Peer Revenue Now accepting positions

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

All of arXiv

Friday 6 November

Computer Vision

Overview of the ImageCLEFmedical 2026 GANs task: Exploring fingerprint detection, privacy leakage, and synthetic medical image generation

The 2026 ImageCLEFmedical GANs Task Exploring Fingerprint Detection, Privacy Leakage, and Synthetic Medical Image Generation continues to investigate privacy and security concerns around using patient data to generate synthetic medical images, now in its fourth edition. This edition introduces three complementary subtasks: the first extends prior editions by asking participants to detect which real images were used in training a Generative Adversarial Network to produce given synthetic outputs; the second is a novel subtask requiring participants to identify which latent vector produced each synthetic image, probing the faithfulness and reversibility of the latent-to-image mapping; the third asks participants to generate privacy-preserving CT slices, quantifying the trade-off between image fidelity and patient privacy. Ground-truth annotations and benchmark datasets of real and GAN-generated lung CT slices are provided for all subtasks. Evaluation is based on Cohen's Kappa for Subtask 1, matching accuracy for Subtask 2, and a combination of FID and Privacy Preservation Score (PPS) for Subtask 3. Overall, 37 teams registered for the task, with 10 teams submitting runs across subtasks and 6 teams submitting working notes papers. This paper presents an overview of the task setup, datasets, evaluation metrics, and summarises the approaches and results of participating teams.
Read the full abstract

All authors Alexandra-Georgiana Andrei, Mihai Gabriel Constantin, Mihai Dogariu, Dzmitry Karpenka, Yuri Prokopchuk, Ahmedkhan Radzhabov, Dan-Cristian Stanciu, Liviu-Daniel Ştefan, Vassili Kovalev, Henning Müller, Bogdan Ionescu

Machine Learning

From LoRA to Language: Recovering Skill Documents from Adapters

Agent skills are short documents that teach a language model how to handle a class of tasks. To save context, these documents are increasingly compiled into low-rank adapters (LoRA), which keep the behaviour but lose the text: an adapter is hard to read, edit, audit or move to another model. We ask whether the skill document can be written back from the adapter and its frozen base model, and we score each recovered document by the share of the original's key instructions it states and by the tasks it solves. Reading the weights alone recovered little with the methods we tried; a learned map from weights to text passed on about one bit of the document. Running the adapter worked much better. The Executing Writer, a small writer trained on top of the adapted model, recovers 39-44% of the key instructions of held-out skills, against 1.3-1.5% with another skill's adapter, and on adapters fitted to 41 external skills its documents restore 85-91% of the original's task gain. These strongest results are on adapters whose training stored the document; adapters trained on behaviour alone are much harder for trained writers (1-2% of instructions). For these adapters we test an agent that runs the base and adapted models and writes a skill from where they differ. On ALFWorld its skills recover 58-71% of a LoRA's gain and carry over to models that cannot load the LoRA, lifting Llama-3.1-8B from 5% to 55%. Whether such a skill comes from the adapter depends on what the adapter learned: when its behaviour cannot be guessed, the adapter is the source; when it can, a reward loop without the adapter does as well.
Read the full abstract

All authors Saranya Dintakurthy Swadha, Pavan C Shekar, Arpit Jain, Aswanth Krishnan

Dynamical Systems

Every countable group has the nonsingular weak Rokhlin property

We prove that for every countable group $\Gamma$, in the space of nonsingular actions by $\Gamma$, there is an action whose orbit under conjugation is dense. This is the nonsingular analogue of the result of Glasner, Thouvenot, and Weiss that every countable group $\Gamma$ is weak Rokhlin, i.e., there is a dense orbit in the space of probability measure preserving $\Gamma$-actions.
Read the full abstract

All authors Garrett Ervin, Boglárka Gehér, Anett Imolay-Kocsis, Máté Pálfy, Jenna Zomback

Theoretical Economics

Optimal robust comonotonic risk sharing under operational constraints

Risk-sharing arrangements such as reinsurance programs and catastrophe pools must satisfy operational constraints, including capacity limits, attachment bands, retention rules and cession limits. Under distortion risk measures, the unconstrained optimum assigns each loss layer to the participant for whom that layer is least costly, but operational constraints break this layerwise rule, and there is no general method to compute the constrained optimum. We show that, for comonotonic allocations, the participants' distortion risk measures and the operational constraints are linear in the amount of each loss layer held by each participant, so the constrained risk-sharing problem on a discretized loss is a linear program that standard software can solve. Its dual gives supporting prices for the loss layers and the marginal value of relaxing each operational constraint. Where the set of binding constraints changes, the constrained optimal value can have a kink, and the smallest and largest optimal multipliers then give the one-sided effects of relaxing and tightening a constraint. To protect the allocation against disagreement between loss models, we also consider the robust version of the risk-sharing problem, that is, minimizing the worst-case total risk assessment over finitely many probability scenarios. This robust problem is again a linear program, and we quantify the cost of protection against model uncertainty from its optimal value. We illustrate the approach with a catastrophe reinsurance tower.
Read the full abstract

