Computer Vision
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Saturday 7 November
The task of visual reasoning relative pose identification (VRRPI) has exposed critical weaknesses in the multi-view 3D spatial reasoning capabilities of modern vision-language models (VLMs). We present a systematic analysis of VLMs' behavior in VRRPI, uncovering stable output biases where VLMs disproportionately predict specific motion patterns regardless of visual evidence. To mitigate these inherent biases, we propose self knowledge re-expression with iterative debiasing (SKR-ID), an annotation-free pipeline that adapts VLMs for VRRPI using only randomly extracted unannotated image pairs. SKR-ID transitions VLMs' output mechanism from generic next-token prediction to task-optimized classification heads. The pipeline proceeds in two stages: (1) a cold-start initialization using debiasing object-centric prompts to anchor motion inference in 2D visual evidence; and (2) a recursive logit-ranking mechanism that enforces statistical parity by using the median logit value as a dynamic threshold. Experimental evaluation across frontier VLMs on standard benchmarks demonstrates that SKR-ID significantly mitigates systematic biases. Our method improves open-source VLMs' Macro-F1 scores from 35.0-45.4% to 53.7-58.0%, without using any annotated data. These results show that SKR-ID can effectively improve 3D reasoning performance from a severely collapsed baseline. The code is publicly available at https://github.com/astudentuser/Codes_of_SKR-ID.
Signal Processing
Soft-then-Hard: Joint Link Association and Precoding Optimization for LEO Satellite Networks
Spatial multiplexing in dense low Earth orbit (LEO) satellite downlinks is hindered by two bottlenecks. First, line-of-sight (LoS)-dominant propagation from a high orbital altitude places the users in a narrow angular region, which limits the spatial resolution achievable by a single satellite. Second, the visible satellites cause severe inter-satellite interference. Both effects call for joint link association and precoding, because the association decides which users are served together and which streams interfere. However, conventional block coordinate descent with hard association updates suffers from premature turning-off, where an inactive link loses precoder power and cannot be reactivated in subsequent iterations. To overcome this limitation, we propose a soft-then-hard update scheme that relaxes the association into continuous weights via iterative scaling, keeping candidate links active. Annealing progressively concentrates the association weights as the precoders are updated, after which a single hard matching determines the serving links. We further extend this framework to the LEO space-time channel, where distinct Doppler shifts provide user separability beyond angular geometry. Numerical results show that, over the space-time channel, the proposed design outperforms the baselines while effectively exploiting this separability.
Read the full abstract
Quantum Physics
A trace costs one quantifier: the exact complexity of perfect play in nonlocal games
Given a finite nonlocal game, can the players win with certainty? We prove that the existence of a perfect normal spatial strategy is $\Sigma^0_2$-complete, even when restricted to games of value one, so it cannot be decided even with an oracle for the halting problem. This is strictly harder than perfect play in the finite-dimensional and commuting-operator models, which are $\Sigma^0_1$-complete and $\Pi^0_1$-complete. Within the commuting model, the existence of a perfect normal strategy with a finite or semifinite von Neumann host is $\Sigma^0_2$-complete, whereas for every fixed type $\mathrm{III}_\lambda$ subtype it has the same $\Pi^0_1$-complete language as unrestricted commuting play. A trace thus costs one existential number quantifier: the hardness constructions force every perfect strategy to choose a finite clock period or a finite tree depth. Spatial $\Sigma^0_2$-completeness also holds when finite entanglement entropy is required, and both forms are formally verified in Lean. For every computable $0<\alpha<1$, perfect spatial play with $\mathrm{Tr}\,h^\alpha<\infty$ for the reduced density $h$ is $\Sigma^0_2$-complete as well; these conditions form a strict ladder between finite rank and finite entropy. We also prove $\Sigma^0_2$-completeness for one-shot zero-error communication with normal entanglement, and construct a finite channel with zero-error codes of a given size each of which needs infinite entanglement entropy.
Read the full abstract
Materials Science
Dopant Cluster Engineering for Tunable Room-Temperature Ferromagnetism in 2D Transition Metal Dichalcogenides
Two-dimensional transition metal dichalcogenides (2D-TMDs) offer a promising platform for realizing dilute magnetic semiconductors. However, simultaneously achieving large magnetic anisotropy energy (MAE) and high Curie temperature (T$_\text{c}$) remains a major challenge. In this work, we employ DFT+U to systematically investigate the thermodynamic stability and magnetic properties of 2D-MoSe$_2$ and 2D-WSe$_2$ doped with transition metal atoms. We found that the paired dopant configurations are more stable than previously attributed isolated dopants. Despite their nearly identical crystal structures, 2D-MoSe$_2$ and 2D-WSe$_2$ show contrasting properties upon doping. All paired dopants in 2D-MoSe$_2$ exhibit long-range ferromagnetic coupling (FMC) primarily due to carrier-mediated exchange interactions, whereas those in 2D-WSe$_2$ show weak exchange coupling. Interestingly, 2Mn$_\text{2W}$ shows a transition from less stable antiferromagnetic coupling (AFMC) in the short range to highly stable FMC in the long-range. The combined effect of spin-orbit coupling, exchange interactions, and orbital hybridization gives rise to a large MAE with maximum in-plane values of $-$3.5 meV for 2V$_\text{2Mo}$ and $-$3.2 meV for 2Mn$_\text{2W}$. Monte Carlo simulations estimate T$_\text{c}$ $\simeq$ 440 K for 2Fe$_\text{2Mo}$ and ensure its high-temperature stability in 2D-MoSe$_2$. Our results establish dopant clustering as an effective route to engineer high-temperature ferromagnetism in 2D-TMDs and highlight their potential for nano-spintronic applications.
