Machine Learning
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Friday 6 November
LLM-based multi-agent systems solve complex tasks through collaboration, but their growing, simultaneously resident key-value (KV) caches create a memory bottleneck. Applying single-model quantization policies independently ignores differences in agents' precision needs, potentially degrading collective reasoning. We introduce Coordinated Mixed-Precision KV Cache Quantization (CoMixKV), which combines online sensitivity estimation with joint agent-layer bit-width allocation under a shared memory budget. To our knowledge, CoMixKV is the first mixed-precision KV-cache quantization framework to jointly allocate precision across simultaneously resident agent caches under such a budget. The estimator derives sensitivity from a local attention-error bound using recent queries, keys, and values, without calibration data, trial quantization, or additional model passes. A shared controller combines these scores with remaining quantizable capacity and byte-level storage costs to minimize total estimated error. Allocations are refreshed as agents start or finish and new observations become available. Selected widths apply only to newly quantized blocks; recent states retain native precision and previously quantized blocks remain unchanged. Across four benchmarks and three LLM backbones at average 2-bit, 3-bit, and 4-bit settings, CoMixKV achieves strong accuracy against the evaluated quantization baselines. At the average 3-bit setting on Gemma-4-31B-it, it achieves 68.35% average accuracy, exceeding the strongest quantized baseline by 0.59%. On Qwen3-32B, it reduces peak KV-cache memory by 32.08% and achieves a 1.38x inference speedup over FullKV.
Machine Learning (Statistics)
Adaptive Budgeted Data Acquisition for Robust Off-Policy Evaluation from Multiple Loggers
We consider off-policy evaluation when additional data can be purchased from several existing logging policies (or loggers) with heterogeneous costs per observation. While existing work on multiple loggers presumes a fixed number of observations from each logger, we study how to allocate a limited budget across them sequentially to evaluate a finite class of target policies. We propose Adaptive Greedy, an algorithm that purchases observations sequentially from the affordable logger that has the most cost-normalized variance reduction across the target policies. It turns out that each such purchase is an 'approximate conditional-gradient step' for the corresponding design problem. Under mild assumptions, we establish that the empirical allocation converges almost surely to the optimal design at an explicit rate. We also derive the asymptotic normality of the final pooled estimator and show that it attains the semiparametric efficiency bound. Simulations and a semi-synthetic application based on the AIDS Clinical Trials Group Study 175 show that Adaptive Greedy outperforms standard allocation strategies, remains accurate if stopped early, and matches the final accuracy of repeated integer programming at a fraction of its computational cost.
Read the full abstract
Statistics Theory
Consistency considerations for rare-event importance sampling
We consider the problem of estimating a small probability $p = P(X \in A)$ by importance sampling with auxiliary density $g$, where $X$ is drawn according to some density $f$. We establish a simple necessary and sufficient condition for the consistency of the importance sampling estimator in triangular schemes where the dimension, the densities $f$ and $g$ and the set $A$ may evolve. It allows to refine the results of Chatterjee and Diaconis [Ann. Appl. Probab. 28 (2018), no. 2, 1099-1135] by establishing the exact minimal sample size required for consistency. Moreover, it allows to study the regime $p \to 0$ which, despite being the relevant mathematical regime for rare-event estimations, has received very little attention so far. We perform an in-depth study of the hyperplane case $A = \{x: x^\top u \geq \alpha\}$ with Gaussian densities, showing that in this case, the optimal importance sampling density considerably improves consistency compared to the standard Monte Carlo estimator. However, we conclude the paper with moment considerations which suggest that, whenever the Monte Carlo and optimal importance sampling estimators both converge, convergence may occur faster for the Monte Carlo estimator.
Read the full abstract
Computer Vision
Contrastive fine-tuning against stain and scanner bias improves robustness and downstream performance across pathology foundation models
Histopathology foundation models achieve strong performance across diverse downstream tasks, yet their representations have been shown to encode scanner- and site-specific biases, creating generalization gaps and potentially limiting real-world utility. We introduce SPECTRA (Scanner and Pathology Embedding Contrastive Tuning for Robust Alignment), a model-agnostic contrastive learning post-training framework which explicitly encourages scanner- and stain-robust representations, to optimize performance and generalization on downstream applications. Using parameter-efficient LoRA adapters trained on matched multi-scanner and multi-stain whole-slide images sourced from the PLISM dataset, our method aligns spatially corresponding regions across acquisition conditions (scanner, stain), using other locations within each candidate condition as negatives. SPECTRA improved PathoROB robustness index, gene-expression prediction on HEST, slide-level classification AUC on CPTAC, and held-out PLISM retrieval for every model. Overall tile-level classification on THUNDER also improved, with exceptions for segmentation and calibration tasks. We open-source our training and evaluation codebases and release adapters and model weights for selected backbones.
