Papers

Quantum steering is an important type of quantum correlation with potential applications in one-sided device-independent (1SDI) quantum information protocols. In this contribution, we explore the time-dependent dynamics of Gaussian quantum steering in coupled lossy optical waveguides using the covariance matrix formalism. Optical dissipation is explicitly incorporated to account for realistic propagation losses in coupled waveguides. We systematically investigate the influence of the optical loss rate, purity, nonclassicality, and squeezing of the input Gaussian states on the generation and evolution of steering. The results demonstrate that increasing nonclassicality and squeezing enhances Gaussian steering, while optical loss and reduced purity progressively suppress it. However, a considerable amount of steering can persist over finite propagation intervals in the presence of optical loss. We further examine that steering is symmetric for input states with identical nonclassicality, while unequal nonclassicalities lead to directional steering asymmetry, offering a way to control the steering direction through input-state engineering. These results demonstrate that coupled optical waveguides can support controllable Gaussian quantum steering under practical loss conditions and offer a promising integrated-photonic platform for asymmetric quantum-information protocols, including 1SDI quantum key distribution (QKD) and quantum communication.

We propose an Urgent Interactive HPC workflow that dynamically integrates high-fidelity 3D nonlinear analysis with surrogate neural networks (NNs) to enable rapid decision-making during large-scale earthquakes. This approach achieves both "Resource Harvesting", which utilizes idle computing capacity during non-emergency periods, and immediate response during crises. By developing two specialized HPC kernels, the proposed method reduces energy-to-solution by 76% and improves throughput by 3.7-fold during normal operations to efficiently construct training datasets, while during emergencies, it couples NN-based inverse analysis with physics-based simulations and dynamic refinement to reduce conventional computational costs by over 97.9%, enabling the generation of highly reliable spatial time-history ground motion distributions within 30 minutes post-earthquake.

We present commonsense ranked search (CoRS), a novel path planner that turns an abstract instruction into a route that follows commonsense. While existing methods respect the considerations written down in advance, a robot working among people must follow those left unstated too, as with a wet floor that a worker avoids without being told. CoRS leverages large language models (LLMs) and vision-language models (VLMs) as commonsense knowledge to reason about these latent considerations in its planning. Given an abstract instruction (e.g., “move carefully”) and visual observations of each region in the environment, CoRS derives a consideration for each region, as in “this wet floor is slippery and worth a detour.” It then compares the considerations between regions to see which of the two the robot should avoid more, as in “the crowd is worse than the wet floor.” These judgments sort the regions into a commonsense ranking, whose costs drive a conventional search that always returns a valid route. We build a benchmark for planning under latent considerations, with three environments, 1350 problems, and five instructions at three levels of abstraction. Experiments show that CoRS discovers the unstated considerations and goes around the ones worth a detour while crossing the rest, a behavior that recent LLM-based planners do not achieve.

We construct the Yukawa couplings of the quark sector as overlap integrals of Gaussian wave functions in extra spatial dimensions. Assuming that the wave function of each chiral fermion is localised at a point in the extra dimensions, while that of the Higgs doublet is flat, the Yukawa matrix elements are determined by the relative positions of the three left-handed quark doublets, $q_1$, $q_2$, $q_3$, and the six right-handed quark singlets, $u_1$, $u_2$, $u_3$, $d_1$, $d_2$, $d_3$. We employ numerical optimisation techniques from machine learning to identify field configurations that satisfy all current experimental constraints on the quark Yukawa sector. In this framework, the large hierarchy of quark masses arises naturally, since the overlap integrals of Gaussian wave functions are exponentially sensitive to ${\mathcal O}(1)$ separations in the extra dimensions.

