatures, heuristic indicators, or static rule-based classifications. Probabilistic Latent Encryption Mapping presents an alternative detection framework that models ransomware-induced encryption behaviors through statistical representations of entropy deviations and probabilistic dependencies in execution traces. Unlike conventional approaches that depend on explicit bytecode analysis or predefined cryptographic function call monitoring, probabilistic inference techniques classify encryption anomalies based on their underlying statistical characteristics, ensuring greater adaptability to polymorphic attack strategies. Evaluations demonstrate that entropy-driven classification reduces false positive rates while maintaining high detection accuracy across diverse ransomware families and encryption methodologies. Experimental results further highlight the framework's ability to differentiate between benign encryption workflows and adversarial cryptographic manipulations, ensuring that classification performance remains effective across cloud-based and localized execution environments. Benchmark comparisons illustrate that probabilistic modeling exhibits advantages over heuristic and machine learning-based detection approaches, particularly in handling previously unseen encryption techniques and adversarial obfuscation strategies. Computational efficiency analysis confirms that detection latency remains within operational feasibility constraints, reinforcing the viability of probabilistic encryption classification for real-time security infrastructures. The ability to systematically infer encryption-induced deviations without requiring static attack signatures strengthens detection robustness against adversarial evasion techniques.

Sudden violations of the slow-roll regime during inflation, a natural prediction of many UV-complete inflationary models, give rise to sharp features in the primordial power spectrum. At large scales, these features provide a unique window into the physics of inflation, with constraints primarily derived from Cosmic Microwave Background observations of linearly evolved primordial fluctuations. However, on smaller scales, it is less clear whether primordial features would survive the late-time nonlinear cosmological evolution, as they are expected to be washed out by mode coupling. In this paper, we run dedicated N-body simulations to tackle this question. We demonstrate that, while oscillatory-like patterns are erased over time by nonlinearities, signatures of primordial sharp features can persist through the nonlinear regime of structure formation. Those take the form of a localised power enhancement or decrease in the matter power spectrum, whose amplitude and position can in principle be used to recover the scale of the primordial feature, and an oscillatory pattern in the halo mass function. While these findings highlight the power for constraining inflationary physics at small scales, they also show the challenges posed by potential degeneracies with other physical processes relevant in the nonlinear regime of structure formation such as non-cold dark matter candidates. Our results open new avenues for probing inflationary physics in large scale structures and galactic physics and emphasise the need for refined theoretical tools to robustly constrain primordial features., Comment: 17 pages, 9 figures, 3 tables

Estimating optimal transport maps between two distributions from respective samples is an important element for many machine learning methods. To do so, rather than extending discrete transport maps, it has been shown that estimating the Brenier potential of the transport problem and obtaining a transport map through its gradient is near minimax optimal for smooth problems. In this paper, we investigate the private estimation of such potentials and transport maps with respect to the distribution samples.We propose a differentially private transport map estimator achieving an $L^2$ error of at most $n^{-1} \vee n^{-\frac{2 \alpha}{2 \alpha - 2 + d}} \vee (n\epsilon)^{-\frac{2 \alpha}{2 \alpha + d}} $ up to poly-logarithmic terms where $n$ is the sample size, $\epsilon$ is the desired level of privacy, $\alpha$ is the smoothness of the true transport map, and $d$ is the dimension of the feature space. We also provide a lower bound for the problem.

We introduce a diagrammatic perspective for Shannon entropy created by the first author and Mikhail Khovanov and connect it to information theory and mutual information. We also give two complete proofs that the $5$-term dilogarithm deforms to the $4$-term infinitesimal dilogarithm., Comment: 33 pages, many figures

In this work, we introduce a novel framework for privately optimizing objectives that rely on Wasserstein distances between data-dependent empirical measures. Our main theoretical contribution is, based on an explicit formulation of the Wasserstein gradient in a fully discrete setting, a control on the sensitivity of this gradient to individual data points, allowing strong privacy guarantees at minimal utility cost. Building on these insights, we develop a deep learning approach that incorporates gradient and activations clipping, originally designed for DP training of problems with a finite-sum structure. We further demonstrate that privacy accounting methods extend to Wasserstein-based objectives, facilitating large-scale private training. Empirical results confirm that our framework effectively balances accuracy and privacy, offering a theoretically sound solution for privacy-preserving machine learning tasks relying on optimal transport distances such as Wasserstein distance or sliced-Wasserstein distance.