Signal Processing

Scalable BLE Channel Sounding Localization with Ranging-Assisted PDoA

Bluetooth Channel Sounding (CS) enables accurate phase-based ranging, but multi-anchor localization requires sequential ranging exchanges that limit scalability and increase measurement latency. This paper proposes Ranging-Assisted Phase Difference of Arrival (RA-PDoA), a scalable BLE localization architecture combining one active anchor performing a standard two-way CS exchange with multiple passive receive-only anchors observing the same transmissions. A channel-recovery framework is derived that enables passive anchors to reconstruct a channel response suitable for super-resolution delay estimation, allowing multipath-robust ranging without participating in a two-way exchange. Coarse arrival-time information is further used to bound the position estimate and reject inconsistent range solutions. A comprehensive evaluation characterizes the impact of multipath propagation, body shadowing, synchronization requirements, oscillator impairments, anchor placement, and deployment inaccuracies, thereby quantifying the practical infrastructure requirements for scalable BLE localization. Simulation results in modeled indoor environments show that RA-PDoA reduces the 90th-percentile localization error from 175.3 cm to 95.5 cm under body shadowing while maintaining an airtime requirement independent of the number of anchors. The results demonstrate that a single active ranging link enables scalable and multipath-robust BLE Channel Sounding localization.
Read the full abstract

Quantum Physics

Symmetry-Resolved Criticality in Dissipative Quantum Phase Transitions

Symmetry breaking in mixed quantum states exhibits a rich strong-to-weak hierarchy, yet the use of symmetry as an operational probe of dissipative criticality remains largely unexplored. In this work, we develop a symmetry-based framework that connects semiclassical phase space, trajectory-level dynamics, and ensemble responses, and demonstrate it in the two-dimensional dissipative XYZ model. We show that symmetry breaking manifests semiclassically as a bifurcation of the Husimi-Q distribution and dynamically as a sharp change in a trajectory-level symmetry residual. At the ensemble level, the weak-symmetry sensitivity to a perturbed symmetry operator exhibits pronounced finite-size scaling, yielding an estimate of the dissipative critical point. Finally, we propose a trajectory-based quantum-circuit algorithm in which mid-circuit measurements condition quantum jumps and allow the proposed indicators to be extracted from circuit observables. Our work provides a symmetry-centered route to diagnosing dissipative phase transitions and critical scaling in open quantum many-body systems.
Read the full abstract

All authors Xingli Li, Yan Li, Jinghui Pi, Wenlin Li, Shuai Geng, Lin Shang, Jiasen Jin, Yangqian Yan

Earth and Planetary Astrophysics

CO gaps in cold transition disks around low-mass stars

Transition disks with large inner dust cavities are promising targets to study planet formation, as these disks are likely hosting massive planetary companions. The structure of gas and dust inside the cavity are crucial for understanding the potential of planet formation in the regions. We investigate the structure of gas and dust in a sample of three cold, inclined transition disks around K and M stars without evidence for an inner dusty disk. We present new high-resolution ALMA Band 6 data of both 230 GHz continuum at 0.02" (~3 au) resolution and the 12CO 2-1, 13CO 2-1, C18O 2-1 isotopologues at 0.12" (~20 au) resolution. We analyze the radial profiles, compute the brightness temperatures, and derive estimates of the gas and dust content of the inner disk at 1 au. All three disks show dust cavities of 30-40 au in radius. The CO emission extends inside the cavity with progressively smaller CO cavity radii for the more abundant isotopologues. The 13CO brightness temperature profile is similar to that of other transition disks and full disks. No inner dust disk is detected in the ALMA continuum, consistent with the lack of near-infrared excess. Considering the accretion rates, the gas-to-dust ratio in the inner disk at 1 au is estimated to be as high as 10^5. No significant differences in the brightness temperature or in the gas-to-dust ratio are found between our disks and other transition disks around more luminous stars, contrary to expectations. This highlights the importance of individual-target studies through thermochemical modeling or multi-transition analyses to constrain the dust and gas structures in the cavity. The observed disk structures could be explained by super-Jovian companion on eccentric orbits, but further modeling is needed to confirm this conclusion.
Read the full abstract