Read the full abstract
Computer Vision
Adapting Frozen Promptable Segmenters to User-Defined Taxonomies via Few-Shot Proposal Learning
Promptable segmenters of the SAM family turn a click or a phrase into an instance mask, so the costly part of a label is now its class. However, each annotation project draws its own class boundary, often through objects these models already know: people from cars, sedans from trucks, one sedan model from another. A generic model's boundary is fixed by its training vocabulary. We present SNAP, an annotation pipeline that learns the project's boundary from the labels the annotator draws anyway. On one frozen SAM2.1 pass per image, we fit a per-project head, SnapHead, that classifies every proposal and refits in seconds after each labeled image. Where SAM 3 is available, the same labels decide class by class whether to keep its boundary or ours. Given the true masks of eight look-alike cell lines, SAM 3 assigns classes at Cohen's kappa = 0.038 and SnapHead at 0.754 (60 labeled images), and when only the class names change, SAM 3 falls to chance and SnapHead's score is unchanged. End to end, with 20 labeled images, SNAP beats a SAM 3 oracle tuned on the evaluation labels on five microscopy projects (0.484 against 0.441 F1), and run unchanged on five held-out datasets, it beats SAM 3 as shipped on four. Our results suggest that frozen foundation features already encode the class boundaries a generic vocabulary misses, and that a project's own labels are enough to read them out.
Read the full abstract
Image and Video Processing
Polarization Aware Deep Learning Framework for SAR Despeckling
Speckle is an inherent characteristic of synthetic aperture radar (SAR) imagery that complicates image interpretation and the extraction of reliable information. Although deep-learning-based despeckling methods have demonstrated promising performance, their applicability across different polarimetric configurations is often limited by the need for specialized architectures or separate training procedures. This paper presents a unified deep-learning framework for despeckling single-, dual-, and full-polarization SAR data using a single trained convolutional neural network. The proposed approach adopts a common nine-channel representation of the polarimetric covariance matrix and a polarization-aware masking strategy to accommodate different configurations without modifying the network architecture. Qualitative and quantitative comparison on three different scenes from different sensors demonstrate competitive performance across the considered experiments and support the feasibility of a unified, polarization-flexible approach to SAR despeckling across different polarimetric configurations.
Read the full abstract
Cosmology and Nongalactic Astrophysics
Gravitational Production of Non-minimally Coupled Vector Dark Matter in Warm Inflation
We study the purely gravitational production of a massive vector dark matter with a non-minimal coupling to the Einstein tensor, $\xi G_{\mu\nu}X^\mu X^\nu$, in warm inflation. The background, $V=\lambda\phi^4$ with a dissipation coefficient linear in temperature, is constrained by CMB observations and parametrized by the dissipation ratio $10^{-4}\le Q_0\le1$. We consider both unavoidable production channels: freeze-in from the Standard Model bath through graviton exchange, and cosmological gravitational particle production (CGPP). For freeze-in, we compute the scattering amplitudes including the contact interaction induced by $\xi$, which enhances the production of light dark matter through the longitudinal mode and suppresses that of heavy dark matter through destructive interference. For CGPP, the slow-roll parameter is not small during warm inflation, which enhances the production of light dark matter by up to three orders of magnitude relative to cold inflation. Combining both channels, CGPP excludes most of the light freeze-in branch, leaving $m_X\lesssim2.4\times10^{-5}$ GeV, a narrow window around $10^{10}$ GeV for strong dissipation, and $m_X\gtrsim10^{14}$–$10^{15}$ GeV. Only the heavy branch remains viable for all $|\xi|\lesssim1$.
