Peer Revenue Now accepting positions

Papers

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

All of arXiv

Monday 9 November

High Energy Physics (Theory)

Semiclassical Asymptotics of Entanglement in Torus-Link States

We investigate the entanglement entropy of topological quantum states associated with $T_{d,\,dn}$ torus links in (2+1)-dimensional SU(2) Chern–Simons theory at level $k$. These links consist of $d$ components with pairwise linking number $n$, and the corresponding link-complement states provide a natural setting to study multipartite entanglement in topological quantum field theories. We analyze the semiclassical limit $k\to\infty$ of the entanglement entropy by deriving the asymptotic behavior of the moments of the reduced density matrix. To achieve this, we decompose the integrable-representation space into left-edge, right-edge, and bulk regions and employ the stationary-phase method to evaluate their respective contributions. We find that the leading asymptotic behavior is governed entirely by the two edge regions, while the bulk contribution is asymptotically suppressed. The resulting expression for the limiting entanglement entropy is given in terms of Hurwitz zeta functions and exhibits a dependence on the residue class of $k$ modulo $n$. We further investigate the dependence of the limiting entanglement entropy on the winding parameter $n$ and the number of link components $d$. In the large-$d$ limit, the entropy reduces to a binary entropy determined by two normalized weights associated with the left and right edges. Finally, we show that the two iterated limits, $k\to\infty$ and $d\to\infty$, commute, reproducing the result obtained previously by taking the large-party limit first.
Read the full abstract

Machine Learning

Operation-aware Multimodal Clinical Decision Support System for ICU Discharge Risk Assessment

Unplanned 30-day intensive care unit (ICU) readmission is a critical benchmark for evaluating healthcare quality. Most predictive models assume readmission risk depends solely on a patient's individual pathophysiological state, ignoring hospital operational context and resource capacity. In practice, discharge decisions are dynamic trade-offs between patient stability and bed turnover pressures. To address this gap, we develop an Operation-aware Multimodal Clinical Decision Support System (CDSS) for ICU discharge risk assessment. The system combines multi-agent collaborative extraction with self-evolving retrieval-augmented generation (RAG) to eliminate template noise from progress notes, while constructing departmental workload indicators alongside admission flow features, continuous vital signs, and sparse medical codes (ICD/DRG). Addressing severe class imbalance (about 10.5% readmissions) and alarm fatigue, we implement data normalization, class reweighting, and decision-threshold calibration. In 20 independent runs across 20,000 ICU stays from MIMIC-III, empirical findings demonstrate that: vital signs alone have limited predictive capacity under real-world class imbalance (PR-AUC of 0.1387 +/- 0.0086); directly concatenating raw text embeddings degrades decision tree performance (XGBoost and LightGBM PR-AUC drop to 0.3893 and 0.4030); whereas an end-to-end multimodal fine-tuning architecture (Transformer E2E FineTune) utilizing layer-wise parameter freezing and differential learning rates reverses text negative transfer, achieving the highest AUROC of 0.8288 +/- 0.0110, PR-AUC of 0.4373 +/- 0.0310, and F1-Score of 0.4757 +/- 0.0196. Operational indicators supply independent risk signals, laying an engineering foundation for high-utility clinical decision support tools.
Read the full abstract

Statistics Theory

On the Instance-Specific Sharp Rates for Constrained Least Squares Estimators in the Sequence Model

We develop a general methodology based on the Feature Space Decomposition framework for establishing explicit, instance-specific sharp rates for the expected estimation error of convex-constrained least squares estimators in the sequence model, without requiring estimates of localized Gaussian mean widths. We apply this methodology to $\ell_q$ ellipsoids, $\ell_1$ balls, and $kB_1^N\cap B_\infty^N$, obtaining matching upper and lower bounds in the nondegenerate regime. Our methodology also extends beyond Gaussian noise to symmetric noise distributions satisfying a Poincaré inequality. We also provide an elementary proof of the instance-specific sharp rate for isotonic regression for any well-specified signal and any noise level. Combining this result with Han's asymptotic reduction, we further obtain explicit instance-specific sharp convergence rates in probability for isotonic regression in Gaussian linear regression under general signal-dependent asymptotics conditions.
Read the full abstract