Magnetic fields regulate the multi-phase interstellar medium (ISM) of spiral galaxies, yet their behavior across ISM phases remains poorly constrained observationally. We present a comparative polarimetric study of the nearby spiral galaxy NGC 6946 using far-infrared (FIR) data at 154 $\mu$m from SOFIA/HAWC+ and radio continuum data at 6 cm from the VLA and Effelsberg telescopes, which probe magnetic fields in the cold dust-emitting and diffuse synchrotron-emitting phases of the ISM, respectively. The synchrotron polarization reveals a coherent large-scale spiral magnetic field, with high polarization fractions in interarm regions and strong depolarization in star-forming arms. The FIR polarization fraction is lower on average and shows stronger spatial variations. Both tracers decrease systematically with increasing star formation activity, indicating that star-formation-driven turbulence reduces magnetic field ordering; the synchrotron polarization fraction is strongly suppressed in the central kiloparsec but stable across the mid disk, with a tendency to increase beyond $R \approx 7$ kpc, while the FIR polarization fraction remains low ($\sim$2-3%) and approximately constant over the radial range where it can be reliably measured ($R \lesssim 5.4$ kpc). Our results indicate that NGC 6946 hosts a globally ordered magnetic field locally modulated by turbulence. This study highlights the complementary nature of FIR and radio polarimetry in probing magnetic fields across ISM phases and provides quantitative constraints on how magnetic field ordering varies between the cold and diffuse ISM components.

Lithographic patterning of nanoscale devices typically depends on photoresists, wet processing and multistep pattern-transfer workflows that impose both environmental and scaling constraints. Focused laser irradiation of (Ni,Co)1+2xTi1-xO3 under a reducing atmosphere drives localized oxygen removal, directly converting the insulating oxide into embedded metallic features without etching or chemical processing. Here we demonstrate a photoresist-free patterning method based on light-induced redox transformations in (Ni0.4Co0.6)3O3 (x = 1) thin films. Our results indicate that the transformation is self-limiting, as the high optical absorption of the emerging metallic phase restricts further energy penetration and confines the reduction front, producing conductive layers with electrical properties comparable to bulk metals. This reversible, orientation-preserving transformation is characteristic of a topotactic redox process, enabling direct nanoscale patterning without photoresists. The process is reversible through re-oxidation and is compatible with standard semiconductor wafer substrates. The role of Ti as a reduction inhibitor is discussed. These results establish a route to direct nanoscale patterning based on controlled redox chemistry, offering a potential alternative to conventional lithography for functional thin-film materials.

A wrist-mounted camera for UMI-style data collection must do two jobs: record the manipulation and localize in the scene. Most handheld devices localize online from workspace-facing views crowded by hands and objects, or add dedicated tracking hardware. Room-scale bimanual capture therefore still tends to instrument the operator or the scene for accurate localization. We present KIWI (Kinematic Interface for the Wild), a capture kit whose only electronics are off-the-shelf cameras. Our core system splits the two jobs across the two lenses of a 360-degree camera. The rear lens faces the room and builds a shared metric map that registers both hands, and an optional head camera, in one frame without workspace co-visibility; the front lens records the manipulation, and offline IMU fusion bridges front-lens tracking loss. Through our quick-release plate, the camera module attaches to chopstick grippers, parallel-jaw grippers, hand-wrist mounts, or robot flanges. Across six bimanual recordings, combining the rear and front lenses failed to localize only 0.1% of query frames, whereas front-only bimanual feature alignment failed on 24.8% of frames and lost one recording entirely; against evaluation fiducials, localization error stayed within 4.5 mm. KIWI's recovered poses were sufficiently consistent for the four wrist streams alone to reconstruct the scene as a 3D Gaussian splat.

Let \[ A_\omega=-\partial_x^2+\rho\dot B_x(\omega), \qquad \rho\ne0, \] be the Dirichlet Schrödinger operator on $(0,\pi)$ with a spatial Gaussian white-noise potential, realized pathwise via quasi-derivatives. Here $\dot B_x$ denotes the distributional derivative of Brownian motion with respect to the spatial variable $x$. For $\kappa\ne0$, we consider the cubic nonlinear Schrödinger equation \[ i u_t=A_\omega u+\kappa |u|^2u. \] We first prove a zero-set theorem for locally real-analytic functions on classical Wiener space. As a consequence, on a single event of probability one, no nontrivial finitely supported integer combination of the random eigenvalues vanishes, and the quartic twist matrix is nonsingular for every finite tangential set. After fixing a path in this event, we combine these qualitative nondegeneracy properties with a partial quartic Birkhoff normal form adapted to the KAM scheme. For every finite nonempty tangential set \(J\) of cardinality \(b\), we obtain a Cantor family of linearly stable, real-analytic, small-amplitude invariant \(b\)-tori. The family is parametrized by a Cantor subset of \([\nu,2\nu]^b\) whose relative measure tends to one as \(\nu\to0\).