Read the full abstract
High Energy Physics (Theory)
Non-Relativistic Supersymmetric Gauge Theories and Soliton Solutions
This paper investigates non-relativistic supersymmetric gauge theories obtained through null reduction, with particular focus on supersymmetric Abelian Higgs models and their soliton solutions. Starting from four-dimensional $\mathcal{N} = 1$ relativistic supersymmetric theories, we apply null reduction to derive Galilean-invariant non-relativistic theories in $2 + 1$ dimensions while preserving gauge symmetry and supersymmetry. The resulting symmetry structure is shown to correspond to the Bargmann extension of the super-Galilean algebra. We first analyze the massive supersymmetric Abelian Higgs model and show that, although it possesses nontrivial vacuum configurations, the mass term obstructs the existence of nontrivial BPS vortex solutions. We then study the massless supersymmetric Abelian Higgs model with multiple charged scalar fields, for which the BPS equations admit vortex-type soliton solutions under suitable charge and winding number conditions. The vacuum moduli space is identified as a weighted projective space, leading naturally to both integer and fractional vortex configurations. By imposing radial symmetry, we derive the corresponding vortex equations and examine their solutions analytically and numerically. Finally, we relate the vacuum structure of the model to the Ginzburg-Landau theory of superconductivity, highlighting how familiar superconducting behavior can emerge from a symmetry-preserving non-relativistic supersymmetric framework.
Read the full abstract
Number Theory
Motivic generating functions and $L$-derivatives of cohomological automorphic representations
Forty years ago, Gross and Zagier proved a remarkable formula computing the central derivative of the Rankin–Selberg $L$-function of a cusp form $f$ of weight $2$ and a character of the class group of an imaginary quadratic field, satisfying the so-called Heegner condition, and deduce that the nonvanishing of the central derivative implies the nonvanishing of the corresponding Mordell–Weil rank. This formula was later generalized in various contexts and degrees, for example, by S. Zhang for $f$ of higher even weights; and by Yuan–Zhang–Zhang for $f$ a Hilbert cusp form of parallel weight $2$ over an arbitrary totally real field $F$ and $\chi$ a Dirichlet character of an arbitrary CM extension $E/F$, allowing arbitrary ramification. In this article, we prove a Gross–Zagier type formula in the context of doubling method, for $f$ a Hilbert cusp form over $F$ and $\chi$ a Hecke character of $E$ of general weights under a balancing condition, again allowing arbitrary ramification. Indeed, we even formulate a general conjecture which we call the motivic arithmetic inner product formula, for cohomological automorphic representations of $\mathrm{GL}_{2r}$ over $E$ of hermitian type of balanced weights.
Read the full abstract
Artificial Intelligence
PatternBloom: Empowering Agentic RAG with Externalized RL-Distilled Graph Patterns
Reinforcement-learning-trained agents for retrieval-augmented generation retrieve content (passages, knowledge-graph fragments, or hyperedges) and supervise training with outcome rewards on the final answer. Across these paradigms the reasoning route through retrieved content never leaves the language model's hidden state, conflating declarative knowledge with the procedural knowledge that recurs across queries of the same logical type; content-only methods consequently plateau as reasoning depth grows. We introduce PatternBloom, an agentic RAG framework that shifts the unit of retrieval from content (what to read) to logic (how to reason). A first RL stage scores the agent's constructed evidence graph by a frozen oracle's information gain through the Information-Density Reward (IDR), a size-normalized mutual-information surrogate. High-reward trajectories are distilled into a Graph Pattern Memory (GPM) of type-abstracted reasoning skeletons, and a second RL stage uses the Pattern-Augmented Reward (PAR) to couple policy to memory, turning procedural reasoning into a learnable, externalized structural prior. A 7B PatternBloom outperforms $27$ baselines by $+15.0$ Avg-OOD EM, with the gain compounding with reasoning depth precisely because procedural knowledge, once distilled, is reused rather than re-derived at every query. We release our code at https://github.com/EllenYiGe/PatternBloom.