Biomolecules

OriDynamics: Towards a World Model for Biomolecular Dynamics

Biomolecular recognition and regulation depend on atomic interactions that change as molecular partners fluctuate and adopt alternative conformations. Modeling these processes requires ensembles that capture both conformational diversity and coordinated atomic fluctuations between interacting molecules. Here we present OriDynamics, a unified framework for all-atom ensemble generation across protein monomers and complexes with small molecules or protein partners. OriDynamics introduces a latent ensemble decomposition that couples broad conformational exploration in a compact structural latent space with local all-atom sampling around generated proposals. Ensemble-level supervision constrains mean positions, fluctuation amplitudes and joint coordinate covariance to capture coordinated atomic variability within and between molecular partners. Across benchmarks of proteins and their complexes, OriDynamics demonstrates superior overall performance in conformational coverage, atomic fluctuation fidelity and interaction geometry compared with leading baselines. Application studies further demonstrate recovery of alternative functional conformations and partner-dependent changes in conformational ensembles alongside sampling of configurations associated with ligand dissociation. These capabilities provide a foundation for a biomolecular world model that represents accessible molecular states and their dependence on interaction context.
Read the full abstract

All authors Jiahao Yang, Dong Liu, Hongliang Li, Ji Chen, Yuehua Liu, Daoyun Ding, Duo An, Jun Li, Jiahua Rao, Ziqiao Zhang, Peicong Lin, Shuangjia Zheng

High Energy Physics (Experiment)

Measurements of $C\!P$ violation with charm hadrons at Belle and Belle II

Recent searches for $C\!P$ violation in charm-hadron decays at Belle and Belle II are presented. The measurements cover $D^+_{(s)}\to\eta h^+$, $D\to\pi\pi$, $D^0\to K_S^0K_S^0$, and $D^0\to\pi^+\pi^-\pi^0$, together with the singly Cabibbo-suppressed baryon decays $\Xi_c^+\to\Sigma^+h^+h^-$ and $\Lambda_c^+\to ph^+h^-$, where $h=\pi,K$. The Belle II analyses use the Run~1 data sample of $428\,\mathrm{fb}^{-1}$, supplemented by $980\,\mathrm{fb}^{-1}$ of Belle data for the $K_S^0K_S^0$ measurements. We report preliminary results for the $\eta h^+$ modes, complementary tests of the isospin structure of $D\to\pi\pi$, and an application of inclusive charm-flavor tagging. The first individual decay-rate asymmetries in these three-body charmed-baryon modes also provide tests of $U$-spin symmetry. All reported asymmetries are consistent with $C\!P$ symmetry within the present precision.
Read the full abstract

Signal Processing

Bistatic UAV-Based 3D SAR Imaging Using Integrated Sensing and Communications Transceivers

Integrated sensing and communication (ISAC) frameworks enable autonomous systems to jointly perform radar sensing and wireless communication using common spectral and hardware resources. Prior art on ISAC-enabled UAV for SAR focused on monostatic configurations for two-dimensional target imaging. In this work, we propose a bistatic UAV tomographic SAR framework in which the IEEE~802.11ad waveform, RF hardware, and spectral resources are jointly utilized for communication and radar sensing for three-dimensional (3D) localization of complex targets. Specifically, we demonstrate how communication packets transmitted from a master ISAC transceiver UAV to a passive receiver UAV are exploited for timing and frequency synchronization, motion compensation, and 3D localization of targets, without requiring auxiliary infrastructure on the passive receiver for these tasks. We evaluate the proposed framework for localizing automotive targets over elevated terrain, with Monte Carlo simulations modeling synchronization and platform motion errors. The results demonstrate improved localization accuracy compared with conventional InSAR-based localization. The interplay between radar and communication performances are further evaluated using diverse modulation schemes.
Read the full abstract

All authors Akanksha Sneh, Krishna Ayyagari, Vivaswan Nawani, Shobha Sundar Ram

Information Retrieval

CHO-Bench: How Do Context Hints Affect Ad Rankings in LLM Advertising?

LLM advertising selects advertisements based on user conversations, where user needs may extend beyond explicit product searches. ChatGPT Ads allows advertisers to provide natural-language context hints describing what they offer, who may find it useful, and when it may be useful, as additional information for ranking advertisements across diverse user conversations. However, how to effectively construct these hints remains unclear. We formulate Context Hint Optimization (CHO) as the problem of constructing context hints to improve an advertisement's ranking under a given budget. Drawing on established marketing frameworks, we develop seven context hint generation strategies covering What, Who, and When, along with a Unified strategy that combines all seven dimensions. We introduce CHO-Bench and conduct over 1.19 million paired evaluations across five datasets using embedding-based and LLM-based rankers. Our results show that context hints improve advertisement rankings under the evaluated offline rankers. While individual strategies achieve similar average gains and Unified offers no consistent advantage, strategy differences become more pronounced under specific generation-ranking configurations, with embedding-based and LLM-based rankers favoring different information dimensions. Increasing the hint budget further improves rankings with diminishing marginal gains. These findings establish an offline benchmark for studying ad-side ranking optimization in LLM advertising and highlight opportunities for model-specific hint selection and budget-aware generation and aggregation.
Read the full abstract

Soft Condensed Matter

Frank Elastic Anisotropy Induced by Anisotropic Attractive Interactions in Nematic Liquid Crystals