All authors A.A. Rota, N. van der Marel, E.F. van Dishoeck, T. Muto

Machine Learning

Not All Entropy Is Equal: Reward-Conditioned Entropy Allocation in Reinforcement Learning with Verifiable Rewards

Policy entropy is widely used to diagnose exploration and collapse in reinforcement learning with verifiable rewards (RLVR), yet its magnitude does not consistently track reasoning performance across optimization objectives. We argue that this ambiguity arises because global token entropy conflates uncertainty in successful and failed trajectories, diversity across correct solution strategies, and variation among realizations of the same strategy. We introduce Reward-Conditioned Entropy Allocation (RCEA), a hierarchical diagnostic framework that separates complete-trajectory entropy from average token entropy, conditions uncertainty on verifier outcome and process validity, measures correct-solution-mode retention, and connects these population-level quantities to local covariance-driven updates. In controlled sweeps over six RLVR objectives at two model scales, entropy trajectories do not induce a stable ordering of reasoning accuracy. Token entropy is higher for verifier-failed than successful trajectories under both broad separately aggregated estimates and a common-question paired analysis, whose confidence intervals exclude equality at 10 of 12 displayed endpoints. A local covariance approximation predicts the direction of adjacent-checkpoint entropy change in 95.5% of nonfailed transitions; a reward-stratified endpoint proxy further shows that positive chosen-token covariance is more tail-concentrated within failed prefixes. Every reportable trained endpoint exhibits some contraction in independently audited correct-solution-mode breadth, whose positive association with large-\(k\) coverage remains exploratory. These results show that entropy collapse should be diagnosed by which uncertainty is removed, rather than by global entropy alone.
Read the full abstract

Data Structures and Algorithms

Additive $\Omega(\log^* OPT)$ Gap for Bin Packing with Three Items per Bin

We prove that bin packing is NP-hard to approximate within an additive factor of $o(\ls\OPT)$, even when all item sizes are arbitrarily close to $1/3$. This gives the first lower bound for bin packing with an additive factor that depends explicitly on the optimum value. In addition, we give a polynomial-time reduction from any bin packing instance into one with item sizes arbitrarily close to $1/3$, preserving its additive configuration-LP gap up to constant factors. Thus, in the sense of additive configuration LP gap and up to a constant, we confirm the conjecture that 3-partition instances are the hardest bin packing instances. In particular, we obtain an unconditional $\Omega(\ls\OPT)$ gap for 3-partition instances. We build on the recent OpenAI construction that uses five items per bin and rules out any constant additive gap.
Read the full abstract

Cryptography and Security

Logit Laundering: Evading Data-Use Auditing in Large Language Models

Data-use auditing in Large Language Models (LLMs) aims to determine whether the given data has been used for training, thereby supporting data ownership safeguards. A dominant solution is data watermarking, which embeds traceable signals into training data and detects them in model outputs. Recent studies have proposed many watermarking-based auditing methods in LLMs. However, systematic investigation from an adversarial perspective remains limited: non-compliant LLM providers would use unauthorized copyrighted datasets for training their own models to enhance model capabilities, while attempting to evade detection by auditors, thereby avoiding legal liability. To fill this research gap, we propose a novel evasion framework, termed Logit Laundering, where the non-compliant LLM provider manipulates model outputs at inference time to undermine detectable watermark signals. Specifically, we design a logit reshaping module to suppress watermark-induced statistical signals by integrating a statistical bias estimator as a plug-and-play component. This design keeps the model parameters unchanged and performs manipulation only in the logit space, thereby enabling auditing evasion while minimizing the impact on model's learned knowledge. To further balance evasion effectiveness and model utility, we introduce a confidence-aware scaling module to adaptively calibrate the suppression strength based on next-token prediction confidence. Extensive evaluations show that Logit Laundering can successfully evade existing various state-of-the-art auditing methods, significantly attenuating detectable signals, while preserving model utility on downstream tasks. Our work highlights the need for exploring more robust auditing methods to guard against evasion behaviors.
Read the full abstract

All authors Ruihan Hu, Wei Luo, Yu-Ming Shang, Jiakai Wang, Chuxuan Zhang, Mei Li, Xi Zhang, Meikang Qiu

Multiagent Systems

Incentive Pathologies in a Live LLM Agent Arena: An Empirical Study of Concentration, Reward-Rule Implementation Discrepancy, and Network Dilution