We introduce a new replicable boosting algorithm which significantly improves the sample complexity compared to previous algorithms. The algorithm works by doing two layers of majority voting, using an improved version of the replicable boosting algorithm introduced by Impagliazzo et al. [2022] in the bottom layer., Comment: 13 pages

Lamellar crystalline materials, whose layers are bond by van der Waals forces, can be stacked to form ultrathin artificial heterostructures, and in particular vertical magnetic junctions when some of the stacked materials are (ferro)magnetic. Here, using the room temperature van der Waals ferromagnet 1$T$-CrTe$_2$, we report a method for patterning lateral magnetic junctions. Exploiting the heat-induced phase transformation of the material into Cr$_x$Te$_y$ compounds ($x/y>1/2$), we use local laser heating to imprint patterns at the micron-scale. Optimizing laser heat dissipation, we further demonstrate the crucial role of the substrate to control the phase transformation. If plain, unstructured poorly heat-conducting substrates allow for direct writing of magnetic patterns, structured $h$-BN layers can serve as heat stencils to draw potentially thinner patterns. Besides, $h$-BN encapsulation turns out to be heat-protective (in addition from protecting against oxidation as it is generally used for), allowing the demonstration of room temperature ferromagnetism in $<$7~nm-thick 1$T$-CrTe$_2$., Comment: accepted in Physical Review Materials, 5 figures

Given a two-dimensional quantum lattice model with an abelian gauge theory interpretation, we investigate a duality operation that amounts to gauging its invertible 1-form symmetry, followed by gauging the resulting 0-form symmetry in a twisted way via a choice of discrete torsion. Using tensor networks, we introduce explicit lattice realisations of the so-called condensation defects, which are obtained by gauging the 1-form symmetry along submanifolds of spacetime, and employ the same calculus to realise the duality operators. By leveraging these tensor network operators, we compute the non-trivial interplay between symmetry-twisted boundary conditions and charge sectors under the duality operation, enabling us to construct isometries relating the dual Hamiltonians. Whenever a lattice gauge theory is left invariant under the duality operation, we explore the possibility of promoting the self-duality to an internal symmetry. We argue that this results in a symmetry structure that encodes the 2-representations of a 2-group.

The Ulam sequence, described by Stanislaw Ulam in the 1960s, starts $1,2$ and then iteratively adds the smallest integer that can be uniquely written as the sum of two distinct earlier terms: this gives $1,2,3,4,6,8,11,\dots$. Already in 1972 the great French poet Raymond Queneau wrote that it `gives an impression of great irregularity'. This irregularity appears to have a lot of structure which has inspired a great deal of work; nonetheless, very little is rigorously proven. We improve the best upper bound on its growth and show that at least some small gaps have to exist: for some $c>0$ and all $n \in \mathbb{N}$ $$ \min_{1 \leq k \leq n} \frac{a_{k+1}}{a_k} \leq 1 + c\frac{\log{n}}{n}.$$

We consider the classical multi-armed bandit problem, but with strategic arms. In this context, each arm is characterized by a bounded support reward distribution and strategically aims to maximize its own utility by potentially retaining a portion of its reward, and disclosing only a fraction of it to the learning agent. This scenario unfolds as a game over $T$ rounds, leading to a competition of objectives between the learning agent, aiming to minimize their regret, and the arms, motivated by the desire to maximize their individual utilities. To address these dynamics, we introduce a new mechanism that establishes an equilibrium wherein each arm behaves truthfully and discloses as much of its rewards as possible. With this mechanism, the agent can attain the second-highest average (true) reward among arms, with a cumulative regret bounded by $O(\log(T)/\Delta)$ (problem-dependent) or $O(\sqrt{T\log(T)})$ (worst-case).