Read the full abstract
Accelerator Physics
Optimizing an RFSoC LLRF Controller for a Low-PRF, Short Pulse System
The Advanced Light Source (ALS) LINAC operates with 3 GHz RF pulses at 0.7 Hz pulse repetition frequency (PRF) and 1.2 microsecond duration, imposing severe limitations on traditional PID and adaptive feedforward control schemes. An RFSoC-based controller prototype was developed and presented in prior work to upgrade the legacy LLRF systems for the LINAC and Subharmonic Buncher. This paper presents sampling scheme optimizations and interleaving calibration techniques for RFSoC based control of pulsed RF systems. We detail bench characterization methodologies to isolate component-level phase noise across individual ADC, DAC, and clock components in the Xilinx ZCU208 hardware ecosystem. Finally, we explore translating open-loop controller bench measurements into controller-added closed-loop LLRF performance predictions.
Read the full abstract
Computer Vision
WordEval: Evaluating Word-Native Operation Fidelity in Document Editing
Microsoft Word is a central environment for office automation, and the faithful execution of its editing operations is an important capability for document agents. Unlike plain-text editing, editing in Word modifies a structured object model, where the requested effect is realized in object properties. Existing benchmarks, however, evaluate office workflows or desktop automation, treating word-processor documents as one task domain among many and verifying outcomes with keyword matching, per-task property checks, or screenshot comparison. We introduce WordEval, an operation-fidelity benchmark that instead centers evaluation on Word editing operations, testing whether each requested edit is realized as intended changes to Word objects and properties. It contains 642 cases across 15 Word feature areas, ranging from text and lists to charts, shape geometry, and SmartArt. 546 atomic cases cover 231 commands, 96 compositional cases combine multiple edits in a single document, and a Word-native oracle checks task-specific object properties and preservation conditions. Cases are constructed from editing-unit specifications, yielding controlled input documents, reference documents, and aligned oracle checks. Each compositional case is linked to corresponding atomic cases, supporting fine-grained diagnosis of editing failures. We evaluate eight agent systems and find that many struggle on the Atomic track, with performance varying across feature areas. Moreover, every system scores lower on Compositional than on Atomic, with gaps that vary substantially by system. Exact atomic-intent links and recorded failure diagnoses provide complementary views of this variation.
Read the full abstract
Social and Information Networks
Do links to retracted papers predict future retractions? An out-of-sample test on the open biomedical citation graph
Tools that screen manuscripts for research-integrity problems flag papers that cite retracted work, and retrospective studies report that such papers are retracted more often. This study tests the signal prospectively on open data. Predictors were computed on the January 2024 OpenCitations graph using the retractions known at that date, frozen, and compared with retractions in the Retraction Watch database from February 2024 to October 2026. The design was pre-specified. Among 10.5 million biomedical papers published 2015-2023, 2.1% cited a paper already retracted, and 4,323 were retracted later (0.041%). Citing a retracted paper raised the later retraction rate from 0.035% to 0.27% with one such reference; excluding reviews (post hoc), it rose to 2.3% with five or more. Within journal and publication year the odds ratio was 2.99 (95% CI 2.73-3.27), and 3.03 when reference-list length was also held fixed (post hoc), so the signal is not explained by retraction waves in journals or by long reference lists. As a screen it is weak: 0.9% of the 1,000 highest-ranked papers were retracted later (2.2% with reviews set aside, post hoc), and 84% of later retractions cite no retracted paper. When the time point was moved into the late-2023 retraction wave, the journal's own retraction rate became the far better screen. NIH-funded papers cited already-retracted work as often as other papers in the same journal and year. Links to retracted papers are a useful flag but not a screen on their own.
Read the full abstract
Superconductivity
Pseudogap in hole doped cuprates: a status report
The pseudogap state, an anomalous momentum dependent partial suppression of low energy electronic excitations persisting well above the superconducting critical temperature in underdoped cuprates, remains one of the most profound enigmas in condensed physics today. This review summarizes over three decades of experimental and theoretical progress. It appears that pseudogap is not a mere precursor to superconductivity, but a distinct, intertwined state of matter characterized by a complex interplay of competing and cooperating electronic orders. We critically examine the bulk probes like resistivity and magnetic susceptibility and the surface hallmark spectroscopic signatures observed via angle resolved photoemission spectroscopy (ARPES) in this review. ARPES reveals a Fermi surface truncated into disconnected segments and scanning tunneling microscopy (STM) visualizes a spatially heterogeneous energy gap around the Fermi level. These observations are contextualized alongside transport anomalies, including resistivity and bulk magnetic susceptibility, to construct a phenomenological portrait of a phase that violates conventional Fermi liquid theory. By mapping the electronic phase diagram, we highlight the abrupt transition at above optimal doping. We conclude with open questions still to be answered. This review underscores the role of the pseudogap as a key to unlocking the unconventional pairing mechanism of hole doped cuprates and the broader physics of strongly correlated electron systems.