We present a mean-field theory to calculate the Frank elastic constants of a nematic liquid crystal with an anisotropic Maier–Saupe (MS) attractive interaction restricted to a cylindrical interaction volume of height $L_z$ and diameter $L_c$. The shape of the interaction volume is characterized by the aspect ratio $\alpha_z=L_z/L_c$, which controls the spatial anisotropy of the attractive interaction. We derive expressions for the splay $(K_{11})$, twist $(K_{22})$, and bend $(K_{33})$ elastic constants and calculate them numerically as functions of concentration and temperature. At the special aspect ratio $\alpha_z=\sqrt{3}/2$, the transverse and axial second moments are equal, yielding the one-constant relation $K_{11}=K_{22}=K_{33}$ over the full ranges of concentration and temperature considered. Away from this point, oblate and prolate interaction volumes produce distinct elastic anisotropies. Changes in concentration or temperature can also reverse the hierarchy of the three elastic constants.
Read the full abstract

Fluid Dynamics

Elastic frontogenesis sets the spectra of smooth elastic turbulence

Most turbulent flows are sustained by inertia, which transfers energy across scales and produces characteristic power-law spectra. Yet dilute polymer solutions also exhibit turbulence when inertia is negligible: in this state, known as elastic turbulence, the velocity field remains spatially smooth while both the velocity and the polymer conformation exhibit robust power-law spectra. How a smooth flow, lacking a hierarchy of eddies, generates such multiscale structure has remained unexplained. Here we show that the missing mechanism is elastic frontogenesis, whereby the flow spontaneously organizes polymer stress into thin sheets that are stretched within their plane and compressed across their thickness, causing their amplitude to grow as they become thinner. Because the fronts are sheet-like structures, this amplification-thinning law uniquely determines the polymer conformation spectrum, while the Stokes response transfers the same geometric scaling to the velocity field, reconciling broadband spectra with spatial smoothness. Being geometric rather than model-specific, elastic frontogenesis provides an organizing principle for viscoelastic flows and the viscoelastic analogue of vortex stretching in inertial turbulence.
Read the full abstract

Computation and Language

BharatSynthQA: A Context-Grounded Question Answering Dataset on Indian Temples and Religious Heritage

India's temples and other places of worship are among its richest repositories of cultural heritage, bringing together religious tradition, history, architecture, and art. Due to the widespread curiosity, people increasingly turn to Artificial Intelligence (AI) assistants for information about them. Current models are rarely trained or evaluated on this domain. We present BharatSynthQA, a passage grounded question answering (QA) dataset focused on Indian temples, which also covers other religious sites connected to them. Our pipeline crawls Wikipedia from topics seeded by authentic heritage sources, filters articles for topical relevance, and locations in India. Further it generates Boolean, factoid, and explanatory QA pairs with a fewshot prompted LLM. An LLM judge filters the pairs on answer correctness, retaining 98,677 of 131,494 generated pairs. Additionally, 300 human-annotated pairs are released separately as a reference set. Human evaluation of the subsample of this dataset showed 89.83% of the questions were formed correctly. Benchmarking five current LLMs shows that factoid token F1 is only 0.18-0.30 without the source passage but 0.73-0.77 with it, indicating that this domain is poorly captured in model knowledge. The dataset is available at https://huggingface.co/datasets/theshivam7/bharatsynth-qa to support research on culturally grounded QA and retrieval-augmented generation for Indian heritage.
Read the full abstract

Computer Vision

UltraVoxelGS: Voxel-First Feed-Forward Gaussian Splatting for 3D Ultrasound Reconstruction

We present UltraVoxelGS, a 3D ultrasound reconstruction method that achieves both fast reconstruction and physically faithful rendering for the first time, bringing live spatial feedback during 2D ultrasound scanning within reach. This is made possible by a feed-forward design that predicts a complete Gaussian scene representation from posed ultrasound images directly, entirely bypassing the per-scene optimization bottleneck that has confined prior methods to offline use. Realising such feed-forward prediction in this setting is non-trivial: unlike natural images, ultrasound slices share at most one-dimensional intersections and exhibit view-dependent appearance due to acoustic propagation and attenuation, violating the dense overlap and photometric consistency assumed by existing feed-forward Gaussian methods. Hence, to achieve fast reconstruction, we introduce a feed-forward voxel-to-Gaussian pass that decouples representation size from input count by lifting all posed slices into a fixed-size voxel support and predicting Gaussian parameters directly. For faithful rendering, we propose ultrasound-aware appearance rendering strategy factorize reflection from cumulative attenuation during training and exploit appearance continuity among spatially adjacent slices for efficient Render-time Appearance Adaptation. On both real-world and simulated datasets, UltraVoxelGS achieves best PSNR/SSIM compared with per-scene optimization baselines while reducing inference from 10-20 minutes to approximately 10 seconds (over 35 times faster) and scales to sequences exceeding 1,000 frames. Code is available at https://github.com/YuebyYue/UltraVoxelGS-camera-ready.
Read the full abstract