Read the full abstract
Artificial Intelligence
META-IR: Meta-Introspective Reasoning for Adaptive Safety Alignment in LLM
Large Language Models (LLMs) remain vulnerable to jailbreak attacks and harmful-content prompts, particularly under adversarial distribution shifts where static alignment pipelines fail to generalize. Although introspective methods such as STAIR strengthen refusal consistency, they depend on fixed self-evaluation policies that cannot rapidly adapt to evolving manipulation strategies. To overcome these limitations, this paper introduces META-IR, a meta-introspective reasoning framework that enhances safety alignment through meta-learning-driven adaptation. It employs a dual-loop architecture: an inner introspective loop that performs structured risk analysis, self-assessment, and reflective correction via SI-MCTS, and an outer meta-learning loop that optimizes the introspection policy across diverse jailbreak task families to enable rapid generalization to unseen attacks. A dynamic jailbreak simulation module further exposes the model to progressively harder adversarial prompts, enabling continual self-improvement without human supervision. Experiments across multiple models (Llama-3.1-8B-Instruct and Qwen-2-7B-Instruct) and a broad suite of safety, helpfulness, generalization, and introspection metrics demonstrate that META-IR substantially improves adaptation speed, robustness, and reasoning stability compared with all baseline models. These results establish META-IR as an effective pathway toward building LLMs capable of sustained, adaptive safety alignment under dynamic and evolving threat environments.
Read the full abstract
Computer Vision
Same Pixels, Different Documents: Measuring Part-Level Editability in Image Vectorization
Image vectorization is often evaluated by rendering fidelity, which cannot distinguish documents that render identically but support different edits. Edit-based evaluation tests outcomes, but an oracle that gathers a part's fragments can conceal the cost of selecting them. We introduce Pass@k, which tests whether a prescribed edit can succeed using objects selectable within k clicks under a specified interaction model. Eight document diagnostics characterize properties relevant to editing beyond these tasks. We assess structural sensitivity through controlled perturbations, practical relevance through designer operations, and reference dependence through alternative drawings. We evaluate twelve open and commercial methods on 211 references, with 635 edit tasks on 93 of them. In each of four reference sets, including two drawn by designers, 64-88% of oracle-passed tasks on parts composed of multiple shapes fail at Pass@1, against 6-40% for single shapes, pooling counts over configurations. Higher fidelity does not guarantee higher Pass@1, while selected diagnostics are associated with designers' effort in our study.
Read the full abstract
Rings and Algebras
Topological Lifting of Mundici Equivalence and Di Nola–Lettieri Equivalence
We prove that summability characterises precisely when a topological MV-algebra can be realized as the unit interval of a lattice-ordered topological abelian group, thereby providing a topological lifting of both Mundici equivalence and DiNola–Lettieri equivalence. Applications include a Birkhoff–Kakutani-type metrisation theorem, a second-countability criterion, and a topological extension of a key continuity lemma from the proof theory of first-order Łukasiewicz logic—following Baaz–Metcalfe's hypersequent approach to approximate completenesss—and provides an $A$-valued topological semantics for the approximate strong completeness of~$\mathbf{G}\text{{\L}}\forall$, where $A$ ranges over summable topological MV-algebras.
Read the full abstract
Machine Learning
Transferring local rib responses to new component geometries in sheet forming with a graph-based Transformer
Sheet metal components often consist of a global geometry, such as the corner radii and forming depth of a corner component, and local features added to it. During forming, both affect material flow. A surrogate learning all their combinations may need large training data. However, the same local feature type is often used on many global geometries. This study therefore partly decouples local features from the global geometry. The local-feature effect is first learned on one global geometry, needing fewer simulations than learning all combinations, then transferred to other global geometries via transfer learning, which can need fewer target simulations than training from scratch. A baseline-conditioned residual surrogate, R-BiLoGo, is proposed for ribs on the side walls of hot-stamped corner components. Its inputs are the baseline response and the geometric perturbation introduced by the ribs. Its outputs are the changes in displacement and thinning from the baseline. R-BiLoGo gave lower errors than a hierarchical graph-based model and a Transolver-based model on the source geometry, and on two of three selected target geometries after transfer. Across the 22 target geometries, transfer learning with 20 target simulations gave errors similar to or lower than training from scratch with 80 target simulations for about half of the targets. With the same 80 target simulations, transfer learning reduced errors for 95% of targets. The effect of the global geometry on transfer was studied with four parameter-free geometric descriptors and controlled variations across 4 corner subclasses. The benefit of transfer decreased with increasing surface distance and high-curvature change between source and target, and for targets shallower than the source. These trends were also seen in other corner types with a different magnitude.