Read the full abstract
Machine Learning (Statistics)
Group-Conditional Conformal Risk Control: Exact Guarantees, Limits, and Overlapping Extensions
We study group-conditional conformal risk control (CRC) for disjoint and overlapping groups under bounded monotone losses. For disjoint groups, we establish exact finite-sample conditional risk control, two-sided bounds on the group-restricted risk, and fundamental limitations arising from local support and calibration-rank granularity. For overlapping groups, we study two exactly valid approaches: group-restricted calibration followed by union deployment, and direct calibration of the deployed prediction-set family. We characterize the union's limiting reconciliation gap, identify conditions for its disappearance, and establish sharp lower bounds for structured losses in specified product-form models. We also demonstrate a strict size advantage for direct calibration in an ordinal model. Experiments show that direct calibration can reduce prediction-set size, but its efficiency depends on group structure, loss type, and shape fitting. Results on historical diabetic-retinopathy scores further illustrate these tradeoffs. Overall, local support, reconciliation slack, and threshold heterogeneity determine efficiency, with neither approach uniformly dominating.
Read the full abstract
Social and Information Networks
Can open citation data detect citation stacking? An audit against Clarivate's journal suppressions
Clarivate suppresses journals from the Journal Citation Reports when self-citations or concentrated citations from another journal ("citation stacking") distort their Journal Impact Factor. Its data are closed. Yearly journal citation networks were built from the OpenCitations snapshot of January 2024 (1.96 billion citations; 85,437 journals) and compared with Clarivate's suppression lists for data years 2015-2023, collected from primary sources. Where reference lists are present, open data reproduce Clarivate's published self-citation shares within ten points for 36 of 50 journals. The failures follow publishers that do not deposit references with Crossref, which also hides several stacking donors. Biodiversity data downloads registered with DataCite inflate naive impact-factor numerators of taxonomy journals. The leading open method, CIDRE, detects none of 15 named Clarivate pairs: it drops large donor journals, and lowering its donor threshold merges the pairs into large groups. A recipient-centric reading of CIDRE's own test flags stacking-listed journals 6 to 14 times more often than other journals, depending on consensus level, and 5 to 7 times more often than size-matched controls, sometimes years before the listing. At equal flag rate it finds more listed journals than a plain 43% donor-share rule (10 vs 3 of 28), and about as many as a rule with a citation floor. A prospective check on the 2024-2025 lists agrees in direction but is not significant (one of six). Flagged pairs show more citations per citing paper and weaker topical similarity than matched pairs. Open audits are feasible if coverage gaps are reported.
Read the full abstract
Optics
Wideband AR-HCF Transmission Under Variable Network Traffic: A System-Level Analysis
Dynamic traffic changes spectral occupancy in wideband optical links, perturbing surviving channels in single-mode fiber (SMF) through nonlinear interference (NLI), inter-channel stimulated Raman scattering, and amplifier loading. This motivates amplified-spontaneous-emission (ASE) idlers to maintain full occupancy. We test whether spectral filling remains necessary in antiresonant hollow-core fiber (AR-HCF) using a span-by-span model of a 96-channel, 9.6-THz extended-C+L link over 960 km. The model jointly tracks signal, ASE, and NLI, including amplifier pigtails, line-resolved CO2 absorption, intermodal interference, amplifier control, and load transients. At equal launch power, reducing occupancy from 100% to 20% changes surviving-channel generalized signal-to-noise ratio (GSNR) by just 0.0016 dB in AR-HCF versus 1.52 dB in SMF. At the AR-HCF throughput-optimal launch of +12.3 dBm/channel, the GSNR shift rises to 0.24 dB, nearly all from NLI in short solid-glass amplifier pigtails, chiefly the output pigtail. These pigtails hold the optimum 8.9 dB below the amplifier ceiling, but removing them yields only about 1% more throughput at that launch due to transceiver, intermodal-interference, and gas-absorption limits. Line-resolved gas modeling distinguishes several decibels of CO2 attenuation, largely offset by per-span equalization, from narrow absorption notches not recoverable by flat gain. Constant-gain control permits a 23% amplifier energy saving by tracking load, whereas constant-output control saves none. An 85% channel drop causes an 8.8-dB uncontrolled gain excursion, reduced to 0.6 dB with an assumed 500-Hz loop. Thus, AR-HCF shifts the constraints on elastic wideband transmission from fiber nonlinearities to amplifier-site components, gas absorption, and control dynamics.
Read the full abstract
Computation and Language
Hibana: Fast Agentic Inference on Unified-Memory Devices
Local AI promises capable personal assistants that run on users' own devices, keeping data and computation under their control. Open-weight models that fit on local devices can now handle practical tasks such as coding. Yet current inference engines fall short. Popular engines such as vLLM and SGLang target datacenter GPUs with dedicated host memory, while local engines underuse the memory hierarchy of unified-memory devices. We present Hibana, an inference engine for unified-memory devices that co-optimizes model representation, memory management, and decoding strategy. Model surgery, the process of editing a trained model's architecture to suit its hardware, changes how the model computes, shrinking its memory footprint and data movement. Unified-memory management preserves each agent's state across turns, so long-running agents can resume instantly instead of recomputing their state. Adaptive speculation generates several tokens per step. Hibana codesigns these features around the shared memory capacity and limited bandwidth of unified-memory systems-on-chip to support local multi-agent workloads. On a single DGX Spark serving Qwen and Nemotron models of up to 125B parameters, Hibana completes up to 3.4$\times$ more agentic tasks per hour and delivers up to 3.9$\times$ higher decoding throughput than other established frameworks such as vLLM, NVIDIA Flare, llama.cpp, and TensorFold. Code is available at https://github.com/sharar-project/hibana.