We audit one live LLM agent arena (AITRAP; 22 autonomous agent seats, ~96,000 reasoning nodes) over nine days (Oct 16-25, 2026), during which the platform deployed two bundled incentive changes: a progressive tax reform (V2.9.3, Oct 19) and a seed-inversion with explorer-side action update (V3.0, Oct 21). These serve as before-and-after markers, not cleanly isolated interventions. Four exploratory predictions are tested against ten data frames (F1-F10) using the Gini coefficient, Top-K concentration share, code-level reconstruction of the continuation-bonus mechanism, and the Mann-Kendall trend test. The Gini analysis is inconclusive at 95% confidence (bootstrap intervals straddle 0.7). The Top-4 node-activity-score share falls from 97.2% (Oct 19, pre-V2.9.3) to 49.4% (Oct 25, post-V3.0; 56.7% under a merge-seat assumption). The continuation bonus, announced as depth-based x10, resolves at code level to a four-condition anti-abuse gate paying x2.85 in the typical case and permanently blocking the self-continuation it superficially rewards; the gate was active ~40 hours before the V3.0 announcement. The events-per-node ratio declines significantly (Mann-Kendall S=-45, z=-3.94, p=8.3e-5) from 7.36 to 7.069. We discuss implications for incentive designers, the announcement-implementation gap, and public peer review as an event record.
Read the full abstract

Analysis of PDEs

Isolated yet highly unstable 2D Euler equilibria

We construct steady solutions of the two-dimensional Euler equations on smooth annular domains that are isolated, modulo scaling, in the $L^\infty$ vorticity topology. They are highly unstable: any Yudovich trajectory remaining sufficiently close to such an equilibrium on an open time interval is a stationary multiple of it and, moreover, at fixed energy, enstrophy and circulations, distinct nearby trajectories cannot return even weakly to the equilibrium. Further, on any non-circular annulus, we show that this kind of isolated, highly unstable equilibria occurs arbitrarily close in the Yudovich norm to smooth equilibria which are themselves non-isolated.
Read the full abstract

Metric Geometry

Sharp isoperimetry in groups of polynomial growth

We study weighted edge boundary and outer vertex boundary on Cayley graphs of polynomial growth of degree $Q>1$. For every finite symmetric generating set, we prove that each exact isoperimetric profile satisfies $I(n)=c n^{(Q-1)/Q}+O(n^{(Q-1)/Q-\beta})$, where $\beta=1/(8Q^2)$. We identify the leading constant with the isoperimetric constant of a horizontal perimeter on the limiting Carnot group, after accounting for the counting density. Its integrand is a norm with polyhedral unit ball, determined by a convex minimization with one real variable for each coset of a torsion-free nilpotent subgroup. For every asymptotically minimizing sequence, we translate its vertex sets so that the characteristic functions of the associated rescaled regions are precompact in $L^1$. The limiting regions minimize continuum perimeter at fixed volume. The exact-cardinality estimate requires no regularity assumption on these regions. For virtually abelian groups, we prove a quantitative stability estimate using a discrete approximation to the Euclidean Wulff body. The relative symmetric difference from a translate is bounded by the square root of the sum of the relative boundary deficit and a cardinality error. We also characterize when bounded additive changes of Cayley metric preserve the leading vertex constant. Failure occurs already in rank two; we construct pairs with the same measured asymptotic cone whose leading constants have an unbounded ratio as the rank grows.
Read the full abstract

High Energy Physics (Theory)

Yang-Mills theories with scalar matter: a first foray using the Curci-Ferrari model

The Curci-Ferrari (CF) model provides an efficient perturbative framework for describing several aspects of the infrared (IR) dynamics of Yang-Mills (YM) theories. We extend this framework to YM theories coupled with scalar matter and investigate its main properties, with particular emphasis on the one-loop two-point correlation functions of scalars, gluons, and ghosts. We analyze the renormalization group (RG) structure of the model, computing its beta functions and studying the corresponding RG flows for different scalar masses. We establish the existence of IR-safe trajectories along which perturbation theory remains under control and identify a novel fixed-point, while recovering the standard CF results in an appropriate limit. We then confront our results with lattice data through a simultaneous fit of the scalar, gluon, and ghost propagators, finding excellent agreement. Our results indicate that, within the approximations considered here, the CF framework gives a successful perturbative description of the IR regime of YM theories coupled with scalar matter.
Read the full abstract

All authors Gustavo L. L. W. Levy, Philipe De Fabritiis, Marcela Peláez, Urko Reinosa

