Machine Learning
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Monday 7 September
Activation steering promises modular control of language model behavior: a behavior such as politeness corresponds to a direction in a model's activations, and adding that direction while it generates should switch the behavior on and leave everything else alone. It does not. We ask what decides which other behaviors move, and by how much, and find that on average it points to the model rather than the behavior being steered. Steers mostly lower what the unsteered answer shows and raise a set of behaviors the model falls back on (on Qwen2.5-7B, for example, chiefly what the judge scores as refusal, sycophancy, and poeticism). Three results across 24 behaviors and ten instruction-tuned models support this, every effect read off the generated text by a language-model judge rather than off a probe. That readout matters: on Qwen2.5-7B all 24 behaviors are linearly decodable, but only 20 change what the model writes. First, on average, directions carrying no behavioral content, of a real steer's size and layer, move the same behaviors in a similar order as real steers do, clearly on five models, weakly on five. Second, much interference runs one way, so it cannot be an overlap between two directions: steering profanity makes the model toxic, while steering toxicity leaves profanity untouched. Third, with a behavior held out entirely, geometry measured on the others explains little of the interference it takes part in. The one-way structure and the failure of geometry hold on all ten models. Base models share the one-way structure, yet only one of four has refusal or sycophancy among the five behaviors steers induce most, so post-training appears to install them. Neither geometry nor random directions predicts one pair of behaviors well. For single-layer additive difference-of-means steering, this implies that disentangling behavior directions cannot by itself make steering modular.
Graphics
CT2Yarn: Yarn-Level Reconstruction of Crochet from Computed Tomography
We introduce CT2Yarn, a human-in-the-loop framework for recovering a single continuous yarn path from micro-computed tomography (micro-CT) scans of real crochet objects. Crochet is a craft that creates complex three-dimensional shapes by interlocking loops formed from a single yarn. Recovering the underlying yarn path from external observations is challenging because of severe self-occlusion. While micro-CT reveals the full internal structure of a crochet object, the volumetric scan alone does not explicitly encode how the yarn traverses the object. This challenge stems from the hierarchical structure of yarn: a yarn consists of multiple twisted plies, and each ply itself consists of twisted fibers. Consequently, local fiber orientations observed in micro-CT scans are not aligned with the overall yarn direction. To recover yarn-level orientations from ply-level fiber orientations, we first estimate local fiber directions using Gabor filtering and convert the volume into an oriented point cloud. We then introduce an anisotropic mean-shift procedure that aggregates local fiber orientations within a neighborhood of the yarn radius into yarn-level orientation estimates. Combined with an automatic topology skeletonization strategy, our method extracts yarn-path fragments. Subsequent fragment linking, junction cleaning, and loop detection merge these trees into a small number of long curves. Where the automatic reconstruction remains ambiguous, a sketch-based user interface enables users to interactively complete the single continuous yarn path. The recovered yarn path enables downstream applications including physically based simulation, ply-level rendering, and stitch-pattern extraction.
Read the full abstract
Materials Science
Quantum hot carrier spectra in plasmonic catalysis
Vibrational activation of admolecules on metal nanoparticles is an elementary step in plasmonic catalysis, yet the underlying dynamics driven by hot carriers is not fully understood in the quantum regime. Using an atomistic description of plasmonic hot carrier generation, we investigate vibrational excitation and dissociation of oxygen on silver nanoparticles as a function of diameter D. As D reduces from the classical to quantum-sized regime, quantized distribution of hot carriers emerges with increasing population in the high-energy regions. These highly energetic hot carriers deliver more efficient vibrational coupling and dissociation. The rate of vibrational excitation shows a linear 1/D scaling, which results from Landau damping. It turns nonlinear at elevated light intensities due to vibrational heating generated by multiple electron scattering. The finding of quantized distribution of hot carrier in plasmonic catalysis opens new avenues for selective control and nonthermal energy conversion.