Read the full abstract
Machine Learning
Dynamic State Estimation from Sparse PMUs via Physics-Informed Graph Neural Operators
This paper proposes a physics-informed graph neural operator for power system dynamic state estimation. From a sparse, intermittent measurement window at a few generator buses, the operator estimates the rotor-angle and speed trajectories of all machines, unmeasured ones included, in one forward pass. A graph-attention encoder processes the measurements, and the swing equations are imposed as a hard constraint or as a soft one with a learned correction that absorbs model errors. It is trained on simulated measurements, without state labels. With several phasor measurement units, masked training hides one from the input but keeps it in the loss. On four transmission systems, simulations show that the operator is the most accurate estimator that neither optimizes per scenario nor uses state labels, except on the smallest system at large model errors, where tuned Kalman smoothers are better. Networks trained on simulated states are more accurate on the two smallest systems but not on the 118-bus system, where the operator is faster and more accurate than hard-constrained moving-horizon estimation. Under model errors, it matches or outperforms an optimization-based benchmark at a fraction of its cost. The results hold for a detailed machine model with exact parameters seen only through voltage phasors.
Read the full abstract
Human-Computer Interaction
Who Benefits, and When? Trajectories of Academic Achievement and Interest in AI-Supported Learning
AI-supported learning environments are increasingly used, yet relatively little is known about how students' achievement and interest develop over time or whether these trajectories vary across learners. This study examined longitudinal trajectories of mathematics achievement and interest in a semester-long AI-supported mathematics learning. Using a mixed-methods longitudinal design, repeated measures of mathematics achievement were collected at four time points and mathematics interest at three time points. Linear mixed models (LMMs) were used to examine overall changes over time, while linear quantile mixed models (LQMMs) were used to examine differences in trajectories across the lower, median, and upper portions of the outcome distributions. Mathematics achievement increased substantially during the intervention, with the largest gains emerging between the second and third measurement occasions. However, the trajectory differed across the achievement distribution. Mathematics interest increased significantly after the first month, but this increase was not sustained through the end of the semester. Qualitative findings further indicated that students experienced the AI platform differently. The study highlights the importance of examining both cognitive and motivational trajectories.
Read the full abstract
Machine Learning
The Fast Tree Transform
The Fast Tree Transform is a locally-aggregated transform defined for arbitrary graph geometries that generalizes both the Hadamard frequency transform and the Haar wavelet transform on the circle. The pure frequency part of the transform approximately diagonalizes the graph Laplacian; this fact makes it useful as a preconditioner in geometric optimization problems. The full transform provides an only slightly-redundant representation of signals on the graph that allows sparse representation of signals that include both edges and textures.
Read the full abstract
High Energy Physics (Experiment)
NuSTEC Review: Current Experimental and Theoretical Challenges of Neutrino-Nucleus Scattering
The precise measurement of neutrino properties is among the highest priorities in fundamental particle physics, involving many experimental and theoretical efforts worldwide. Since the experiments rely on the interactions of neutrinos in the nuclear environment, the planned advances in the scope and precision of these experiments require a commensurate effort in understanding and modeling the hadronic and nuclear physics of these interactions, as incorporated in neutrino event generators. This modeling is essential to every phase of the experimental analyses, and the theoretical uncertainties play an important role in interpreting experimental results. In this Review of current challenges, we summarize the open questions in neutrino-nucleus interactions at GeV-scale energies, and discuss in detail the potential impacts of these unknowns on the measurement of neutrino properties, using the determination of oscillation parameters as a central example. After an Executive Summary and a concise Overview of the issues, we address how the neutrino event generators treat these challenges, how electron-nucleus interactions can help understanding these challenges and the important unanswered questions for each underlying physics process—from quasi-elastic to deep-inelastic scattering. Across every topic, we find that current challenges can be met only through active support and collaboration among specialists in strong interactions and electroweak physics, including theorists and experimentalists from both the nuclear and high-energy physics communities.
Read the full abstract
Machine Learning (Statistics)
Beyond Discrete Ancestry Groups: Distribution-Aware In-Context Learning for Cancer Outcome Prediction Across the Genetic Ancestry Continuum
Machine learning models for omics-based prediction of cancer outcomes are commonly built on aggregated data without distinguishing ancestry groups, or under an ancestry-discrete design, which partitions samples into ancestry groups and fits a separate model within each group. Recent evidence shows that this approach is inherently limited: predictive accuracy declines with genetic distance from the training samples, even within a single ancestry group, because the data distribution shifts continuously rather than only between predefined categories. In this paper, we propose distribution-aware in-context learning (DA-ICL) for ancestry-continuum modeling. Our method extends conventional ancestry-discrete modeling to ancestry-continuum modeling, capturing the nonstationary omics-endpoint relationship across the genetic continuum while enabling efficient modeling through an in-context learning foundation model. We provide theoretical analysis demonstrating the superiority of the ancestry-continuum over ancestry-discrete modeling, along with empirical results supporting the proposed framework. Source code is available at https://github.com/ai4pm/DA-ICL-Continuum.