We consider a combinatorial reconfiguration problem on a subclass of quadrangulations of surfaces called square-tiled surfaces. Our elementary move is a shear in a cylinder that corresponds to a well-chosen sequence of diagonal flips that preserves the square-tiled properties. We conjecture that the connected components of this reconfiguration problem are in bijection with the connected components of the moduli space of quadratic differentials. We prove that the conjecture holds in the so-called hyperelliptic components of Abelian square-tiled surfaces. More precisely, we show that any two such square-tiled surfaces of genus $g$ can be connected by $O(g)$ powers of cylinder shears., Comment: 41 pages, 25 figures

Over an arbitrary field $\mathbb{F}$, let $p$ and $q$ be monic polynomials with degree $2$ in $\mathbb{F}[t]$. The free Hamilton algebra of the pair $(p,q)$ is the free noncommutative algebra in two generators $a$ and $b$ subject only to the relations $p(a)=0=q(b)$. Free Hamilton algebras are models of free products of two $2$-dimensional algebras over $\mathbb{F}$. They can be viewed as the most elementary nontrivial noncommutative algebras over fields. It has been recently observed that the free Hamilton algebra has surprising connections with quaternion algebras. Here, we exploit these connections to investigate its zero divisors, group of units, maximal ideals, finite-dimensional subalgebras, and its automorphism group., Comment: 161 pages

The rapid growth of artificial intelligence (AI), particularly Large Language Models (LLMs), has raised concerns regarding its global environmental impact that extends beyond greenhouse gas emissions to include consideration of hardware fabrication and end-of-life processes. The opacity from major providers hinders companies' abilities to evaluate their AI-related environmental impacts and achieve net-zero targets. In this paper, we propose a methodology to estimate the environmental impact of a company's AI portfolio, providing actionable insights without necessitating extensive AI and Life-Cycle Assessment (LCA) expertise. Results confirm that large generative AI models consume up to 4600x more energy than traditional models. Our modelling approach, which accounts for increased AI usage, hardware computing efficiency, and changes in electricity mix in line with IPCC scenarios, forecasts AI electricity use up to 2030. Under a high adoption scenario, driven by widespread Generative AI and agents adoption associated to increasingly complex models and frameworks, AI electricity use is projected to rise by a factor of 24.4. Mitigating the environmental impact of Generative AI by 2030 requires coordinated efforts across the AI value chain. Isolated measures in hardware efficiency, model efficiency, or grid improvements alone are insufficient. We advocate for standardized environmental assessment frameworks, greater transparency from the all actors of the value chain and the introduction of a "Return on Environment" metric to align AI development with net-zero goals.

Suspensions of self-propelled objects represent a novel paradigm in colloidal science. In such active baths traditional concepts, such as Brownian motion, fluctuation-dissipation relations, and work extraction from heat reservoirs, must be extended beyond the conventional framework of thermal baths. Unlike thermal baths, which are characterized by a single parameter, the temperature, the fundamental descriptors of an active bath remain elusive, especially in confined environments. In this study, buoyant, passive tracers are employed as generalized probes to investigate an active bath comprising motile bacteria confined within a droplet. We demonstrate that momentum transfer from the bath to the tracer can be effectively described as colored noise, characterized by temporal memory and an enhanced effective diffusivity significantly larger compared to thermal Brownian motion values. Using a stochastic analytical framework, we extract the temporal memory and diffusivity parameters that define such an active bath. Notably, the diffusivity scales linearly with bacterial concentration, modulated by a factor representing the role of confinement, expressed as the ratio of the confining radius to the probe radius. This finding, while still awaiting a complete theoretical explanation, offers new insights into the transport properties of confined active baths and paves the way for a deeper understanding of active emulsions driven by confined active matter.

The use of quantum stochastic models is widespread in dynamical reduction, simulation of open systems, feedback control and adaptive estimation. In many applications only part of the information contained in the filter's state is actually needed to reconstruct the target observable quantities; thus, filters of smaller dimensions could be in principle implemented to perform the same task.In this work, we propose a systematic method to find, when possible, reduced-order quantum filters that are capable of exactly reproducing the evolution of expectation values of interest. In contrast with existing reduction techniques, the reduced model we obtain is exact and in the form of a Belavkin filtering equation, ensuring physical interpretability.This is attained by leveraging tools from the theory of both minimal realization and non-commutative conditional expectations. The proposed procedure is tested on prototypical examples, laying the groundwork for applications in quantum trajectory simulation and quantum feedback control.