Optics

Reconstructing intracell polarization with calibrated sapphire-window optics for collimated blue-light generation in a rubidium cell

We study the polarization dependence of collimated blue light (CBL) generated near 420~nm by resonant four-wave mixing in thermal rubidium vapor in a sapphire-window cell driven at 780 and 776~nm. Sapphire windows can transform the polarization of pump light before it reaches the vapor, complicating the interpretation of polarization-dependent nonlinear emission. We measured the 780-nm pump Stokes state at an external before-cell reference plane to calibrate the common HWP/QWP preparation optics and recorded 266 off-resonant external output Stokes states after the complete sapphire-window rubidium cell at 90, 100, and 110~$^\circ$C in two propagation orientations. Direct paired-wavelength measurements show a mean before-cell 780/776 separation of $4.29^\circ$ and temperature- and orientation-specific complete-cell rotations of $22.20^\circ$–$32.02^\circ$, with rms residuals of $4.83^\circ$–$12.87^\circ$. Using this calibrated co-prepared pump description, we analyzed 2574 measurements of total spectrally filtered 420-nm CBL intensity. Across the six temperature–orientation data sets, a complete nine-coefficient second-order response yielded cross-validated $R^2$ values of 0.636 to 0.828. These paired-wavelength measurements establish, for each temperature–orientation data set, a fixed relative Stokes-space map between the two co-prepared pump trajectories, with the residual of this mapping defining the experimental uncertainty of the correlated two-pump reduction used in the response model. Thus, the resulting framework combines a reconstructed vapor-entrance polarization coordinate with direct two-wavelength calibration of the assembled cell and provides a practical basis for polarization-controlled nonlinear-optical experiments in sapphire-window vapor cells.
Read the full abstract

Analysis of PDEs

Asymptotic behavior of solutions to the critical $p$-Laplace equation near isolated singularities

In this paper, we study the asymptotic behavior of weak solutions to the following critical $p$-Laplace equation \[ -\Delta_p u=u^{p^*-1}, \quad u\ge0 \quad\hbox{in }B_1\setminus\{0\}, \] where $\Delta_pu=\diver(|\nabla u|^{p-2}\nabla u)$ is the classical $p$-Laplacian with $1<p<n$, $p^*=\frac{np}{n-p}$ is the Sobolev critical exponent and $B_1\setminus\{0\}\subset \mathbb{R}^n$ is the punctured unit ball. We develop a nonlinear potential-theoretic method to establish sharp upper bounds for solutions, without using the method of moving planes or moving spheres. Furthermore, we use Steiner symmetrization to study the radial symmetry of global singular solutions, and thereby classify all Fowler singular solutions. Finally, we prove that every local singular solution is asymptotic to a global Fowler singular solution.
Read the full abstract

Statistical Methodology

FlexRL: an R package for Flexible Record Linkage and Linked Data Quality Assessment

FlexRL is an R package for probabilistic record linkage and inference on linked data. It provides a method to link observations from the same entities across two data sources that lack a unique identifier but share partially identifying variables. This method fits a hierarchical latent variable model of the data generation process, rather than classifying pairs of records from the comparison of their values, and accommodates registration errors and variables changing over time. Since linkage may affect subsequent analyses, the package also provides tools to assess the linked data quality as functions of the linkage decision rule. Two false discovery proportion estimators are implemented: a probabilistic approach contingent on the record linkage model and an augmented approach relying on an auxiliary estimation using synthetic data. Several measures of divergence of the linked sample from the source population are provided. These tools are agnostic to the underlying record linkage model and can be applied to any method. Together, they connect the output of the linkage to the validity of the subsequent analyses. The implementation combines standard R functions and S3 methods with Rcpp computations, resulting in fast execution times and a low memory footprint.
Read the full abstract

Functional Analysis

Trace and Regularized Index of Callias Operators on Hyperbolic Space

Let $d \geq 3$ be odd. On hyperbolic space $\mathbb{H}^d$, we consider Callias-type operators, i.e. Dirac-Schrödinger operators of the form $D = ic \nabla + A$, where $c \nabla$ is the Dirac operator and $A$ a function with values in the self-adjoint operators, and provide an explicit formula for the iterated trace $\text{tr}_{L^2(\mathbb{H}^d,H)} \text{tr}_{S}( e^{-tD^*D} - e^{-tDD^*})$, $t>0$, that appears in the formulation of the Witten index. Thereby, the operator $D$ is not assumed to be Fredholm.
Read the full abstract

Plasma Physics

Different exact-law formulations to calculate the energy cascade rate in magnetohydrodynamic turbulence with and without mean magnetic fields