Read the full abstract
Computation and Language
LongCapHarness: Capability-Guided Auto-Harnessing for Long-Context Reasoning
Existing methods enhancing an LLM's long-context reasoning capability often rely on data synthesis and follow-up LLM post-training. However, the data synthesis strategy and the post-training recipes are empirically driven and computationally intensive. In this work, we propose a new framework, dubbed LongCapHarness, to instead harness the underlying capabilities for long-context reasoning. The harnessing process recursively revises and updates capability primitive implementation, capability state and control logic through recursive self-improvement on the development set. With neither data synthesis nor LLM post-training, LongCapHarness is applied to Llama3-8B and transferred to Qwen2.5-3B. LongCapHarness substantially outperforms recent data-synthesis-based baselines on HELMET and RULER benchmarks; its estimated development cost is less than 50% of the estimated training costs of NExtLong and PolicyLong. We hope these results motivate further research into harnessing the latent capabilities of LLMs to solve tasks beyond their explicit training.
Read the full abstract
Signal Processing
Coexistence of LEO Satellite Communication Systems under Regulatory Protection Constraints
This paper investigates the coexistence of two low Earth orbit (LEO) satellite communication systems sharing the same frequency band: a primary system with regulatory precedence over the spectrum and a secondary system that must not cause excessive interference to the primary system's ground users. This spectrum sharing paradigm was recently formalized by the United States Federal Communications Commission (FCC) through short-term and long-term protection criteria that limit primary-link unavailability and throughput degradation due to interference inflicted by the secondary system. To facilitate coexistence, we propose a satellite selection mechanism that associates the secondary system's satellites with clusters of ground cells at each handover to maximize secondary-user throughput while abiding by these regulatory constraints. Our mechanism enforces both constraints through a cumulative assessment that combines past primary-link performance with predicted performance over the upcoming handover interval. Simulation results show that our proposed method achieves up to 31% higher secondary-user throughput than prior work that imposes interference-to-noise ratio (INR) constraints, even when its constraints are tuned to satisfy the FCC's throughput-based protection criteria.
Read the full abstract
Optimization and Control
Zeroth-Order Langevin Dynamics for Stochastic Nonconvex Optimization: Curvature Regularization and Non-Asymptotic Guarantees
Zeroth-order methods provide a way to optimize stochastic nonconvex objectives using only noisy function evaluations. However, the smoothing inherent in their gradient estimators effectively modifies the objective being optimized. In particular, Gaussian smoothing with smoothing parameter $\eta>0$ induces, to leading order, an implicit Hessian-trace regularization, which motivates the study of the curvature-regularized objective of the form $f(x)+(\eta^2/2) tr(\nabla^2 f(x))$. We develop a non-asymptotic theory in the nonconvex setting for a class of algorithms called stochastic zeroth-order Langevin dynamics (ZOLD), which use only noisy function evaluations and do not have access to first- or higher-order derivative information, and derive excess-risk guarantees for the curvature-regularized objective. We first control the finite-time distribution of the ZOLD iterates in Wasserstein distance with respect to the Gibbs measure associated with the Gaussian-smoothed objective. An explicit coupling between the smoothing and inverse temperature parameters then quantifies the proximity of this measure to the Gibbs measure associated with the corresponding Hessian-trace-regularized objective. These distributional estimates in Wasserstein distance lead to excess-risk bounds for the curvature-regularized objective. Our analysis accommodates assumptions under which the stochastic function evaluations are allowed to have non-uniform dependence on the data stream. In addition, the constants in the non-asymptotic bounds do not rely on log-Sobolev or Poincaré inequalities as in prior zeroth-order Langevin analyses. To the best of our knowledge, this is the first non-asymptotic analysis of a zeroth-order Langevin method that simultaneously establishes excess-risk guarantees for a curvature-regularized objective and Wasserstein guarantees for the associated Gibbs measure in the nonconvex setting.