Implicit Neural Representation (INR) leverages neural networks to represent discrete signals such as images as continuous ones, where the network weights serve as a compact form of the signal itself. Most existing INR methods adopt Multi-Layer Perceptrons (MLPs) as their backbone. Since these models render each pixel independently, they inherently fail to exploit the spatial correlations that exist between neighboring pixels. In contrast,convolutional INRs can process pixels in parallel while inherently accounting for inter-pixel dependencies, making them a more natural fit for representing images. Nevertheless, convolutional INRs remain relatively underexplored, and the majority of them rely on fixed architectural settings, leaving little room for image-specific adaptation. In this paper, we investigate network customization for convolutional INRs. We replace conventional filters with irregular directional kernels, whose allocation is guided by the directional energy in the image spectrum, i.e., directions exhibiting stronger energy are assigned a larger number of kernels, enabling content-tailored convolution settings. These kernels are further reformulated via an orthogonal basis to achieve a superior sparse representation. Moreover, we introduce an annealed Gumbel-Softmax-based mechanism for kernel-level activation function selection, which gives the most suitable activation function for each convolution kernel. Extensive experiments demonstrate that our method, namely C$^{2}$-INR, achieves superior performance against state-of-the-art approaches under comparable parameter budgets across a wide range of image processing tasks, including representation, inpainting, and super-resolution.

During a core-collapse supernova (CCSN), the emitted electron antineutrino ($\bar{\nu}_e$) burst initiates inverse beta decay (IBD) in the outer hydrogen layer of the stellar envelope. This leads to the production of a characteristic 511 keV photon signal, termed the Gamma-Ray Echo. The same neutrino burst, when detected on Earth in large neutrino observatories such as Hyper-Kamiokande, will offer information about the $\bar{\nu}_e$ content of the neutrino flux reaching Earth. In this work, we show that comparing the $\bar{\nu}_e$ content inferred from the gamma-ray echo with that inferred from terrestrial neutrino observations during a nearby CCSN event can offer a promising probe of large neutrino emission anisotropies and long-baseline propagation effects. In the latter case, the combined SN-Earth observations effectively act as a near-far detector configuration for neutrino propagation over an astrophysical baseline. We find that, for a near-Earth supernova, anisotropies or non-standard propagation effects at the level of tens of percent can be tested using gamma-ray telescopes with effective areas of $\mathcal{O}(10^4)~\mathrm{ cm^2}$. Therefore, our results motivate the development of the next generation of large effective area gamma-ray telescopes operating in the MeV gap.

Natural language generation (NLG) ranges from highly constrained tasks, such as machine translation, to open-ended tasks like dialogue generation. Between them lie peer-review generation and image captioning, where one input admits several valid, overlapping outputs. We measure this intermediary region with Von Neumann entropy ($\hat{H}\approx0.6$ vs. 0.15 translation, 0.92 dialogue). Single-reference training discards valid perspectives, while multi-reference training introduces high-variance, unaligned targets. To tackle this, we propose target-side consolidation, that is, merging all references for an input into a single target which preserves every distinct information unit. To this end, we introduce MERC-67K, pairing 36,537 papers and 31,014 images with their original and consolidated references. In human evaluation, models trained using consolidated reviews obtain an average rank of 1.70 (out of 4) for coverage against 2.51-2.98 for baselines, and win 63.5-74.0% of pairwise comparisons. Models trained using consolidated captions obtain an average rank of 1.87 for overall caption quality (accuracy and naturalness) against 2.64-2.82 for baselines and obtain the highest reference-free LongCLIPScore (0.839 vs best baseline's 0.816). For reviews, multi-reference training often under-performs zero-shot inference, and consolidation's gains over multi-reference training hold across Llama3.1, Mistralv0.3, and Qwen2.5. Consolidation uses $\sim$4$\times$ (reviews) and 5$\times$ (captions) fewer individual references during optimization. Dataset and code will be released upon acceptance.