Read the full abstract
Physics and Society
Beyond node strengths: the topological content of weighted bipartite networks
Weighted bipartite networks encode the topology and intensity of interactions between two disjoint sets of nodes. However, it is unclear whether the observed pattern of connections is already explained by node interaction volumes or contains additional structural information. Here we introduce a hierarchy of maximum-entropy models for continuous weighted bipartite networks that preserves node strengths alone, strengths and global density, or strengths and the complete degree sequence. Across economic, technological, ecological, social and lexical systems, we find that strength constraints alone are generally insufficient to reproduce empirical structures. Yet the missing topological information is often low-dimensional: a single constraint controlling network density frequently provides the most parsimonious description, whereas networks with stronger residual degree heterogeneity require node-specific constraints. Beyond downstream inference, we also show how the choice of topological resolution changes which interactions and projected relationships are identified as statistically significant. Our results indicate that topology is essential, but often compressible, in the statistical modeling of weighted bipartite networks, and provide a scalable computational framework for fitting, sampling, model selection and validation within a common workflow.
Read the full abstract
Cosmology and Nongalactic Astrophysics
Optimal summary statistics to constrain w0waCDM with the DESI BGS
The Dark Energy Spectroscopic Instrument (DESI) is the first stage-4 spectroscopic instrument installed on the 4-m Mayall Telescope at Kitt Peak National Observatory in Arizona. It contains 5000 robotically positioned fibers, allowing the measurement of as many galaxy spectra in one observation (~20 minutes). The data acquisition (2021–2028) aims at constraining the cosmological parameters, and especially the parameters describing the evolution of the dark energy with time. However, standard clustering analysis based on 2-point statistics only capture the Gaussian information of the matter density field, while the analytical models used to match these statistical probe are limited to quasi-linear scales, as they are based on perturbation theory. Therefore, the non-Gaussian and non-linear information of the density field is not exploited. To go beyond these limitations, several alternative clustering methods have been developed that rely on a simulation-based pipeline. In this presentation, we focus on the DESI Bright Galaxy Survey, a magnitude-limited sample of bright galaxies in the redshift range of 0.1 < z < 0.4. We present how accessing the non-linear and non-Gaussian information of the density field allows around 40% improvement on the constraints of some $\Lambda$CDM parameters. We are also extending the method to constrain w0waCDM in order to improve DESI's sensibility at low redshift where the Universe is dominated by dark energy.
Read the full abstract
Superconductivity
Pelletization Pressure Engineering of the Superconducting Properties of Fluorine-Doped PrFeAsO
We investigate the influence of the pelletization pressure on the structural, microstructural, and superconducting properties of fluorine-doped PrFeAsO0.7F0.3 (Pr1111). Precursor powders are pressed into pellets at pressures ranging from 174 to 1131 MPa, after which the applied pressure is released and all samples undergo an identical two-step conventional synthesis process at ambient pressure (CSP-AP). All samples predominantly exhibit the tetragonal P4/nmm structure, and the lattice parameter a and unit-cell volume V decrease with increasing pelletization pressure up to an intermediate-pressure range (~900 MPa) before increasing at higher pressures. The actual fluorine-content derived from energy-dispersive X-ray spectroscopy (EDX) also increase with pelletization pressure, while Raman spectra show only modest variations in the characteristic vibrational modes associated with Fe and As. The superconducting response exhibits a non-monotonic dependence on the pelletization pressure. Among the investigated samples, the sample pelletized at 870 MPa exhibits the most favorable overall superconducting properties, with a superconducting onset temperature of 44.2 K, ~5 K higher than that of the lowest-pressure sample. It also displays a narrower superconducting transition, the highest relative sample density, a favorable effective magnetic critical current density at high fields, and a relatively large effective thermally activated flux-flow activation energy. At higher pelletization pressures, the superconducting response deteriorates, coinciding with increases in the lattice parameters, increased secondary-phase contributions, and reduced relative density. These findings establish the pelletization pressure as a relevant processing parameter for modifying the structural, microstructural and superconducting properties of Pr1111 during CSP-AP.