At scales much larger than the ion inertial length and ion-gyroscale, space-plasma turbulence is often described within the framework of incompressible magnetohydrodynamics (MHD). Using high-resolution direct numerical simulations with $1024^3$ grid points, we perform the first systematic numerical comparison of three exact-law formulations for the total energy cascade in homogeneous, but not necessarily isotropic, MHD turbulence: the Yaglom-Monin (YM), Banerjee-Galtier (BG), and two-point correlator formulations. Despite directional variations, the angle-averaged energy transfer rates for all three formulations show excellent agreement for zero, moderate, and strong background magnetic fields. These results clearly suggest the usefulness of alternative BG and correlator formulations for estimating the energy cascade rates in turbulent space plasmas when the required spatial derivatives are accessible through multi-spacecraft measurements. We also argue that the isotropic Politano-Pouquet (PP) formulation does not provide a true directional transfer rate and therefore its application as a single-direction energy transfer rate estimator leads to inaccurate values, especially under anisotropic conditions. Decomposing the exact formulations into RMHD and non-RMHD contributions, we also show that the RMHD cascade rates cannot be obtained merely by increasing the strength of the mean magnetic field in incompressible magnetohydrodynamic turbulence. Finally, a term-by-term analysis of different formulations reveals how different nonlinear interaction channels coexist and collaborate with each other to maintain the same energy cascade rate across the inertial scales.
Read the full abstract

Computer Vision

The Role Of Recurrent Refinement In Medical Image Segmentation

While feed-forward segmentation networks perform well on large cohorts, they degrade in constrained clinical settings with few training samples and imaging artifacts. Human vision, by contrast, remains robust under these conditions, a property partly attributed to the recurrent and feedback connectivity of the visual cortex, which refines visual information over time. Inspired by this, we append lightweight recurrent refinement blocks to segmentation backbones, namely Conditional Random Fields (CRFs), modern Hopfield networks and convolutional GRUs, and study their behaviour as the number of iterations increases at test time (including beyond the training horizon). Evaluated on two open-source datasets, Kvasir- SEG and ARCADE, our recurrent blocks add less than 1.5% parameters and yield gains on clean data (up to 1.4 pp) and under noise (up to 17 pp) while reducing HD95. Code is available at https://github.com/joaodsmarques/recurrencehelpsegmentation .
Read the full abstract

Signal Processing

Pinching Antenna Systems (PASS) for Air-Ground Integrated Sensing and Communications (ISAC)

This paper proposes a multi-user single-target air-ground integrated sensing and communication (ISAC) system, in which a ground pinching-antenna system (PASS) transmits communication and sensing signals and an unmanned aerial vehicle (UAV) receives the target echo. The communication and sensing beamforming vectors, pinching antenna positions, and UAV position are jointly optimized to minimize the weighted Cramér-Rao bound (CRB). A penalty dual decomposition (PDD) algorithm is developed for the resulting nonconvex problem. The auxiliary variable and beamforming subproblems admit globally optimal solutions in semi-closed form, while the positions of the pinching antennas and the UAV are updated by successive convex approximation (SCA). Under mild conditions, every accumulation point of the sequence generated by our proposed algorithm is a Karush-Kuhn-Tucker (KKT) point of the reformulated problem. Numerical results show that the proposed PASS-UAV design achieves a lower weighted CRB than conventional multiple-input multiple-output (MIMO) and PASS benchmarks.
Read the full abstract

All authors Chaoran Chen, Haochen Li, Jincheng Zhu, Zhiwen Pan, Mark F. Flanagan

Quantum Physics

Optimal performance of driven-dissipative many-body quantum batteries beyond complete nonreciprocity

Proposals for nonreciprocal quantum batteries identify the point of complete nonreciprocity, where coherent and reservoir-mediated couplings balance, as the operating point for enhanced charging. We show that, for batteries built from interacting qubits, this point is, in general, suboptimal. The optimal design is set by a more intricate interplay between the coherent coupling and the local and collective dissipation rates, of which complete nonreciprocity is only a special case. We consider a many-body battery charged by a driven many-body charger, with each qubit coupled to a local bath and the complete setup immersed in a shared engineered reservoir. For a weakly driven two-qubit setup, we prove the Dzyaloshinskii-Moriya interaction to be the optimal coupling, and show that the battery capacity is maximized at a coupling strength fixed by the local and collective dissipation rates, which coincides with complete nonreciprocity only for a specific balance between them. The shared reservoir enhances the capacity for couplings weaker or stronger than both local dissipation rates, and can increase the extractable work even where it reduces the capacity. At strong drive, with the Dzyaloshinskii-Moriya interaction between the charger and the battery, the capacity is bounded by $1/2$. For qubits arranged in a star network with a single central battery qubit, these results remain qualitatively unchanged, while the maximum capacity of a battery charged by $N_A$ weakly driven charger qubits saturates with $N_A$ in the presence of the shared reservoir, approaching its limit as $N_A^{-1}$. In contrast, the maximum capacity of $N_B$ qubit batteries charged by a single-qubit charger decays as $N_B^{-2}$.
Read the full abstract