Read the full abstract
Computer Vision
PRG-Fusion: Orchestrating Generative Priors with Reconstruction Evidence for Driving View Synthesis
Synthesizing photorealistic driving videos along specified trajectories is essential for scalable closed-loop simulation. However, reconstructed scenes exhibit spatially varying degradation under viewpoint shifts, while generative refinement can alter reliable content and introduce geometric inconsistencies. We propose \mbox{PRG-Fusion}, a framework for driving view synthesis that uses reconstruction evidence to orchestrate generative priors across regions. Our key insight is that reconstruction reliability and available support should jointly determine how previously unseen regions are generated and reliable content is preserved. We derive pixel-wise Preserve, Repair, and Generate (PRG) labels from reconstruction evidence to distinguish reliable appearance, degraded but supported content, and regions with insufficient support. At inference, these labels serve as a unified routing policy for region-aware spatiotemporal synthesis, orchestrating 3DGS appearance preservation, LiDAR-guided structural correction, and video-prior-driven content completion across Preserve, Repair, and Generate regions, respectively. We then follow a two-stage training paradigm, first establishing geometric control from sparse LiDAR projections and subsequently learning appearance control from dense 3DGS renderings. Extensive experiments on Waymo demonstrate that \mbox{PRG-Fusion} achieves state-of-the-art overall performance in novel trajectory video synthesis, with superior visual quality and geometric fidelity while maintaining competitive view consistency under large trajectory shifts. Our project page is available at https://pre-cici.github.io/PRG-Fusion/
Read the full abstract
Machine Learning
Particle Dynamics of Flow Matching and Classifier-Free Guidance from a Stagewise Geometry Perspective
Flow matching, together with classifier-free guidance (CFG), is widely used in generative modeling, yet much of the theoretical understanding remains distribution-wise. Since practical sampling follows individual trajectories, distribution-level guarantees alone do not fully capture how trajectories interact with the data geometry or how guidance reshapes it. To overcome this limitation, we establish a unified stagewise geometric theory of attraction and absorption for both continuous dynamics and explicit Euler discretization. Specifically, with $t\in[0,1]$ running from noise to data, we show that unconditional flow trajectories are successively attracted toward a neighborhood of the global mean, the data convex hull, and a neighborhood of a possibly nonconvex local cluster. Across these stages, the corresponding distance satisfies a common contraction estimate, yielding an ${O}(1-t)$ decay of the distance in the final stage. For CFG, the same structure persists with an extrapolated mean, an inflated conditional convex hull, and, near the target cluster, the restored local geometry of conditional flow matching. We further show that a general time schedule $a(t)$ replaces the $O(1-t)$ decay by $O(1-a(t))$. Together, these results provide a unified particle-level geometric account of flow matching and CFG across continuous and discrete sampling.
Read the full abstract
Optimization and Control
First-Order Optimization under Uniform Nondegeneracy: Geometry, Computation, and Information
Strongly convex minimization and its natural indefinite extension to strongly convex–strongly concave minimax problems combine quantitative control of curvature with a prescribed curvature orientation. We disentangle these two roles by retaining uniform nondegeneracy alone: curvature remains uniformly separated from zero but may have either sign, with no prescribed positive–negative splitting. Surprisingly, a large part of the familiar theory nevertheless re-emerges. We first derive an intrinsic formulation through first-order secant inequalities, making the gradient on $\mathbb{R}^d$ a global bi-Lipschitz homeomorphism and yielding a unique stationary point. We then pair signed Moreau envelopes to construct a smooth scalar merit that recovers the missing descent geometry at both zeroth and first order, and the resulting paired proximal descent method achieves global linear convergence with dimension-free first-order oracle complexity. Meanwhile, we show that this tractability can break down on restricted domains: merely assuming the existence of a stationary point in the domain may lead to the curse of dimensionality, even with access to an infinite-order oracle. To overcome this information barrier, we introduce certified feasibility, an observable localization condition that enables feasible continuation. Together, these results establish a first-order optimization theory under uniform nondegeneracy that spans geometry, computation, and information.