This work disproves the 1-leaky positive semidefinite edge-deletion conjecture and replaces it with a sharp theorem. For every leak level $\ell$ and edge $e$, one has $|Z^+_{(\ell)}(G)-Z^+_{(\ell)}(G-e)|\le 2$. More generally, if two graphs differ only on edges with both endpoints in $S$, their parameters differ by at most $|S|$. An endpoint-sensitive refinement recovers an increase of at most one whenever some minimum set for $G-e$ contains an endpoint of $e$. Both signs are sharp for every positive leak level. Joining two copies of $K_{\ell+1}$ by a bridge gives $Z^+_{(\ell)}(G)=2\ell$ and $Z^+_{(\ell)}(G-e)=2\ell+2$. For every $\ell\ge 2$, a connected clique-leaf pair of order $2\ell+3$ gives the opposite difference. The remaining positive-difference one-leak case is attained by connected graphs on nine vertices with $Z^+_{(1)}(H)=4$ and $Z^+_{(1)}(G)=6$. Explicit forcing sequences, fort certificates, and an exact verifier check the finite extremal example and stress-test the general results.

Simulations for learning-based autonomous colonoscopic navigation focus mainly on fully actuated capsule robots, failing to capture the contact-rich navigation of long and flexible clinical colonoscopes. We present the Autonomous Robotic Colonoscopy Gym (ARCGym), an open-source reinforcement learning environment and benchmark for image-based navigation in clinically derived deformable colon anatomies. ARCGym supports multiple types of colonoscope robots, spanning capsule robots and flexible endoscopes, with this work focusing on flexible endoscopes including magnetic-driven tip actuation and clinically used proximally translational actuation. This work includes five CT-reconstructed colons representing typical clinical scenarios, a set of clinically meaningful navigation subtasks, and unified success metrics. We introduce a reward combining depth-based lumen alignment with a lumen-visibility score to improve learning under occlusions. Experiments across tasks, robots, and anatomies show that autonomous navigation remains challenging for both magnetic-driven and proximal-insertion flexible robots, with proximal-insertion actuation remaining an open problem.

We consider minimum-time control for the linear system $$ \dot{z}(t)=Az(t)+Bu(t),\qquad \|u\|_{L^\infty(0,\infty;\mathbb{R}^m)}\leq 1, $$ with $A\in\mathbb R^{n\times n}$ and $B\in\mathbb R^{n\times m}$. While the individual point-target and ball-target problems are classical, we study a different question: when are their optimal controls exactly compatible, in the sense that, for every nonzero initial state $x$ and all sufficiently small $\varepsilon>0$, the optimal control for the tolerance ball $\overline{B}_\varepsilon(0)$ is precisely the restriction of the point-target optimal control? We prove that this {\it{exact truncation}} property is equivalent to the rigidity conditions $$ B B^\top=\beta I_n,\qquad A+A^\top=2aI_n, \qquad \beta>0,\; a\leq 0, $$ and also to Euclidean radiality of the point-target minimum-time function. Thus exact truncation holds precisely when the sublevel sets of the point-target minimum-time function are Euclidean balls centered at the origin, matching the geometry of the tolerance targets. Moreover, the local property automatically extends to every $0<\varepsilon<|x|$, and the resulting optimal times and point-target optimal feedback are both explicit.

The hypersensitivity and turnpike phenomena in the optimal control of an inverted pendulum are investigated from a dynamical-systems perspective. We show that, for a fixed terminal time and a fixed terminal state optimal control problem, (1) the hypersensitivity originates from the fractal structure of the set of initial adjoint variables in the associated Hamiltonian dynamics, (2) the turnpike arises from slow dynamics in the vicinity of a degenerate center manifold, and (3) the escape channels are formed by normally hyperbolic invariant manifolds (NHIMs). As a consequence, small perturbations in the initial adjoint variables lead to qualitatively distinct extremal trajectories, resulting in severe numerical instability in trajectory optimization. Both the fractal structure and the invariant sets are characterized numerically and analytically.

Equilibria of the Hamiltonian system associated with Pontryagin's minimum principle exhibit an exact doubling of the natural spectrum and, under a simple pairing condition, a Jordan block at every simple purely imaginary eigenvalue. Consequently, the classical Lyapunov Center Theorem does not apply to the augmented system, and no periodic orbit with nonzero optimal control bifurcates at linear order. We establish this mechanism in general and show that an optimal-control-induced periodic family emerges at cubic order in a Lindstedt–Poincaré expansion. The mechanism is illustrated in closed form for the pendulum and evaluated numerically for the planar $L_2$ equilibrium of Hill's restricted three-body problem, where the third-order approximation is validated against an independently computed family of periodic orbits.