Analysis of PDEs

BMO Vorticity Strong Ill-Posedness for the 2d Incompressible Euler Equation

For the two dimensional incompressible Euler equation, Yudovich, and independently Serfati, extended the classical existence and uniqueness class to include vorticity with mild singularity. This class is sometimes called the generalized Yudovich class. Then, Vishik extended the uniqueness class to include vorticity in BMO, allowing for stronger singularities than the generalized Yudovich class. In this work, we prove strong ill-posedness of vorticity in BMO.
Read the full abstract

Computer Vision

Watching Itself Watch: Self-Auditing Visual Reliance for Video Reasoning

Video reasoning hinges on a small number of visually decisive tokens embedded in otherwise largely language-driven chains of thought. Yet prevailing post-training methods such as GRPO overlook this structure, assigning coarse trajectory-level rewards while requiring multiple rollouts and often costly external judges. We introduce Visually-Aware On-Policy Self-Distillation (VIOS), a post-training framework that identifies where a model under-utilizes visual evidence and concentrates learning at those positions. Our key insight is that a model can audit its own visual reliance: the divergence between its next-token distributions with and without the video reveals how strongly each prediction depends on the visual input. Measuring this reliance for both a student and an answer-privileged teacher exposes a per-token reliance gap, identifying positions where the teacher exploits visual evidence that the student overlooks. VIOS translates this signal into dense, targeted supervision through three complementary components: an exponential moving average (EMA) teacher that co-evolves with the student, a contrastive objective that encourages predictions to depend on the video, and a reliance-gap gate that focuses learning on visually under-utilized tokens. Crucially, VIOS requires only a single on-policy rollout from a single model, without external rewards, judges, or larger teachers. Across 12 benchmarks spanning general, temporal, spatial, knowledge-intensive, and long-video understanding, VIOS consistently improves three model backbones and outperforms contemporary video-reasoning systems, while training in over 9x fewer steps, 60x fewer FLOPs, and 111x fewer generated tokens. Project page: https://vision.cs.utexas.edu/projects/VIOS/
Read the full abstract

Optimization and Control

A Bearing-rate-free Implementation of a Scalar Estimator for Bearing-only Target Localization and Circumnavigation

This paper studies the problem of guiding an agent to localize and circumnavigate a stationary target at a prescribed radius using bearing measurements. Direct implementation of a classical scalar signed-distance estimator requires explicit evaluation of bearing rate signals that can be difficult to obtain in practice. To eliminate this dependence, we propose a bearing-rate-free (BRF) realization of the scalar estimator that recovers the signed-distance estimate algebraically, without numerical differentiation or a separate filter for estimating the bearing rate. We prove, under consistent initialization, that the proposed BRF realization inherits the guarantees previously established for the classical realization, including exponential convergence of both the target estimation error and radial tracking error, as well as the agent-target collision avoidance. Simulations show that, with noise-free bearings, the BRF realization achieves numerical accuracy comparable to that of an exact-bearing-rate benchmark and, at the higher noise levels tested, exhibits smaller radial tracking oscillations than the backward difference implementation of the classical estimator.
Read the full abstract

Robotics

The Relative Value of Demonstration Data: Execution Time Limits and Operator Mixtures in Behavior Cloning

Robot demonstration datasets combine demonstrations from teleoperators of uneven experience, so collection teams need to know how many demonstrations from one group match those from another. We measure this with a data multiple: the ratio of the numbers of demonstrations two groups need for their policies to reach the same success rate. On robomimic's simulated Square and Can tasks, we train behavior-cloning policies on demonstrations from the two teleoperators robomimic labels Better and the two it labels Worse, varying dataset size and the share of Worse demonstrations. On Square, within robomimic's 25-second time limit, an all-Worse training set needs 1.55 times as many demonstrations as an all-Better set. Most of this difference is late completion: policies trained on Worse demonstrations finish later, as the Worse teleoperators do, and with a 100-second limit the multiple falls to 1.11. On Can, most of the difference remains at 100 seconds. If each Worse demonstration were worth 1/1.55 of a Better one, as in an all-Worse set, a half-Better, half-Worse set on Square would need 1.22 times as many demonstrations as an all-Better set. In fact, it needs 13% more than that, and this excess persists at every time limit from 20 to 200 seconds. The excess remains when training draws demonstrations uniformly instead of in proportion to their length. On Square, within the 25-second limit, switching to these Worse teleoperators is cheaper at equal success only if they earn less than a third of the Better teleoperators' hourly pay, because Worse demonstrations also take longer to record.
Read the full abstract