Machine Learning

DexDICE: Reconciling Safety and Performance via Dual-Expert Offline Imitation Learning

Ensuring high returns while providing strong safety guarantees in offline imitation learning (IL) is a fundamental challenge: existing approaches either rely on explicitly specified cost functions, which are difficult to design for complex real-world safety constraints, or require expert demonstrations that are both safe and high-performing, which are often too costly to collect. We introduce dual-expert imitation learning, which leverages two complementary data sources: (i) safety-compliant demonstrations that faithfully follow safety guidelines but may be suboptimal, and (ii) performance-oriented demonstrations that achieve high returns but may violate safety guidelines. To address this problem, we propose DexDICE (Dual-expert offline imitation learning via stationary DIstribution Correction Estimation). Instead of conventional distribution constraints, DexDICE combines DICE-style offline learning with safe support constraints, enabling agents to exploit high-return behaviors while remaining within safety-compliant regions. We prove that the ideal solution satisfying this constraint exactly has zero unsafe-region visitation with high probability. We evaluate DexDICE on a re-curated DSRL benchmark and validate it on a real-world mobile-robot navigation task trained directly from human teleoperation demonstrations, where it substantially outperforms baselines and remains robust to data scarcity and quality degradation.
Read the full abstract

All authors Seokin Seo, Sung Kuk Shyn, Minji Seo, Yisoo Lee, Jongmin Lee, Hongseok Yang, Kee-Eung Kim

Statistical Methodology

A hybrid model for zero-inflated proportion data

This paper proposes a hybrid model for analyzing zero-inflated proportion data by distinguishing two types of zeros. The first type arises from sampling variability, whereas the second reflects intrinsically unsuitable covariates. Accordingly, the hybrid model combines a regression component for suitable observations with an estimator of the suitability region that identifies unsuitable covariates. Using a debiasing technique, we establish the consistency of both the regression and classification components, together with the asymptotic normality of the regression coefficient and a convergence rate for the support estimator. Comprehensive simulation studies verify our theoretical findings. We further apply the model to two real datasets, where it outperforms most competing methods.
Read the full abstract

All authors Yangzi Zheng, Qian Li, Binyan Jiang, Pinjun Dong

Statistical Methodology

Debias Online Expectile Regression for High-Dimensional Stream Data and Statistic Inference

Expectile regression is a powerful tool for capturing heterogeneous structures in data and has attracted increasing attention in large-scale data analysis due to its computational efficiency. However, statistical inference for high-dimensional online expectile regression remains underexplored. In this paper, we propose a debiased online expectile regression method for high-dimensional streaming data. The proposed method incorporates Lasso regularization to perform variable selection and alleviate estimation instability caused by high dimensionality. It further adopts an online updating strategy in which recursively updated cumulative statistics are used to approximate the global gradient, and a proximal gradient descent algorithm is employed to update model parameters. To enable valid inference when only the current batch and historical summary statistics are available, we also develop an online precision matrix estimation procedure. We establish the consistency and asymptotic normality of the proposed debiased estimator. Extensive numerical studies and real-data applications show that the proposed method achieves performance comparable to the oracle and offline estimators while maintaining low storage cost.
Read the full abstract

Numerical Analysis

EDP-PI-DOSnet: An Energy Dissipation Preserving Neural Solver for Two-Phase Flow Problems

Preserving the energy dissipation properties of gradient flow equations is important for the stability and reliability of numerical approximations, particularly in long-time simulations. In this paper, we propose a class of energy-dissipation-preserving physics-informed deep operator splitting networks (EDP-PI-DOSnets) for solving gradient flow equations. The proposed framework incorporates classical structure-preserving numerical schemes into neural network architectures, allowing the corresponding discrete energy dissipation laws to be preserved. In particular, suitable constraints are imposed on the convolution kernels to ensure that the learned discrete operators are symmetric and negative semidefinite. Furthermore, several network formulations are developed based on the Crank–Nicolson (CN), scalar auxiliary variable Crank–Nicolson (SAV/CN), and relaxed discrete variational derivative (R-DVD-1) schemes. Meanwhile, a physics-informed loss function based on the residuals of the governing equations is employed. Numerical experiments show that the predicted solutions are in good agreement with the reference solutions while satisfying the corresponding discrete energy dissipation laws. The results also suggest stable behavior during long-time recursive prediction and the potential applicability of the proposed framework to two-phase flow problems.
Read the full abstract