Read the full abstract
Medical Physics
Noise Propagation and Non-Uniqueness in Three-Material Dual Energy CT with a Path Length Constraint
Objective: Using a path length constraint with dual energy computed tomography (DECT) data, a sample can be decomposed into three basis materials. We aim to analyze the stability of such a system by presenting a noise propagation framework for optimizing the contrast to noise ratio (CNR) of virtual monochromatic images (VMIs) and by studying instances where sinogram decompositions are not unique. Approach: We extend a noise propagation framework developed for two-material DECT to three materials. Using the Jacobian of the system, we identify tube voltage pairings with good noise properties and obtain analytical expressions for VMI CNR. We apply spectral theorem to identify the best CNR possible in a linear combination of basis images. By using a reduced two-dimensional problem, we develop the sampling isoline recovery algorithm (SIRA) to study non-unique sinogram decompositions in this system. Main results: Using our noise propagation framework with a realistic breast phantom consisting of adipose, glandular tissue, and iodine, we find an ideal tube pairing of 34 kVp and 69 kVp, and a maximum VMI CNR of a spiculated lesion of 9.9. By allowing arbitrary linear combinations of basis material images, we find a CNR of 10.2 can be achieved. In a realistic phantom with a cardiac plaque, we find non-unique water, bone and iodine decompositions, and show that qualitatively accurate reconstructions can be obtained using SIRA. Significance: We developed a framework for direct optimization of VMI CNR in a three-material DECT decomposition and showed that an arbitrary linear combination of basis material images can achieve a higher CNR than a strict VMI. We demonstrate the presence of non-unique sinogram decompositions in three-material DECT with a data constraint.
Read the full abstract
Statistical Methodology
Riemannian Ordinary Least Squares
Many scientific responses take values in Riemannian manifolds, with predictors that are scalar- and/or manifold-valued. A regression analysis in such settings must provide accurate prediction with hypothesis testing for individual coefficients. Yet no current manifold-regression framework accommodates both predictor types and enables per-coefficient hypothesis tests. We propose Riemannian Ordinary Least Squares (ROLS) as a tangent space conditional-mean model, which is the ordinary least squares (OLS) analogue for Riemannian manifolds. ROLS is well-defined under an explicit injectivity condition that keeps the logarithm maps single-valued, with cut-locus conditions for branch-sensitive cases outside it. Within the local geometric setting it admits a closed-form estimator, an $O(n^{-1})$ excess prediction-error bound with explicit curvature bias, and asymptotic $t$- and Wald tests for individual coefficients. With synthetic simulations and three real-data case studies, ROLS is shown to be competitive with the state-of-the-art baselines. In addition, the per-coefficient tests agree with effects reported in literature. We envision ROLS to be a standard tool for manifold-valued regression with coefficient-level inference to support scientific discovery, which can be interpreted in the same way as Euclidean OLS. The source code is available at https://github.com/XiaoyuChenPub/ROLS.
Read the full abstract
Astrophysics of Galaxies
B-fields In STar-forming Region Observations (BISTRO): Population-Level Universality and Environmental Structuring of 450/850 $\mu$m Polarization Signatures
We observed 69 molecular cloud regions at 450 and 850 $\mu$m with the POL-2 polarimeter on the James Clerk Maxwell Telescope and matched the maps to a common $12.6''$ beam. Polarization fractions are typically lower at 450 $\mu$m. The median of 30 measured ratios is $P_{450}/P_{850} = 0.69 \pm 0.02_{\rm sys} \pm 0.03_{\rm stat}$. Across 35 regions with reliable spectral indices of total ($\alpha$) and polarized ($\gamma$) intensity, the medians over sources are $\langle\alpha\rangle = 3.33$ and $\langle\gamma\rangle = 2.85$-$3.04$. The median $\gamma - \alpha$ is $-0.57$ to $-0.45$, with $\gamma < \alpha$ in 24-27 regions. These ranges span the two extreme placements of $3\sigma$ limits for polarized intensity below the noise floor. The position angles at the two wavelengths agree within $10^\circ$ in 10 of 13 regional medians, disfavoring a geometric origin that would rotate the angles. The medians of $\gamma-\alpha$ in 9 low-mass, 19 high-mass and 7 Central Molecular Zone (CMZ) regions are not ordered low-mass $<$ high-mass $<$ CMZ (Jonckheere-Terpstra $p = 0.43$-$0.70$). A difference among classes is not robustly detected. The Kruskal-Wallis $p = 0.056$-$0.078$ becomes $p = 0.026$-$0.041$ when the Rosette Molecular Cloud alone is excluded, although the high-mass class has the lowest median in every treatment. The classes follow a distance sequence (the beam spans 0.008-0.51 pc), coupling environment and physical scale. Within clouds, the 450 $\mu$m polarization fraction decreases with Herschel column density in all 7 fields with sufficient dynamic range. Coefficients from pooled pixels mix trends within clouds, differences between clouds, and the influence of OMC 1. They therefore do not define a single law for individual pixels. The wavelength dependence across sources provides a benchmark for grain alignment. We cannot order it by environment and do not propose a unique mechanism.