Blind and low vision (BLV) screen reader users construct mental models of user interfaces (UIs) through incremental screen reader interaction, a time-consuming and cognitively demanding process complicated by modern interfaces that may not convey dynamic content accessibly. We interviewed eleven BLV screen reader users about UI understanding and derived design objectives that informed Coral, a context-aware browser extension. Coral synthesizes interface and user context to generate screen reader-narrated gists which holistically notify users of interface states and changes relevant to their likely goals and ongoing interaction. In a comparative evaluation with eight BLV screen reader users, participants used Coral to form initial expectations and interpret interface changes, and reported spending less time and effort manually verifying interaction outcomes. Together, our findings provide deeper insight into how BLV screen reader users navigate gaps in their UI understanding, and how intelligent support that combines interface and user context can bridge these gaps.

Expert demonstrations often specify what a robot should do, but not how fast it can do it. Imitation Learning (IL) inherits demonstration timing, while directly accelerating the learned motion can fail when faster execution changes the robot-object dynamics. We study faster-than-demonstration execution as a dynamics-aware control problem and introduce Expert-Play Contouring Control (EPCC), which combines slow expert demonstrations with fast, non-expert play. Expert demonstrations train a latent trajectory generator whose predictions are reparameterized into a time-independent contour of successful task progression, while play trains a world model (WM) of fast-action outcomes. At deployment, our proposed planner uses the WM to optimize actions that makes maximize progress along the expert-derived contour while penalizing deviation from the intended task evolution. Averaged across three visuomotor manipulation tasks, EPCC achieves a $2.0\times$ the throughput of the IL baseline, including $2.2\times$ that of the throughput of the strongest acceleration baseline on a task with interaction-sensitive object dynamics. Our analysis shows that the gains concentrate where faster execution changes robot-object evolution. Together, our results highlight a simple yet effective principle: demonstrations provide task intent, while play data provides the dynamic coverage needed to execute that intent faster.

Gödelian incompleteness yields arithmetical sentences \(A\) such that \(\mathsf{PA}+A\) and \(\mathsf{PA}+\neg A\) are both consistent. Are such extensions equally legitimate? I propose graded epistemic arithmetical non-pluralism: justification for choosing between them varies with the set-theoretic strength of \(A\). I defend \(\mathsf{PA}+\mathsf{Con}(\mathsf{PA})\) and show that Friedman's concrete incompleteness results pose a serious challenge to non-pluralism about first-order arithmetic. Resolving the selection problem for such sentences turns on justifying large cardinals. Defending graded non-pluralism thus engages Gödel's programme and the justification of very large cardinals.

Dialectal Arabic machine translation (MT) remains challenging despite recent progress in Arabic language technologies, particularly because effective translation requires modeling not only semantic content but also dialectal variation, conversational context, speaker and addressee characteristics, and sociolinguistic appropriateness. Moreover, conventional MT metrics provide limited insight into the linguistic errors produced by dialectal systems. We present the AlexandriaX 2026 Shared Task on Dialectal Arabic MT, which addresses these challenges through three complementary subtasks: (1) context-aware English-to-Dialectal Arabic dialogue translation across 13 Arabic varieties, (2) cross-dialect Arabic translation in the financial domain covering six Arabic dialects, and (3) span-level MT error detection and classification using linguistically motivated error categories across five Arabic varieties. The shared task attracted 38 registrations for Subtask 1, 33 for Subtask 2, and 35 for Subtask 3. Twelve unique teams submitted their system description papers, all of which we accepted for publication. The best system on Subtask 1 achieved 30.42 spBLEU in the constrained track and 33.49 spBLEU in the unconstrained track. On Subtask 2, the top system obtained 28.40 spBLEU. On Subtask 3, the best system achieved an overall score of 49.82, outperforming the 24.29 baseline. Taken together, the results of the leading systems across the three subtasks highlight the benefits of explicit dialect modeling, context-aware generation, retrieval and reranking, and specialized approaches to interpretable MT error analysis. All the resources of AlexandriaX 2026 shared task are publicly available, including data, baselines, and evaluation code on our project page: https://alexandriax.dlnlp.ai.