For every positive integer $k$, we define the $k$-treedepth as the largest graph parameter $\mathrm{td}_k$ satisfying (i) $\mathrm{td}_k(\emptyset)=0$; (ii) $\mathrm{td}_k(G) \leq 1+ \mathrm{td}_k(G-u)$ for every graph $G$ and every vertex $u \in V(G)$; and (iii) if $G$ is a $(

This paper investigates the challenges of developing large language models (LLMs) proficient in both multilingual understanding and medical knowledge. We demonstrate that simply translating medical data does not guarantee strong performance on clinical tasks in the target language. Our experiments reveal that the optimal language mix in training data varies significantly across different medical tasks. We find that larger models with carefully calibrated language ratios achieve superior performance on native-language clinical tasks. Furthermore, our results suggest that relying solely on fine-tuning may not be the most effective approach for incorporating new language knowledge into LLMs. Instead, data and computationally intensive pretraining methods may still be necessary to achieve optimal performance in multilingual medical settings. These findings provide valuable guidance for building effective and inclusive medical AI systems for diverse linguistic communities.

We report on the development of an optimized and verified decision procedure for orthologic equalities and inequalities. This decision procedure is quadratic-time and is used as a sound, efficient and predictable approximation to classical propositional logic in automated reasoning tools. We formalize, in the Coq proof assistant, a proof system in sequent-calculus style for orthologic. We then prove its soundness and completeness with respect to the algebraic variety of ortholattices, and we formalize a cut-elimination theorem (in doing so, we discover and fix a missing case in a previously published proof). We then implement and verify a complete proof search procedure for orthologic. A naive implementation is exponential, and to obtain an optimal quadratic runtime, we optimize the implementation by memoizing its results and simulating reference equality testing. We leverage the resulting correctness theorem to implement a reflective Coq tactic. We present benchmarks showing the procedure, under various optimizations, matches its theoretical complexity. Finally, we develop tactics including normalization with respect to orthologic and a boolean solver, which we also benchmark. We make tactics available as a standalone Coq plugin.

Introducing a controlled and strong anharmonicity in mechanical systems is a present challenge of nanomechanics. In cavity optomechanics a mechanical oscillator may be made anharmonic by ponderomotively coupling its motion to the light field of a laser-driven cavity. In the regime where the mechanical resonating frequency and the single-photon coupling constant are small compared to the decay rate of the cavity field, it turns out that the quantum electromagnetic fluctuations of the laser field drive the oscillator into a high-temperature thermal state. The motional state may however be highly non-Gaussian; we show that a precise tuning of system parameters may even lead to a purely quartic effective potential for the mechanical oscillator. We present a theory that predicts the measurable signatures left by the mechanical anharmonicity. In particular, we obtain analytically and numerically the mechanical displacement spectrum, and explore the imprints of the mechanical anharmonicity on the cavity light field., Comment: Submission to SciPost

Improving robustness to uncertainty and rejection of external disturbances represents a significant challenge in aerial robotics. Nonlinear controllers based on Incremental Nonlinear Dynamic Inversion (INDI), known for their ability in estimating disturbances through measured-filtered data, have been notably used in such applications. Typically, these controllers comprise two cascaded loops: an inner loop employing nonlinear dynamic inversion and an outer loop generating the virtual control inputs via linear controllers. In this paper, a novel methodology is introduced, that combines the advantages of INDI with the robustness of linear structured $\mathcal{H}_\infty$ controllers. A full cascaded architecture is proposed to control the dynamics of a multirotor drone, covering both stabilization and guidance. In particular, low-order $\mathcal{H}_\infty$ controllers are designed for the outer loop by properly structuring the problem and solving it through non-smooth optimization. A comparative analysis is conducted between an existing INDI/PD approach and the proposed INDI/$\mathcal{H}_\infty$ strategy, showing a notable enhancement in the rejection of external disturbances. It is carried out first using MATLAB simulations involving a nonlinear model of a Parrot Bebop quadcopter drone, and then experimentally using a customized quadcopter built by the ENAC team. The results show an improvement of more than 50\% in the rejection of disturbances such as gusts.

This paper introduces AFRIDOC-MT, a document-level multi-parallel translation dataset covering English and five African languages: Amharic, Hausa, Swahili, Yor\`ub\'a, and Zulu. The dataset comprises 334 health and 271 information technology news documents, all human-translated from English to these languages. We conduct document-level translation benchmark experiments by evaluating neural machine translation (NMT) models and large language models (LLMs) for translations between English and these languages, at both the sentence and pseudo-document levels. These outputs are realigned to form complete documents for evaluation. Our results indicate that NLLB-200 achieved the best average performance among the standard NMT models, while GPT-4o outperformed general-purpose LLMs. Fine-tuning selected models led to substantial performance gains, but models trained on sentences struggled to generalize effectively to longer documents. Furthermore, our analysis reveals that some LLMs exhibit issues such as under-generation, repetition of words or phrases, and off-target translations, especially for African languages., Comment: under review