Read the full abstract
Machine Learning
fNIRSAtlas: A Large-Scale Benchmark for Functional Near-Infrared Spectroscopy Classification
Functional near-infrared spectroscopy (fNIRS) is a promising modality for brain-computer interfaces and cognitive state decoding, but progress in machine learning for fNIRS has been limited by small, isolated datasets and inconsistent evaluation protocols. We introduce fNIRSAtlas, the largest open benchmark for fNIRS classification to date, comprising 23 classification tasks from 19 public data collections, spanning diverse experimental paradigms, including motor execution, motor imagery, mental arithmetic, working memory, and emotion recognition. fNIRSAtlas provides a fully automated pipeline from raw data to evaluation and defines standardized benchmark tasks for both cross-subject and within-subject classification, ensuring reproducible comparison of methods. We evaluate a variety of baseline models from fNIRS and related neuroimaging and time-series machine learning literature on the benchmark. Our results demonstrate that performance varies substantially across tasks and that methods validated on limited data – especially deep learning approaches – often fail to generalize beyond their original evaluation. fNIRSAtlas serves as an extensible, open, and reproducible benchmark to support more robust and cumulative progress in machine learning for fNIRS data. Our code and data can be found at https://fnirsatlas.github.io.
Read the full abstract
Number Theory
A singular homeomorphism between Pierce expansions and continued fraction expansions
In this paper, we study the map $\Theta$ which sends each $x\in[0,1]$ to the number whose continued fraction digits are the first Pierce expansion digit of $x$ and the gaps between two consecutive Pierce expansion digits of $x$. We investigate the regularity of $\Theta$ and $\Theta^{-1}$, including the optimal Hölder exponent of $\Theta^{-1}$, and the local and Hausdorff dimensions of the push-forwards of the Lebesgue measure under $\Theta$ and $\Theta^{-1}$. We also study the differentiability of $\Theta$ at rational and irrational points, and calculate the Hausdorff dimension of the set of points at which $\Theta$ has an infinite derivative.
Read the full abstract
Signal Processing
Real-Time Trans-Oceanic Fiber Sensing via Convergence of Long-Reach OFDR, High-Speed Optical Networking, and RDMA-Capable Edge-GPUs
We demonstrate real-time network-integrated bi-directional fiber sensing over a 4400-km subsea cable. Streaming and fusing two OFDR instruments via RDMA over a live transoceanic 130Gbaud coherent link and GPU edge-compute enabling parallel capture of multiple spatial resolutions from ~10m to ~100m.
Read the full abstract
High Energy Physics (Phenomenology)
Critical Temperature in Nonextensive Hadron Models: A Pressure-based Approach
The determination of the critical temperature $T_c$ in strongly interacting matter is central to understanding the transition between confined hadronic matter and deconfined quark-gluon degrees of freedom. In high energy nuclear collisions, intrinsic fluctuations, memory effects, and long-range color interactions can drive the fireball evolution away from purely Markovian thermalization, making nonextensive statistical descriptions relevant near equilibrium. We investigate the critical temperature of nonextensive hadron models using a pressure-based formulation derived from Tsallis entropy. Here we show that the qualitative structure of the pressure equation is controlled by both the nonextensive parameter $q$ and the model-dependent bag contribution $B^{\text{model}}$. For $q=0$, the pressure equations are monotonic and admit a unique positive root, whereas for $q=1$ the seventh-order term vanishes and analytic solutions can be obtained. For $q>1$, the negative $T^7$ contribution modifies the pressure behavior and can generate multiple positive roots at finite volume. %Cosmological bounds require $q \approx 1$, meaning $q \simeq 1.001$ in proton-proton phenomenology is a system-dependent fit parameter. For $q\geq1$, $T^{\text{model}}_c$ is $\left(102.3\leq T_c^{\text{MIT}}\leq106.1,\ 86.5\leq T_c^{\text{Dual}}\leq 87.8,\ \text{and} \ 113.1\leq T_c^{\text{Fuzzy}}\leq 121.1\right)$ MeV. At $q=0$, $T^{\text{model}}_c \sim 55$ MeV marks the nuclear liquid-gas transition. A weighted comparison constrained by the MIT bag model at $q=1$ gives $(T_c^{\text{w}}\approx102)\ \text{MeV}$, with Fuzzy and Dual QCD terms entering as model-propagation corrections. The interval $(0.99\leq q\leq 1.002)$ therefore provides the most reliable domain for a common physical interpretation and a composite estimate of the critical temperature.