Different tasks performed by legged robots impose distinct torque and speed requirements on actuators. Existing robotic actuators are generally optimized at the component level for metrics such as torque or power density, without explicit task guidance. System-level optimization across components such as motors, gearboxes, and sensors is challenging because of the high computational cost and coupling among mechanical, electrical, and electromagnetic behaviors. Consequently, improvements in individual components may not translate into better robot performance in a specific task. To this end, we present a systematic optimization framework for task-oriented co-design of actuator hardware and control. First, surrogate models are employed to accelerate motor evaluation and support global exploration of the coupled design space. Then, a hierarchical mixed-variable optimization strategy is adopted, combining discrete enumeration with continuous search over dimensions and real-valued indices. These indices are rounded to select admissible values for the remaining discrete choices before each evaluation. Within this search, rated output torque density and task performance are jointly optimized, with Bezier-parameterized joint torque profiles determined for each hardware candidate. Finally, the effectiveness of the proposed framework is validated through actuator fabrication and experiments on a two-degree-of-freedom jumping leg. Based on its measured mass, the fabricated prototype achieves a nominal rated output torque density of 35.7 N m/kg, approximately 60% higher than that of a widely used commercial geared joint actuator, while being 18.6% lighter. Under matched bench conditions, it achieves 12.0% greater jump height at twice-rated torque. Together, these results demonstrate a systematic route from task requirements to actuator design and control.

Since the discovery of charge density wave (CDW) and superconductivity, kagome metal AV3Sb5 (A = K, Rb, Cs) provides a new platform for exploring novel many-body quantum phenomena. In CsV3Sb5, a stripe-like CDW with commensurate wave vector q = 3/8 was observed under moderate pressures, which leads to a peculiar superconducting double-dome behavior in pressure-dependent phase diagram. Previous density functional theory (DFT) calculations indicate that the pressure-induced stripe-like CDW is beyond conventional phonon softening scenario, suggesting a nontrivial role of electronic correlations. However, an in-depth understanding for the pressure-induced unconventional CDW remains elusive. Here, we performed pressure-dependent 51V nuclear magnetic resonance (NMR) measurements on KV3Sb5 and RbV3Sb5. Although the superconducting double-dome behavior is absent in pressurized KV3Sb5 and RbV3Sb5, a pressure-induced CDW phase, ascribed to a possible incommensurate triple-Q CDW, is identified by NMR spectra in both materials, indicating that the pressure-induced unconventional CDW beyond DFT calculations is a common feature for kagome metal AV3Sb5. In contrast to the stripe-like CDW, the pressure-induced incommensurate triple-Q CDW does not strongly suppress the superconducting temperature (Tc) but coincide with an almost plateau behavior at intermediate pressure regime in the pressure-dependent superconducting phase diagram. Furthermore, by systematically analyzing the Korringa relation between Knight shift and nuclear spin-lattice relaxation rate in AV3Sb5, van Hove singularities (vHSs) driven electronic fluctuations are revealed as an effective knob for the pressure-induced unconventional CDW. Finally, our present findings underscore the pressure-induced unconventional CDW as a novel correlated quantum state in kagome metal AV3Sb5.

LLM-based machine translation evaluation can closely match human judgments, but in practice it remains largely diagnostic, with the signals rarely translating into direct quality improvements under real production constraints. We propose a two-stage, evaluator-guided automatic post-editing framework that turns MQM-style evaluation into targeted repairs: a retrieval-augmented LLM evaluator outputs structured, span-level MQM diagnoses under an explicit edit contract, and a separate LLM post-editor applies minimal edits restricted to those diagnoses. This separation improves controllability and reduces paraphrastic drift compared to one-stage "judge-and-refine" baselines. In a systematic study involving seven LLMs spanning three model providers and seven languages, our best configuration consistently improves both COMET-22 and COMETKiwi scores over one-stage post-edit methods, while the evaluator's error spans and severities show strong agreement with human MQM annotations and human editor preferences.