Statistical Methodology

Navigating Selection Bias in Genome-wide Association Studies (GWAS)

Selection bias arises when inclusion in a genetic study depends on traits, or causes of traits, that are associated with genotype. It is present in volunteer-based biobanks, case-control designs and studies of disease progression, and genome-wide association study (GWAS) summary statistics from affected samples can propagate this bias into post-GWAS analyses, such as polygenic scores (PGS), Mendelian randomization (MR) and genetic correlation analyses. Despite broad potential impact of selection bias it is sometimes ignored in genetic studies. Many methods exist to detect and correct selection bias, but they rely on different assumptions and require specific data. We organise ten methods applicable to GWAS by the decisions an analyst faces: whether to diagnose or correct, which data are available (individual-level or summary-level), and which form of selection is operating (non-representative participation, case-control ascertainment, index events, sex-differential participation or survival to recruitment). For each method, we describe its principle, applicability, limitations and data requirements. A companion GitHub repository implements nine methods in R and prepares input for the external SPREG software. It provides worked examples using simulated data with known true effects for each form of selection. An interactive tool supports method selection. We conclude with open methodological challenges.
Read the full abstract

All authors Ciarrah-Jane S Barry, Tahereh Gholipourshahraki, Zoltán Kutalik, Bjarni J Vilhjálmsson, Esben Agerbo

Analysis of PDEs

Kolmogorov Obstacle Problems: Euclidean Free Boundary Regularity under a Cone Condition

We study obstacle problems for Kolmogorov-type operators that diffuse in one variable and transport in another. Such problems describe American options on running averages. Under a cone estimate, a strong monotonicity condition on the solution in the diffusive direction, we show that the free boundary is a graph in the diffusive direction with Hölder continuous first derivatives in the Euclidean sense, and that every free boundary point is regular. The key observation is that a Euclidean Lipschitz free boundary is flat at the intrinsic scale. For the American fixed-strike Asian option we verify the cone estimate, so that its exercise boundary has the same regularity.
Read the full abstract

Probability

Fluctuations of heat kernels and exit times on random graphs and application to the critical long-range percolation

Let $\mathcal{G}$ be a graph equipped with random conductance. We give sufficient conditions for the Markov chain on $\mathcal{G}$ to exhibit large scale fluctuations in its on-diagonal heat kernel and in exit times. The conditions also imply non-tightness of the height process w.r.t. the quenched measure, while it is tight w.r.t. the annealed measure. As an application of the general theory, we prove large scale fluctuations for the simple random walk on $1$-dimensional critical long-range percolation.
Read the full abstract

Statistics Theory

Robust Bayes-like estimation of the distribution of a determinantal process

We consider the problem of estimating the distribution of a determinantal process (DP) on a measured space from the observation of $n$ independent copies of it. Our estimator is obtained by sampling from a $T$-posterior distribution associated with a suitable prior, and we establish non-asymptotic risk bounds for the Hellinger loss which hold with high probability. These bounds remain valid when the data are only independent rather than i.i.d., and when the model is misspecified, which makes the estimator robust, for instance, to the presence of a small fraction of outliers. For determinantal projection processes of rank $k$ whose kernels are associated with $k$-dimensional subspaces of a $D$-dimensional space, Haar priors on Grassmannians lead to a rate of order $\sqrt{k(D-k)\log n/n}$, and model selection extends this result to infinite-dimensional settings. For general DPs, we introduce priors based on the spectral decomposition of the kernel that combine uniform distributions on the eigenvalues with Haar measures on Stiefel manifolds. When the underlying space is finite with $N$ elements, we obtain a rate of order $N\sqrt{\log n/n}$, without any assumption on the kernel, and derive that the $2^{N}$ principal minors of the kernel can be estimated simultaneously. Finally, for translation-invariant DPs on the torus $\mathbb{T}^d$, we obtain rates that only depend on the smoothness of the kernel at the origin and that are adaptive with respect to it. Combined with a new lower bound on the Hellinger distance between such DPs, these rates also yield estimators of the kernel in $\mathbb{L}_2$.
Read the full abstract