Read the full abstract
Metric Geometry
The inscribed $4$-polytope of maximal volume with nine vertices
We prove that a $4$-polytope with nine vertices inscribed in the unit sphere $\mathbb{S}^{3}$ has volume at most $\frac{5}{24}\sqrt{10 + 2\sqrt{5}}$, and that, up to orthogonal transformations, equality holds only for the convex hull of a square and a regular pentagon inscribed in $\mathbb{S}^{3}$ in orthogonal $2$-planes through the origin. For the combinatorial type of this polytope, whose boundary complex is the join of a $4$-cycle and a $5$-cycle, we write the signed volume as the pairing, through the Hodge star, of the area vectors of the two cycles, and bound each area vector by finite Fourier analysis; the bound holds for every choice of nine points on the sphere. Since an optimal polytope is simplicial, by a result of Horváth and Lángi, it remains to exclude the other simplicial types, which we do by sum-of-squares certificates checked in exact arithmetic.
Read the full abstract
Mesoscale and Nanoscale Physics
Exceptional Topological Bound States in the Continuum
Topological phases, exceptional points, and bound states in the continuum (BICs) are three key concepts in condensed matter and classical wave physics. Their pairwise combinations have yielded a rich variety of intriguing singular modes, but the triple merger–exceptional topological BICs (ET-BICs)–has yet to be explored. In this Letter, we theoretically propose and experimentally demonstrate a universal approach for constructing such elusive exceptional structures. The system consists of four coupled topological chains: y-mirror stacking generates pairs of topological BICs, while z-stacking with biased loss yields a generalized mirror symmetry under which each pair–separated in different subspaces until their collapse–merges into an ET-BIC at a critical loss. Experimentally, using an acoustic platform with electrically tunable loss, we not only directly observe the coalescence of eigenenergies but also the coalescence of eigenstates, offering a complete verification of the ET-BIC formation. The triply synergistic ET-BICs may pave the way for wave devices that simultaneously demand robustness, sensitivity, and BIC confinement.
Read the full abstract
Superconductivity
Roadmap on Nanoscale Superconductivity for Quantum Technologies
In 2025, the Year of Quantum Science and Technology, we celebrate a century of quantum mechanics, witnessing a surge in activities that illuminate its inherent strangeness and drive technological innovation. Superconductivity, discovered 114 years ago, stands as a prime example, offering direct and compelling evidence of macroscopic quantum phenomena. Beyond its ability to conduct immense currents without loss, superconductivity reveals the quantum realm operating on a scale we can directly observe and manipulate. The macroscopic quantum coherence, where an ensemble of particles is described by a single wave function, leads to remarkable consequences: dissipation-less current and flux quantization - the basic properties exploited in superconducting quantum circuit fabrication. This Roadmap has been inspired by intensive discussions and collaborations emerging from the European COST Action CA21144 (SuperQuMap - Superconducting Nanodevices and Quantum Materials for Coherent Manipulation). The roadmap outlines the network's concrete activities, driving advancements in superconductor-based quantum technologies and charting future directions. Spanning fundamental research to practical applications, the roadmap incorporates insights from industry partners developing quantum computation. It begins by exploring quantum materials, highlighting how topology and electronic correlations could catalyze a quantum leap in technology. We then delve into manipulating the superconducting phase, leveraging advancements in magnetism, 3D fabrication, and tunable correlations. Further, we showcase the advanced microscopy techniques - such as angle-resolved photoemission spectroscopy and scanning probes - used to visualize quantum behavior. Finally, and crucially, we detail the quantum devices developed within the network, and their transformative impact on modern quantum computing approaches.