LLM agents are increasingly used for security tasks: vulnerability discovery, exploit reproduction, and patch generation. Improving them at the model level demands expert demonstrations or computable rewards, which security tasks rarely offer: traces are costly, failures hard to diagnose, rewards sparse, and non-computable. Efforts thus shift to the harness and context, but manual tuning needs task-specific expertise and scales poorly, while automated methods rely on scarce ground truth, stronger optimizer models, or unguided propose-and-evaluate loops that reduce to costly trial and error. We introduce SelfOp, an algorithm that automatically improves a frozen security agent's task context (instructions, skills, and reference documents), without modifying its execution harness and model weights. SelfOp casts context optimization as chain-rule-inspired textual gradient descent: from a single instance's outcome, it propagates error signals backward through the evaluator, the agent's trajectory, and the context artifacts that shaped its behavior, yielding per-instance textual gradients. Gradients are accumulated across instances by clustering, ranking, and filtering, and committed only under cross-instance consensus. A convergence detector monitors the gradient signal itself and stops once the context has absorbed the generalizable information in the training data, without held-out validation data. We evaluate SelfOp on CyberGym, a benchmark of real-world vulnerability reproduction tasks. With fewer than 200 training examples, SelfOp yields a 17-point self-improvement for GPT-5.4-mini (with Codex), enough to surpass the frontier GPT-5.4 baseline by 6 points, and an 18.5-point self-improvement for GPT-5.4 itself. The optimized skills also transfer across models, highlighting that SelfOp-optimized skills learn generalizable task knowledge not model-specific patterns.

We propose a quantum reservoir computing framework based on a hybrid cavity magnomechanical system in the linearized Gaussian regime. The reservoir combines microwave-cavity, magnon, and mechanical degrees of freedom, and is extended by an auxiliary cavity acting as an input port, with time-dependent signals encoded in its detuning. The covariance matrix of the quadrature fluctuations provides the features for a trained linear readout. Using linear-memory, nonlinearmemory, and parity-check benchmarks, we find strong temporal memory together with more limited nonlinear processing, whose balance is controlled by the reservoir evolution time, and we show that intermode correlations substantially enhance the information accessible to the readout. The same architecture reconstructs a time-dependent signal encoded in the auxiliary-cavity detuning, with an accuracy governed by the interplay between the internal couplings and the encoding strength, and robust against Gaussian detuning noise. Accounting for finite measurement statistics reveals a trade-off between encoding strength and the precision of the covariance estimation, so that the optimal encoding depends on the available measurement budget. These results establish hybrid cavity magnomechanical systems as a promising platform for continuous-variable quantum reservoir computing with potential applications in signal probing.

Undergraduate programmes in intelligent medical engineering are expanding, yet laboratory curricula lag behind the multimodal, long-tailed, and distributionally shifting realities of clinical AI. This design paper presents an advanced experimental teaching system that translates an ongoing multimodal deep learning research project on endometrial carcinoma into a structured undergraduate lab sequence. We identify three educational gaps (modality, authenticity, and deployment) and derive four pedagogical principles from constructive alignment, experiential learning, the research teaching nexus, and the CDIO framework. The curriculum comprises four progressive tiers plus an engineering layer, with 32 laboratory units over 64 contact hours, delivered via a custom virtual clinical workstation using de-identified multi-institutional data. Each tier maps to a specific technical bottleneck, prerequisite coursework, and criterion-referenced deliverables. Data governance, safety, and assessment protocols are specified. Learning outcome data will be collected across two implementation cycles.

Image-to-layer decomposition converts a flattened image into editable RGBA layers, enabling element-level editing in design workflows. Existing diffusion-based systems typically adapt large pretrained text-to-image (T2I) models and introduce RGBA autoencoders or variable-layer architectural modules. We revisit this design choice and ask whether layer decomposition truly requires these heavyweight components. We introduce PixelART, a pixel-space rectified-flow Transformer trained from scratch for image-to-layer (I2L) decomposition. PixelART directly denoises regional RGBA pixel patches using a single-stream multi-modal diffusion Transformer, avoiding RGBA-VAEs, pretrained T2I backbones, and layer-specific decoders. We identify a key property of the task: high-noise timesteps determine layer assignment and coarse layer organization, while low-noise timesteps mainly refine color, alpha, texture, and boundaries. Based on this observation, we propose a terminal-boosted timestep sampling strategy to increase training coverage in the high-noise layer assignment regime. Trained on 4M multi-layer design templates, PixelART achieves state-of-the-art layer decomposition and composite reconstruction on Design-Multi-Layer-Bench and LICA with over 80% fewer parameters, 98% lower latency, and 85% lower memory than the recent Qwen-Image-Layered model. Ablation experiments show that pixel-space $\mathbf{x}$-prediction, terminal-boosted timestep sampling, and data/model scaling are critical, while T2I pretraining brings marginal benefits to the I2L task.