Read the full abstract
Statistical Methodology
Brier-UDAM for Paired Prediction-Rule Comparisons: Latent-Risk Interpretation, Observable Contrasts, and Inference
The Brier risk summarizes probability-prediction error but does not explain why two prediction rules differ. We develop Brier-UDAM, a framework for paired binary prediction-rule comparisons that separates the Brier-risk difference into mean-bias, dispersion, and Brier-scale alignment contributions. Although the observable contrast identity follows algebraically from Yates-type moment decompositions, the framework gives these terms a latent-risk interpretation and shows that paired contrasts can be identified from observed outcomes and paired predictions without estimating the conditional event-risk function. For two rules evaluated in the same target population, the exact identity $\Delta R = \Delta M + \Delta D + \Delta L$ holds, and the corresponding plug-in estimators reproduce the empirical paired Brier-risk difference exactly in finite samples. We derive influence-function-based inference for first-order regular contrasts and a projection-based confidence procedure for the mean-bias contrast when first-order inference degenerates. Simulations confirmed interpretable component behavior, recovery of the target contrasts, and generally satisfactory inference in regular settings; projection intervals retained coverage near degeneracy but were conservative. In a large public-data analysis of 30-day hospital readmission, models with similar held-out Brier risks showed substantially larger and opposing dispersion and alignment contributions, revealing cancellation that was hidden in the aggregate score. Brier-UDAM therefore provides a diagnostic account of paired Brier-risk differences, while calibration, discrimination, and information-based assessments continue to address their distinct performance questions.
Read the full abstract
Analysis of PDEs
Local Well-Posedness of the Unsteady Boundary Layer System for Dilatant Power-Law Fluids
We establish local-in-time existence and uniqueness of monotone classical solutions to the two-dimensional nonstationary boundary layer equations for dilatant power-law fluids in a periodic half-space, for every fixed exponent $n>1$. The result holds under Oleinik's monotonicity condition and suitable weighted Sobolev, algebraic decay, and compatibility assumptions. The main difficulties are the tangential derivative loss caused by the normal velocity and the degeneracy at infinity of the principal coefficient $n\omega^{n-1}$ in the nonlinear vorticity diffusion $\partial_y^2(\omega^n)$, where $\omega=\partial_y u$. We adapt the Masmoudi-Wong energy method and use a good unknown to control the tangential derivative loss. For the nonlinear diffusion, we separate the estimates near the boundary from the treatment of degeneracy in the far field. Nonlinear boundary reduction and localized trace estimates control the boundary terms, while weighted estimates exploit the compensating decay of vorticity derivatives to control the diffusion commutators. Together with tangential regularization and weighted maximum and minimum principles, these estimates yield bounds uniform in the regularization parameter and preserve weighted monotonicity. Under these assumptions, the result provides an affirmative local-in-time answer to the existence and uniqueness question in Open Problem~11 of Oleinik and Samokhin.
Read the full abstract
Materials Science
Metastable polar order and phase competition in Carpy-Galy LaTaO4
Carpy-Galy LaTaO$_4$ undergoes a structural sequence linking the antipolar monoclinic $P2_1/c$ ground state and the polar orthorhombic $Cmc2_1$ phase through an incommensurately modulated regime. First-principles calculations identify a metastable polar monoclinic $P2_1$ phase as a commensurate intermediate in the phase-transition landscape. As a common subgroup of $P2_1/c$ and $Cmc2_1$, $P2_1$ provides a symmetry-connected route between the established polymorphs. Mode-resolved energy surfaces for these symmetry-breaking pathways show that coupling between the respective primary order parameters and an isosymmetric relaxation stabilizes the $P2_1$ minimum along both pathways. Consequently, the $P2_1/c\rightarrow P2_1$ transformation has a finite energy barrier, while a restricted soft-mode model retains competing $P2_1$ and $Cmc2_1$ basins over an illustrative range of harmonic stiffness. Lattice-metric and neutron-diffraction comparisons indicate a shared displacement character between the calculated commensurate state and the experimental IC-o modulation, supporting the use of $P2_1$ as a commensurate structural reference for IC-o. Finally, composition-dependent energetics suggest chemical routes for tuning polar-antipolar competition.