Causal survival analysis combines survival analysis and causal inference to evaluate the effect of a treatment or intervention on a time-to-event outcome, such as survival time. It offers an alternative to relying solely on Cox models for assessing these effects. In this paper, we present a comprehensive review of estimators for the average treatment effect measured with the restricted mean survival time, including regression-based methods, weighting approaches, and hybrid techniques. We investigate their theoretical properties and compare their performance through extensive numerical experiments. Our analysis focuses on the finite-sample behavior of these estimators, the influence of nuisance parameter selection, and their robustness and stability under model misspecification. By bridging theoretical insights with practical evaluation, we aim to equip practitioners with both state-of-the-art implementations of these methods and practical guidelines for selecting appropriate estimators for treatment effect estimation. Among the approaches considered, G-formula two-learners, AIPCW-AIPTW, Buckley-James estimators, and causal survival forests emerge as particularly promising.

We investigate the Cr electronic structure and excitations in CrSBr, a layered magnetic semiconductor, using a combination of resonant x-ray spectroscopic techniques. X-ray absorption spectroscopy (XAS) and resonant inelastic x-ray scattering (RIXS) spectra collected at the Cr $L_{2,3}$ edges reveal significant linear dichroism, which arises from the distorted octahedral environment surrounding the Cr$^{3+}$ ions. The origin of the bright excitons observed in this compound is examined through a comparison of the d-d excitations identified in the RIXS spectra, the x-ray excited optical luminescence (XEOL) spectra, and previously reported optical spectroscopic and theoretical studies. To further understand these phenomena, we develop a multiplet model based on a crystal electric field (CEF) approach that accounts for the local environment of Cr ions. This model successfully reproduces several experimental features, while also suggesting strong hybridization effects between Cr $3\textit{d}$ orbitals and ligands that are not fully captured by the present framework. These findings advance our understanding of the electronic structure and excitonic behavior in CrSBr and provide a foundation for future $\textit{in-situ}$ and $\textit{operando}$ studies of CrSBr-based devices for spintronic and optoelectronic applications., Comment: 10 pages, 10 figures

Online test-time adaptation (OTTA) of vision-language models (VLMs) has recently garnered increased attention to take advantage of data observed along a stream to improve future predictions. Unfortunately, existing methods rely on dataset-specific hyperparameters, significantly limiting their adaptability to unseen tasks. In response, we propose Online Gaussian Adaptation (OGA), a novel method that models the likelihoods of visual features using Gaussian distributions and incorporates zero-shot priors into an interpretable Maximum A Posteriori (MAP) estimation framework with fixed hyper-parameters across all datasets. We demonstrate that OGA outperforms state-of-the-art methods on most datasets and runs. Additionally, we show that combining OTTA with popular few-shot techniques (a practical yet overlooked setting in prior research) is highly beneficial. Furthermore, our experimental study reveals that common OTTA evaluation protocols, which average performance over at most three runs per dataset, are inadequate due to the substantial variability observed across runs for all OTTA methods. Therefore, we advocate for more rigorous evaluation practices, including increasing the number of runs and considering additional quantitative metrics, such as our proposed Expected Tail Accuracy (ETA), calculated as the average accuracy in the worst 10% of runs. We hope these contributions will encourage more rigorous and diverse evaluation practices in the OTTA community. Code is available at https://github.com/cfuchs2023/OGA .

The zero-shot capabilities of Vision-Language Models (VLMs) have been widely leveraged to improve predictive performance. However, previous works on transductive or test-time adaptation (TTA) often make strong assumptions about the data distribution, such as the presence of all classes. Our work challenges these favorable deployment scenarios, and introduces a more realistic evaluation framework, including: (i) a variable number of effective classes for adaptation within a single batch, and (ii) non-i.i.d. batches of test samples in online adaptation settings. We provide comprehensive evaluations, comparisons, and ablation studies that demonstrate how current transductive or TTA methods for VLMs systematically compromise the models' initial zero-shot robustness across various realistic scenarios, favoring performance gains under advantageous assumptions about the test samples' distributions. Furthermore, we introduce StatA, a versatile method that could handle a wide range of deployment scenarios, including those with a variable number of effective classes at test time. Our approach incorporates a novel regularization term designed specifically for VLMs, which acts as a statistical anchor preserving the initial text-encoder knowledge, particularly in low-data regimes. Code available at https://github.com/MaxZanella/StatA.