Read the full abstract
High Energy Physics (Phenomenology)
Phenomenological explorations of a quasi-degenerate $Z'$ boson
We do a general model-independent phenomenological analysis of the curious case of a $Z'$ boson with mass close to that of the Standard Model (SM) $Z$ boson. From tree-level calculation alone we show that presence of such a quasi-degenerate or `practically' degenerate $Z'$ boson affects both the cross-section and forward-backward asymmetry of the process $e^+ e^- \to f \bar{f}$ where $f$ (with $f\neq e^-$) can be a lepton or quark. Most importantly the effect on forward-backward asymmetry is found to be more prominent away from the $Z$-pole. We provide constraints on such a $Z'$ from the measurement of $N_{\nu}$, the number of light neutrino species.
Read the full abstract
High Energy Physics (Theory)
A Free-Field Realization of Boundary Symmetries in Discrete Jackiw–Teitelboim Gravity
We investigate a finite-periodic Miura free-field realization of the classical Virasoro–dilaton sector relevant to sl(2,R) Jackiw–Teitelboim (JT) gravity. Starting from a single logarithmic Cartan field, we use the established discrete Miura map and its free-field Poisson bracket to construct the lattice spin-two current. The dilaton is then realized as a periodic composite of the same Cartan variables, selected by an inhomogeneous difference equation and its monodromy condition. This construction yields an explicit induced mixed Poisson bracket without postulating an independent dilaton bracket. We examine the associated transfer problem and quadratic adjoint invariant, and test the finite Fourier algebra against the continuous Virasoro–dilaton transformation, in which the dilaton has conformal weight minus one. The construction supplies a free-field counterpart of the JT boundary transformation laws; it does not by itself identify the Miura Poisson structure with the canonical BF boundary-charge algebra.
Read the full abstract
Image and Video Processing
Noise-Regularized Training for Learned Image Compression
Learned image compression (LIC) is trained almost exclusively on clean images: the only stochastic perturbation in the loop is an additive uniform proxy injected in the latent domain to keep the entropy model differentiable. Outside compression, by contrast, a long line of work—from Tikhonov regularization and contractive autoencoders through denoising score matching and modern diffusion models—has established that Gaussian noise at the input of a network with a clean target acts as a structured probe of the data manifold rather than as data augmentation. Building on this view, we introduce noise-regularized training for LIC: at every training step we add Gaussian noise to the encoder input while keeping the distortion target, the latent uniform proxy, the hyperprior, and the entropy coding pipeline unchanged. The recipe is architecture-agnostic and adds essentially no training or inference overhead. Across backbones and operating regimes it yields consistent BD-Rate gains: -2.60% on a reproduced ELIC and -0.85% / -1.05% / -2.38% on the Base/Medium/Large scales of a variable-rate DMCI codec, with DMCI-Large reaching -22.02% against VTM-22.0. A second-order expansion of the noisy RD loss accounts for these gains: because the noise enters upstream of the analysis transform, it converts the clean RD objective into a structured $O(\sigma^{2})$ regularizer comprising an information-weighted encoder-Jacobian penalty (rate side) and a Frobenius contractive penalty on the end-to-end reconstruction map (distortion side). A finite-difference Lipschitz analysis shows corresponding sensitivity reductions of ~12% for the encoder and ~8% for the end-to-end codec, while an exact-JVP test further shows preferential contraction of high-information directions. Code and trained models are available at https://github.com/Googolxx/NoiseRegu-LIC.
Read the full abstract
General Mathematics
Double sums involving quotients of binomial coefficients
This paper is the third part of our detailed investigation of double sums where the inner sum is binomial but incomplete. We prove many identities for these double sums involving quotients of binomial coefficients and additional parameters. In addition, we evaluate some quadratic incomplete sums using Legendre polynomials.
Read the full abstract
Numerical Analysis
Operator-Adapted Multiscale Kernels for High-Contrast Elliptic Inverse Problems
We consider the joint recovery of the state and diffusion coefficient in a high-contrast elliptic equation from sparse, noisy state measurements. We seek the coefficient in a parameterized family defined relative to a prescribed reference coefficient. We construct a generalized multiscale finite element basis from the reference coefficient and use it to define a finite-rank kernel for the state. A weak Galerkin residual enforces the equation without pointwise differentiation across material interfaces. For each trial coefficient, the state is obtained by solving a linear least-squares problem. Variable projection eliminates the state coordinates from the objective, and the reduced gradient is evaluated without an adjoint PDE solve. We split the state reconstruction error into multiscale approximation, penalty relaxation, and parameter estimation terms. Under an admissible patch decomposition, the squared multiscale approximation error is bounded by a local spectral tail estimate with a contrast-independent interpolation constant. For a fixed coefficient and discretization, the penalty relaxation error is of order $O(\varepsilon^2)$ when the residual weight is proportional to $\varepsilon^{-2}$. The parameter contribution is bounded by local Lipschitz continuity of the projected state. In 50 synthetic tests based on a reference medium of contrast $10^4$, the multiscale space with the weak residual gives mean relative state and log-coefficient errors of 2.37% and 4.43%, respectively, the lowest among the four configurations tested.