Video foundation models now reach human-level accuracy on physical-reasoning benchmarks, yet such tasks require predicting unobserved physical outcomes. Do these models perform human-like forward simulation, or do they exploit statistical regularities in visible scenes? Accuracy alone cannot distinguish these strategies. We introduce a distributional evaluation framework that treats model seeds and human raters as populations, enabling comparison of consensus, uncertainty, and strategy. On the Physion benchmark, we evaluate three ViT-L architectures (V-JEPA2, VideoMAEv2, DINOv2). V-JEPA2 narrows the accuracy gap to ~1 percentage point (73.2% vs. 74.2%), yet model-human disagreement reaches 26.4%, far exceeding human-human disagreement (4.8%), with substantially lower agreement (kappa ~ 0.48 vs. 0.91). The divergence follows forward-simulation demands: models outperform humans on geometric reasoning (linking, +11.8 pp) but underperform on gravitational dynamics (rolling, -11.8 pp) and causal chains (dominoes, -10.5 pp). Strategy fingerprinting confirms all three architectures share non-human strategies while none aligns with humans. Attribution analysis suggests that unobservable outcome features, rather than visible scene properties, predict this divergence, consistent with models relying more on scene-level statistical regularities than on explicit forward simulation, a systematic divergence that accuracy alone cannot reveal. Code is available at https://github.com/fanhong-li/model-human-divergence.

We present the catalogue of quasi-periodic pulsations (QPPs) detected in hard X-ray (HXR) observations of X-class solar flares obtained with the HEL1OS instrument onboard Aditya-L1 during July 2024 - March 2026. The catalogue comprises 34 flares spanning GOES classes X1.1 - X7.1. QPPs are detected and studied using a uniform procedure based on Ensemble Empirical Mode Decomposition (EEMD) and wavelet analyses. Statistically significant QPPs are identified in 25 events (~74%), demonstrating that QPPs are a common property of X-class flares. The detected periods range from approximately 1 to 3 min, with a pronounced concentration between 1.3 and 1.7 min, suggesting the existence of a preferential timescale that may be associated with kink or slow magnetohydrodynamic oscillations in the flaring region. The dominant periods remain nearly independent of photon energy across the 8 - 20 keV range, while the modulation depth increases systematically with photon energy following a power-law relation with an exponent of about 1.5. QPPs detected in different HXR energy channels exhibit strong phase coherence and positive cross-correlation coefficients, indicating modulation by a common physical mechanism. Several intense flares display HXR modulation depths of up to 25%, providing evidence for nonlinear effects. The observed statistical properties support models in which QPPs are generated by repetitive magnetic reconnection, either spontaneous or periodically modulated by magnetohydrodynamic oscillations. This study establishes HEL1OS as a powerful instrument for quantitative studies of flare QPPs during Solar Cycle 25.

There are renewed efforts to build a lunar base and house a team of astronauts on the Moon for extended periods. However, important challenges remain, such as a rapid-response fire-hazard management system. In the event of a fire or toxic leak inside an early crewed lunar habitat, a single pressurized cylinder must be handled within minutes. However, suited astronauts and radio-delayed support from Earth cannot guarantee such response times. We describe a prototype fire-fighting and emergency response safety technology that embeds intelligence in the habitat itself: using thin ceramic QR codes affixed every few meters, each carries a local floor map, safe sensor limits, and step-by-step hazard responses. A low-power rover decodes a plate in a matter of seconds, polls the adjoining temperature-gas sensor cluster, and acts immediately, eliminating the need for a global map . In a representative layout, the interior is divided into roughly one code per two square meters. Simulations show that rapid-response firefighting both simplifies the task and avoids costly resource use and indeterminate outcomes. Because knowledge is spread across passive plates, damage is localized, procedures can be updated by replacing a single code, and processor demands remain minimal. However, important challenges remain, such as a rapid-response fire-hazard management system. In the event of a fire or toxic leak inside an early crewed lunar habitat, a single pressurized cylinder must be handled within minutes. However, suited astronauts and radio-delayed support from Earth cannot guarantee such response times. We describe a prototype fire-fighting and emergency response safety technology that embeds intelligence in the habitat itself: using thin ceramic QR codes affixed every few meters, each carries a local floor map, safe sensor limits, and step-by-step hazard responses.