All authors Jiaxuan Guo, Xingyu Ren, Youran Sun, Jingxin Yang, Kejia Zhang, Haizhao Yang

Machine Learning

Decision Aware Origin-Destination Demand Estimation with Structurally Constrained Data

Origin-destination (OD) demand plays an important role in transportation system design and management applications. Conventional OD estimation methods typically judge estimation quality based on how estimated demand aligns with observed traffic data and historical demand patterns. We argue that data alignment alone is insufficient; how errors in OD estimates influence downstream decision quality should also be considered. This paper develops a decision aware OD demand estimation framework that leverages structurally constrained traffic data to characterize an admissible demand region, within which we identify OD estimates that would lead to least regret in the downstream decision. Therefore, our OD estimates are both data-consistent and decision-relevant. Methodologically, we approximate the admissible demand region induced by physical network constraints through a geometric representation, which regularizes the search over feasible OD estimates. Within this regularized region, we use a convex surrogate of downstream regret to obtain a tractable decision aware optimization problem. Numerical experiments show that the proposed framework consistently reduces decision regret relative to conventional estimation benchmarks, demonstrating the value of aligning OD demand estimation with downstream decision quality.
Read the full abstract

Dynamical Systems

Rigidity and existence of profiled expansive solutions of the $N$-body problem with homogeneous potentials

We study expansive solutions of the $N$-body problem with homogeneous potentials of degree $-p$, where $p>0$. We establish asymptotic rigidity for arbitrary solutions with admissible profiles. For general admissible profiles and $0<p<2$, we obtain convergent power-log expansions whose exponents are governed by the leading profiles. In the hyperbolic case, we derive a more explicit analytic representation for all $p>0$, extending Chazy's Newtonian result. Together with the author's recent results [Wu, 2026], this yields a bijective parametrization of hyperbolic solutions near infinity by their asymptotic velocities and offsets. We also establish the existence of free-time minimizing solutions with arbitrary prescribed initial configurations and admissible asymptotic profiles for $0<p<2$, covering the hyperbolic, parabolic and hyperbolic-parabolic regimes. In the hyperbolic case, the result extends to all $p>0$, including possibly collisional initial configurations in the strong-force regime $p \ge 2$. Furthermore, we prove that when $p \neq 2$, every solution with an admissible asymptotic profile is eventually a free-time minimizer.
Read the full abstract

Group Theory

Extreme amenability and failure of the Lévy property for the trace-class special unitary group

We prove that the trace-class special unitary group, equipped with the trace-norm topology, is extremely amenable but is not a Lévy group, answering questions posed by Pestov. The failure of the Lévy property follows from a uniform non-concentration estimate for every nontrivial compact subgroup. Extreme amenability is established through a transverse spectral-gap estimate and trace-norm duality, yielding asymptotic invariance of Lipschitz observables under each fixed finite-dimensional special unitary subgroup. A limiting argument then gives fixed points for continuous actions on arbitrary compact Hausdorff spaces. We also formulate an abstract criterion for extreme amenability based on asymptotic quadratic invariance along a dense chain of compact subgroups. Neither argument uses concentration results for other Schatten exponents.
Read the full abstract

All authors Sergio Luigi Cacciatori, Takashi Shioya, Pietro Ursino, Shigeaki Yokota

General Relativity and Quantum Cosmology

Single and repetitive Penrose processes in genuine uncharged Kerr-Bertotti-Robinson black hole

We establish an analytical framework for the Penrose process in the genuine uncharged Kerr-Bertotti-Robinson (KBR0) black-hole spacetime within a Bertotti-Robinson (BR) electromagnetic vacuum, consistently linking particle decay, radial transport, thermodynamic states, and geometric reconstruction by distinguishing coordinate Killing energy from thermodynamic mass-generator energy and performing multibranch parameter mapping from thermodynamic states to background metric parameters. Furthermore, we present the analytical solution for the decay of equatorial massive neutral particles at a common radial turning point inside the ergoregion, the orbital accessibility criteria, and the six physical constraints required for a complete extraction event. Numerical calculations demonstrate that for both near-extremal and lower-spin inputs, in both single and repetitive Penrose processes, the net energy return on investment ($\eta_{\text{EROI}}$) and the rotational energy utilization efficiency ($\eta_{\text{EUE}}$) decrease with increasing external magnetic field, whereas the rotational energy reservoir consumption rate per unit incident energy ($\eta_{\text{thermo}}$) increases. For the inputs considered, the feasible domain and the upper bound of the initial magnetic field for single extraction are constrained by two boundary mechanisms: the obstruction of parent-particle infall and the loss of negative energy in the captured fragment. Finally, for repetitive extraction processes within the surveyed regime, all sequences terminate due to orbital accessibility obstructions, explicitly local-slope failure and far-field incident-branch closure, with the final black-hole state retaining most of its initial rotational energy reservoir. Moreover, the black-hole state evolves along a common integral curve independent of the parent particle's specific energy, which solely alters the sequence endpoint.
Read the full abstract