Read the full abstract
Image and Video Processing
Frequency-consistent synthesis and selective transfer for medical image analysis
Complementary imaging is not always available, and realistic synthesis does not necessarily improve downstream analysis. Here we introduce a training framework linking medical image synthesis to task learning through adaptive frequency consistency and selective feature transfer. Image-specific frequency constraints encourage consistency of structure and detail. Features are transferred only when a teacher using synthetic images achieves a lower task-error score than the student by a task-specific margin on a labelled training sample. Evaluations across five datasets show improved translation fidelity with three adversarial and diffusion-based translators trained without paired supervision, alongside gains in brain-tumour segmentation on tumour-containing slices and lesion centre-point detection. All 105 reported paired comparisons yielded point estimates favouring the proposed methods, including 36 external translation comparisons on post?treatment MRI without retraining. These findings support synthetic modalities as task-oriented complements to acquired scans when complementary imaging is unavailable.
Read the full abstract
Algebraic Geometry
On deformations of foliated complex analytic structures IV-1: Obstructed and unobstructed foliated deformations
We study obstructions of deformations of foliated complex analytic structures. We describe obstructions in terms of Cech resolutions. From foliated complex analytic families in Part I and unfolded complex analytic families in Part III, we deduce the integrability conditions. We identify infinitesimal deformations and obstructions in terms of Dolbeault resolutions or Dolbeault type bicomplexes from the integrability conditions. Then we describe the correspondences of infinitesimal deformations and obstructions between Cech resolutions, and Dolbeault resolutions or Dolbeault type bicomplexes.
Read the full abstract
Theoretical Economics
Analysis of Supply Information, Code Bidding, and Efficiency in Repetitive Simultaneous Ascending Online Auctions: A Study of India's eNAM Digital Agricultural Market
This paper analyzes a Repetitive Simultaneous Ascending Auction (RSAA) mechanism employed in India's eNAM digital agricultural market. RSAA is characterized by full seasonal supply transparency and a strategic seller option to reject equilibrium bids. We develop a model of the RSAA game with inter-round strategic dependence and endogenous seller entry, and we find conditions for collusion formation. Using transaction-level data, we empirically validate the emergence of code bidding, a tacit collusive strategy sustained by low bidder concentration and shared supply information. The results show that demand-side coordination systematically reduces price competition. Moreover, the coexistence of supply transparency and strategic rejection introduces significant allocative inefficiencies despite the ascending-price format. These findings offer novel insights for auctions and market design, particularly concerning collusion vulnerability and efficiency in agricultural and digital commodity markets.
Read the full abstract
Computers and Society
The Challenge of Modeling the Complexity of Use for the Measurement of Digital Maturity in Education
Measurement of teachers' digital maturity can be useful in diagnosing technology adoption in education and in supporting teachers' professional development. However, there is a lack of easily usable tools in the existing literature. By investigating the most effective ways to measure teachers' digital maturity, our research aims to fill this gap. The focus of this article is on a data-driven approach to the assessment of digital maturity in education. Following a process similar to teaching analytics (data collection, data modeling, data understanding and visualization, and contextual understanding), we draw on diary data collected from 12949 French primary school teachers using a virtual learning environment (VLE) during the 2022–2023 school year. Specifically, the paper presents results related to modeling and visualizing data. To this end, we propose a critical analysis of two automatic classification techniques, one supervised rule-based and the other unsupervised (hierarchical clustering). In addition to providing different scales to measure, we have proposed a new approach to diagnose and visualize digital maturity. In this way, the method provides insight into the diversity and intensity of technology use, contributing to a better understanding of digital maturity and providing a practical tool for assessing day-to-day teaching practices.