Read the full abstract
Artificial Intelligence
Represented but Not Used: Value-Policy Dissociation in LLM Agents
Large language model (LLM) agents need to choose actions whose consequences unfold over multiple future steps, yet poor decisions do not reveal whether the model failed to represent those consequences or failed to use them. Mechanistic interpretability has studied internal representations of environment state and plans, but sequential decision making also requires evaluating how candidate actions change future outcomes. Reinforcement learning formalizes this as action value. We study this computation in Qwen2.5-7B-Instruct, Qwen2.5-32B-Instruct, and GPT-OSS-20B across a partially observed Grid environment and WebShop, a simulated web shopping environment. We find that the relative values of candidate actions are linearly decodable from intermediate hidden states, including at states where the policy prefers a worse action. We call this gap between value representation and policy choice value-policy dissociation. To investigate how represented value relates to action selection, we compare direct interventions along the value readout direction with Natural Value Transport (NVT), a transformation learned from hidden state changes that naturally result from differences in action value. While readout-direction interventions have limited and inconsistent effects, NVT more consistently influences action preferences, particularly at later layers. We also test a closed-loop controller that uses the value readout as a sensor and NVT as an actuator, observing improvements in task outcomes. These results suggest that LLM agents can represent useful prospective value without reliably translating it into policy choice.
Read the full abstract
Astrophysics of Galaxies
HINSPHERES: Spherically Symmetric Radiative Transfer Modeling of HINSA and the Radial HI-to-H$_2$ Transition in Molecular Clouds
The transition from atomic hydrogen (HI) to molecular hydrogen (H$_2$) is the first step in the evolution from diffuse interstellar gas to dense star-forming cores. HI narrow self-absorption (HINSA), produced by cold atomic gas mixed with molecular material, provides a direct tracer of the HI-to-H$_2$ transition in molecular clouds. Existing HINSA analyses typically model HINSA as a single Gaussian component with a uniform spin temperature. This approach cannot recover the radial HI distribution and yields large errors for young cores whose profiles deviate from a Gaussian shape. Here we present hinspheres, an open-source framework that models the core as onion-like spherically symmetric concentric shells, solves radiative transfer along the line of sight and jointly fits the HI abundance, temperature, and kinematic parameters. The framework recovers all parameters with a mean relative error of 4.6-6.8% on our synthetic data with different ages. A combined resolution-noise grid shows that the median recovery error remains below 16% for cores within 1 kpc observed at FAST resolution with noise levels below 0.2 K. Applying it to the dark cores L1574 and L1544, we reveal radial HI abundance profiles that span about three orders of magnitude, which were missed by the single-Gaussian method. The derived chemical ages are $6.5\times10^6$ and $8.7\times10^6$ yr, respectively. Compared with the commonly used single-Gaussian method, hinspheres characterizes the radial profile of HINSA structures more accurately, making it well suited to identifying and analyzing young cores undergoing the HI-to-H$_2$ transition.
Read the full abstract
Machine Learning
Bayesian Low-Rank Posteriors for Scalable Membership Inference
Membership inference attacks (MIAs) aim to determine whether a sample was used during the training of a target model. Strong MIAs rely on reference (or shadow) models to estimate target-conditional score distributions, but training such ensembles is computationally expensive for modern large-scale Vision–Language Models (VLMs). In this work, we propose a scalable alternative that replaces explicit reference-model training with a Bayesian approximation over low-rank adaptation parameters. Specifically, we model uncertainty only within the LoRA subspace and construct a posterior from a single training run, from which we sample a diverse set of virtual reference models. We instantiate this approach using Stochastic Weight Averaging Gaussian (SWAG), enabling efficient approximation of target-conditional score distributions at a fraction of the cost of conventional shadow-model ensembles. We evaluate the resulting attack on three downstream-adapted VLMs across two privacy-sensitive visual question answering tasks, namely Document VQA and Medical VQA. To isolate the effect of pretraining knowledge, we additionally introduce a controlled synthetic dataset. Our approach outperforms standard reference-based attacks while requiring only a single reference-model fine-tuning run, making membership inference more practical for large VLMs.
Read the full abstract
Artificial Intelligence
Decentralised Drone Swarm Formation under Silent Failures with ACO Pheromone Claims and PSO Tuning
We study decentralised drone formation, without a central planner, under the conditions that make it hard: nobody knows how many drones are alive, failures are silent, radio is short-ranged and lossy, and actuation is noisy. In our controller, ant colony optimisation (ACO) allocates drones to picture positions through claims that behave as reinforced, evaporating pheromone and are gossiped between neighbours, so evaporation doubles as failure detection. A particle swarm optimisation (PSO) law flies the drones, a noise-buffered Voronoi filter keeps them apart, and PSO also tunes the controller offline. We prove that claims of destroyed drones vanish within bounded time, that conflicts between drones that exchange tables are resolved consistently, that survivors cover the picture to within a factor of two, and that drones cannot collide while their safety cells are feasible. On held-out seeds and with two tuned configurations, the controller formed the picture in 95-99% of runs in the hardest regime, restored it after a silent loss of half the swarm in 96-100% of runs at normal difficulty, transferred unchanged to eight unseen shapes and to 300 drones, and overlapped in 5 of 21,088 runs. A CBBA-style baseline that shares the evaporating claims but uses argmax without swaps was at least as reliable as ours once tuned with the same budget, and repaired faster, whereas ours generalised better to unseen shapes. The benefit thus comes from the pheromone dynamics and from tuning, not from ACO's stochastic construction or PSO's social terms. A centralised oracle is faster.