All authors Ruo-Cen Wu, Hanjun Zou, Meng Xu, Yun-Long Zhang

Classical Analysis and ODEs

Pointwise estimates for an anisotropic product type singular integral on the sphere

Let $\Gamma=\{1,\dots,d\}$, $d\geq 2$. For each $j\in \Gamma$, let ${\ell}_{j}\geq 1$ be an integer and let $\alpha_{j}\in (1/2,1)$. We obtain a pointwise estimate for the integral \begin{equation*} \int_{\mathbb{S}^{d-1}}\prod_{j\in \Gamma} |x^{\ell_{j}}_{j}-\xi_{j}|^{-\alpha_{j}} d \sigma(x), \end{equation*} in the region \begin{equation*} \{\xi\in \mathbb{R}^{d}: \xi_{j}\neq 0,\;j\in\Gamma,\;\; \sum\nolimits_{j\in F}|\xi_{j}|^{2/\ell_{j}}\neq 1,\; F\subseteq \Gamma\}. \end{equation*} These estimates are used to prove a Tomas - Stein type restriction theorem for the oscillatory integral operator $Tf(x)=\int_{\mathbb{R}^d}e^{\dot{\imath}h(x,y)}f(y)dy$, with the phase $h(x,y)=\sum\nolimits_{j\in\Gamma} a_{j} x_{j}^{\ell_{j}} y_{j}^{m_{j}}$, where, for each $j\in \Gamma$, $a_{j}\neq 0$ is a real number and $m_{j}>1$ is an integer.
Read the full abstract

Robotics

Flip Anything with Agentic Robotics Self-Learning

Flipping is a fundamental robotic manipulation skill that requires reorienting a target object in a stable pose on a planar support surface by rotating it 180$^\circ$ about an axis parallel to the support surface – think of flipping a coin from heads to tails. Although flipping is a common objective in industrial tasks such as parts inspection, mechanical assembly, and tool use, coding a bimanual robot with grippers to flip diverse objects is a significant challenge. In this paper, we show that a coding agent can autonomously develop and refine object-flipping strategies through iterative learning in physics simulations across diverse object geometries. Progressive skill acquisition, from basic handover to toppling and pushing, robust grasping and sensing, and thin-object manipulation with recovery, improves task success from 24.7% to 95.0%. The resulting graph-based code integrates high-level strategy selection with reactive control, enabling multi-stage flipping on physical robots.
Read the full abstract

Robotics

Reliability-Adaptive Multimodal Relative Pose Tracking for Underwater Leader-Follower Robots

Underwater leader-follower formation control requires continuous relative-pose estimation despite degraded sensing conditions and limited onboard computation. This paper presents a reliability-adaptive multimodal perception framework for six-degree-of-freedom relative-pose tracking between two moving underwater robots. The framework integrates complementary visual, acoustic-ranging, and hydrodynamic measurements while accounting for their distinct physical characteristics and time-varying reliability. It adaptively weights modality contributions and models temporal dependencies with linear computational complexity, enabling efficient pose estimation on embedded hardware. Extensive experiments were conducted using a custom-built AUV platform, with 648 pool trials spanning two leader vehicle types, three illumination conditions, and nine trajectory patterns. Evaluation yields an average translational error of 3.8 cm and an average orientation error of 2.4{\deg}. On an onboard NVIDIA Jetson AGX Orin, the fusion network achieves an inference latency of 17.62 ms, providing a 2.69-fold speedup over the evaluated Transformer-based fusion baseline. Closed-loop experiments further demonstrate stable formation maintenance with a moving leader under onboard computational constraints. A multimodal dataset containing measurements from all 648 trials will be made publicly available.
Read the full abstract

All authors Mingyang Yang, Zeyu Sha, Xiaorui Wang, Jun Wang, David Chen Wang, Feitian Zhang