It has been recently shown that cosmological models with scale-dependent primordial non-Gaussianities (sPNG) could provide a possible path to solve current cosmic tensions. Moreover, it has been pointed out that some of these models might mimic the effects of Warm Dark Matter (WDM) for several observables at low redshift. Here, we confirm the qualitative similarity of the matter power spectrum for sPNG and WDM models, but also point out differences in the halo mass function and void size function. We then jointly simulate WDM and sPNG together. Such simulations allow us to demonstrate that the joint impact of WDM and sPNG is close to the linear superposition of their respective effects at low redshift, at the percent level. We finally propose a model with mixed hot and cold dark matter together with sPNG, that reproduces the $\Lambda$CDM power spectrum at redshifts $z \leq 3$ but is still distinct in terms of halo statistics., Comment: 13 pages, 7 figures, comments welcome :)

Pressure is a powerful thermodynamic parameter for tuning the magnetic properties of van der Waals magnets owing to their weak interlayer bonding. However, local magnetometry measurements under high pressure still remain elusive for this important class of emerging materials. Here we introduce a method enabling in situ magnetic imaging of van der Waals magnets under high pressure with sub-micron spatial resolution. Our approach relies on a quantum sensing platform based on boron-vacancy (V$_\text{B}^-$) centers in hexagonal boron nitride (hBN), which can be placed in atomic contact of any type of two-dimensional (2D) material within a van der Waals heterostructure. We first show that the V$_\text{B}^-$ center can be used as a magnetic field sensor up to pressures of a few GPa, a pressure range for which the properties of a wide variety of van der Waals magnets are efficiently altered. We then use V$_\text{B}^-$ centers in a thin hBN layer to perform magnetic imaging of a van der Waals magnet under pressure. As a proof of concept, we study the pressure-dependent magnetization in micrometer-sized flakes of $1T$-CrTe$_2$, whose evolution is explained by a shift of the Curie temperature. Besides providing a new path for studying pressure-induced phase transitions in van der Waals magnets, this work also opens up interesting perspectives for exploring the physics of 2D superconductors under pressure via local measurements of the Meissner effect., Comment: 10 pages, 5 figures

Multiple Instance Learning (MIL) methods allow for gigapixel Whole-Slide Image (WSI) analysis with only slide-level annotations. Interpretability is crucial for safely deploying such algorithms in high-stakes medical domains. Traditional MIL methods offer explanations by highlighting salient regions. However, such spatial heatmaps provide limited insights for end users. To address this, we propose a novel inherently interpretable WSI-classification approach that uses human-understandable pathology concepts to generate explanations. Our proposed Concept MIL model leverages recent advances in vision-language models to directly predict pathology concepts based on image features. The model's predictions are obtained through a linear combination of the concepts identified on the top-K patches of a WSI, enabling inherent explanations by tracing each concept's influence on the prediction. In contrast to traditional concept-based interpretable models, our approach eliminates the need for costly human annotations by leveraging the vision-language model. We validate our method on two widely used pathology datasets: Camelyon16 and PANDA. On both datasets, Concept MIL achieves AUC and accuracy scores over 0.9, putting it on par with state-of-the-art models. We further find that 87.1\% (Camelyon16) and 85.3\% (PANDA) of the top 20 patches fall within the tumor region. A user study shows that the concepts identified by our model align with the concepts used by pathologists, making it a promising strategy for human-interpretable WSI classification.

''Positive geometries'' are a class of semi-algebraic domains which admit a unique ''canonical form'': a logarithmic form whose residues match the boundary structure of the domain. The study of such geometries is motivated by recent progress in particle physics, where the corresponding canonical forms are interpreted as the integrands of scattering amplitudes. We recast these concepts in the language of mixed Hodge theory, and identify ''genus zero pairs'' of complex algebraic varieties as a natural and general framework for the study of positive geometries and their canonical forms. In this framework, we prove some basic properties of canonical forms which have previously been proved or conjectured in the literature. We give many examples and study in detail the case of arrangements of hyperplanes and convex polytopes., Comment: 51 pages, 11 figures