Soft Condensed Matter

Topology to Elasticity Mapping in Thermal Glass

Building on our previous work on thermal nonaffine metrics and thermal strain fields, we examine how local steric geometry controls thermally driven rearrangements in an amorphous glass-forming solid. As a structural descriptor, we employ the topological steric "theta-mismatch" order parameter, which quantifies local deviations of bond angles from the steric reference geometry and is used to distinguish low- and high-mismatch particle populations. We then compare their thermal nonaffine response, thermal deviatoric strain, and thermal volumetric strain over a broad temperature range at several state points. High-Delta-theta sites preferentially exhibit larger thermal nonaffine motion and stronger thermal deviatoric strain, particularly in the deeply jammed regime, indicating that geometrically frustrated regions are preferentially associated with localized mechanical distortions. The thermal volumetric response exhibits a distinct mismatch-dependent crossover: low-Delta-theta sites predominantly contribute to the small-strain response, whereas high-Delta-theta sites contribute more strongly to the large-strain tail. The contrast between low- and high-mismatch populations is most pronounced in the deeply jammed regime, remains identifiable through the intermediate viscoelastic regime, and progressively weakens upon heating toward the liquid state. These results establish a direct connection between local steric frustration and thermally activated mechanical response, showing that thermal nonaffine metrics and thermal strain fields provide a physically transparent bridge between local structure and thermally driven rearrangements in amorphous glasses.
Read the full abstract

Differential Geometry

Lower bound of the Calabi functional and non-Archimedean K-energy

We prove Donaldson's conjecture which describes the infimum of the Calabi functional by the normalized non-Archimedean K-energy. The proof makes use of proximal equation for the K-energy. To obtain approximate smooth solutions, a continuity path of prescribed scalar curvature equation is coupled with a version of Chen's $\e$-subgeodesic. It does not require a global smooth solution to the Calabi flow.
Read the full abstract

Machine Learning

Attention at Absolute Zero

Attention scores candidate supports of an answer, which we call witnesses, and reads the scores at a temperature. We build a higher-order attention layer whose state lives on pairs of items: for a pair (a, b), each third item k is a witness judged through its legs (a, k) and (k, b). It reads witnesses at zero temperature, keeping the best and the number tied with it, passes on only bits comparing such counts, rebuilds its pair state from the input and these bits at every loop, and is trained from answers alone. Each choice removes one way soft attention fails at size: dispersed weight, unresolved gaps between learned scores, and a drifting soft state. In maximum retrieval it is right on at least 99.95% of sets 1,000 times larger than in training, on all ten seeds; trained with a soft count, the same network falls to 42% at 4,096 items. On CLRS-30 without hints, a general form of the layer (half its channels read at zero temperature, numbers embedded, state carried) beats the best published model, which uses hints on some tasks; the full cold layer is exact on every test input of 9 of the 12 tasks whose answers depend on numbers only through comparisons. From yes-or-no answers it decides 2-SAT and two-colouring exactly at 500 variables and, as our theory predicts, not parity equations. Beside a frozen 7B language model it chains family relations exactly at 1,000 statements after training on 16, where the model alone falls to 44%; finite checks on the weights prove every block that learned correct at every size, on acyclic contexts and given what the model reads. On CLUTRR it is right on every story of five to ten relations given the graphs, learning the 62 rules exactly; from text, with the model frozen, no hand-written rule and only a small read-in trained, it lifts an open 7B model from 33% to 65%, above every system we found without hand-written rules except frontier reasoning models.
Read the full abstract

2611.11937 · 9 Nov · v3: corrects the shared-block column of Supplementary Table 4 (now the final shared block, 97.20%, as in the text). 51 pages, 9 figures, 1 Extended Data figure, 4 supplementary tables. Replaces v1, "Attention Is Only as Good as Its Witnesses". Code (Apache 2.0): https://github.com/NeoVand/cold-attention ; demo: https://neovand.github.io/cold-attention ; models: https://huggingface.co/mohsenvand

Machine Learning

Constrained Variational Policy Optimization for Hybrid Action Spaces

Constrained reinforcement learning in hybrid action spaces must coordinate discrete modes and mode-conditioned continuous controls under long-horizon cost constraints. Existing constrained hybrid-action methods typically optimize factorized policies directly in parameter space, whereas Constrained Variational Policy Optimization (CVPO) separates nonparametric improvement from parametric projection in its original continuous-action formulation. We introduce Hybrid Constrained Variational Policy Optimization (H-CVPO), extending this variational framework to hybrid action spaces. Under boundedness and strict feasibility, its constrained E-step has zero duality gap and, when the dual optimum is attained at positive temperature, yields a closed-form joint target distribution over hybrid actions through exponential reward-cost reweighting. Using the exact hybrid KL decomposition, we formulate a factorized M-step that fits both policy components to this target under separate discrete and conditional-continuous trust regions. For exact E/M updates, we prove fixed-reference surrogate improvement and conditional reward-cost bounds with a sufficient feasibility-margin condition. For approximate updates, we establish the existence of a dual minimizer under continuity on a compact domain and use residual-based admissibility to determine whether to project the target. Ablations identify cost-aware reweighting as an important contributor to constraint control. Extensive experiments in simulated hybrid-control environments show that H-CVPO achieves favorable reward-cost trade-offs compared with hybrid-action baselines.
Read the full abstract

Quantum Physics

AI-Driven Discovery of Entanglement Witness in Spin Hamiltonians

Entanglement is a crucial resource in quantum science and engineering, but detecting it in many-body states from experimentally accessible observables remains challenging. Here we introduce EntNet, an AI-driven framework to discover compact combinations of spin observables that accurately distinguish entangled from fully separable many-body mixed states in one-dimensional spin-$1/2$ chains governed by short-range Hamiltonians. We construct local-unitary-invariant representations of measurable Pauli correlators up to four-body order, sparsely select 16 candidate invariant feature channels, and learn their spatial structure using a Transformer architecture with rotary positional embeddings. Using only expectation values of these low-order observables, EntNet achieves $>98\%$ recall on certified entangled states with near-zero false-positive rates on separable states. Remarkably, it exhibits transferability across system sizes and distinct Hamiltonian families, including frustrated and disordered spin models, with minimal loss of accuracy. Beyond detection, EntNet provides physical insight into entanglement detection by revealing a selective spatial organization of pair correlations, while its finite-temperature decision boundary tracks the excitation gap of Jordan-Wigner quasiparticles. These results establish a transferable and physically interpretable framework for data-driven discovery of entanglement witnesses from local correlation functions.
Read the full abstract

Computer Vision

Scale-Consistent Monocular Fisheye Visual Odometry via Spherical Bundle Adjustment-Guided Unsupervised Depth Learning

The wide field of view of fisheye cameras and the simplicity of a monocular setup make monocular fisheye visual odometry (VO) attractive for robotic localization and mapping. Nevertheless, scale drift remains a major challenge over long trajectories. Unsupervised depth learning can help to mitigate this issue, but the pronounced distortion of fisheye images poses challenges for accurate depth estimation. This article presents S$^2$D-FVO, a monocular fisheye VO framework integrating spherical bundle adjustment (SBA)-guided unsupervised depth learning to improve scale consistency and depth estimation accuracy. Given that pixel reprojection residuals do not uniformly reflect angular deviations between viewing rays and that the first-order linearization of a scalar angular reprojection residual is insensitive to perturbations along one tangent direction, SBA employs tangent-space angular reprojection residuals derived from the spherical logarithmic map. These residuals retain sensitivity along both tangent directions, with their norms equal to the corresponding scalar angular residuals. A median depth-based scale regularization term is further integrated into SBA to constrain the reconstruction scale. Building on these formulations, SBA replaces the pose estimation network during unsupervised depth learning, providing more reliable relative poses for view synthesis. In turn, depth estimates from the well-trained depth estimation network provide scale references for SBA, improving scale consistency while allowing individual 3-D landmark positions to be refined. Extensive experiments conducted on multiple public datasets, as well as two newly collected large-scale datasets, demonstrate that the proposed framework outperforms representative existing approaches in both depth and trajectory estimation accuracy and is applicable to different camera models.
Read the full abstract

All authors Wanhui Luo, Hongbo Zhao, Yi Zhou, Qijun Chen, Yiming Nie, Rui Fan

Algebraic Geometry

Gersten resolutions for logarithmic forms

Let $X$ be a smooth variety over a field $k$ of characteristic 0. We show that (two versions of) the Zariski sheaves of regular $k$-linear logarithmic differential forms on $X$ (in the most naïve sense, i.e. sums of products of logarithmic differentials) have Gersten property (and therefore coincide). It turns out that the values of these sheaves are free abelian groups. As a consequence, for any integer $n>0$, these sheaves modulo $n$ also have Gersten property. The proof is an elementary application of the Voevodsky's $(-1)$-machinery and consists of assembling several well-known results.
Read the full abstract

Human-Computer Interaction

ExCritique: An Interactive Visual Interface to Evaluate AI-Generated Explanations at Scale

Vision-language Models (VLMs) can generate natural language explanations for complex decision-making scenarios. However, these explanations can sometimes be misleading, sounding convincing while containing incorrect or unsupported information. Automated explanation-quality metrics help researchers and practitioners evaluate large collections of explanations, but their scores alone provide limited support for identifying failure cases and understanding the reliability of these metrics. This paper presents a case study on how interactive visualizations can help researchers and practitioners evaluate both AI-generated explanations and automated explanation-quality metrics across large datasets. Based on formative interviews with nine AI researchers and practitioners, we designed an interactive visualization tool named ExCritique, situated in the task of Visual Question Answering (VQA). ExCritique visualizes the distributions of multiple automated quality metrics, allows users to explore thousands of data items, filter items and inspect the evidence behind individual scores. We compared ExCritique with a baseline interface without interactive visualizations in a controlled study involving 16 AI researchers and practitioners. Our findings suggest that the interactive visualizations in ExCritique helped users identify explanation-quality issues more efficiently and better understand automated quality metrics, supporting an effective way to measure the reliability of AI-generated explanations.
Read the full abstract