Read the full abstract
Machine Learning
PCSDiff: Diffusion-Based Bias Correction and Super Resolution Toward Practical Operational Medium-Term Precipitation Forecast
Medium-range precipitation forecasts are impaired by persistent systematic biases, lead-time-dependent error accumulation, and coarse spatial resolution, restricting their reliability for flood-drought risk assessment. Existing AI correction techniques lack dedicated modeling for multi-day dynamic bias evolution and proper meteorological constraints, often generating over-smoothed rainfall structures, and cannot meet operational deployment demands. This work introduces PCSDiff, a cascaded task-decoupled diffusion framework targeting 10-day precipitation bias correction and downscaling. To jointly counteract temporal error drifts and reconstruct physically plausible local precipitation details, PCSDiff integrates the Precipitation Intensity-aware Multi-branch Decoder (PIMD) module for dynamic multi-day error mitigation using synoptic-temporal features, followed by a two-phase conditional diffusion super-resolution module to restore fine-scale precipitation patterns. Evaluated against CMA-CRA observations over China after global-data training, PCSDiff cuts RMSE by 16.1% and lifts ACC by 13.9% relative to raw ECMWF forecasts at 3-10-day lead times, and consistently outperforms mainstream deep-learning baselines on both general and extreme-precipitation metrics. Benefiting from a streaming inference pipeline, our method achieves low-latency rolling forecasting for practical meteorological operations.
Read the full abstract
Artificial Intelligence
When and Why LLM Causal Priors Help: Closed-Loop Prior Selection for Amortized Causal Inference
Causal effect estimation asks how an outcome would change under an intervention, and medicine, economics, and public policy all treat it as a foundational task. Prior-data fitted networks (PFNs) amortize the task: a model trained on large numbers of programmatically generated synthetic causal tasks reads a new problem's observational data into context and returns an interventional-effect estimate in a single forward pass. The capability of such models is largely determined by the synthetic training prior, which is currently designed by hand, a bottleneck acknowledged by both Do-PFN and CausalPFN. Large language models (LLMs) can now “draw” plausible causal graphs for a given domain, suggesting that LLM-distilled graphs could serve as prior material. Whether injecting such graphs helps at all, where any gain comes from, and when injection helps. Practice has so far relied on manual trial and error. We propose a closed-loop prior selection framework that casts prior injection as a budget-constrained optimization over a candidate prior pool. Candidates undergo cheap post-training and are scored by a composite metric dominated by real-domain generalization; the winner then receives full training and paired statistical validation. On a 7.34M-parameter Do-PFN, the framework's winner attains a formally significant $2.75\times$ gain on the primary evaluation domain, and its error falls below that of the uninjected official base. Generalization on an adjacent monitoring domain improves significantly, and no monitored capability degrades. Mechanism experiments show that the gain depends on the semantic content of the distilled graph rather than its structural diversity alone does not produce it (directional evidence). With this framework and this regularity in hand, the use of LLM causal priors stops being manual trial and error and becomes an empirically verifiable selection problem.
Read the full abstract
Distributed Computing
Unified AI Gateway: A Framework for Joint Model Routing and KV Cache Management
Large language model (LLM) inference increasingly spans models that differ in size, capability, price, and provider. This shift creates two costs for developers. One is the integration cost of choosing among and switching between many models. The other is the inference cost of rebuilding a KV cache when it is unavailable or incompatible with the selected model. We define and analyze the Unified AI Gateway as a system setting for an edge-deployed AI traffic hub. It coordinates model routing, KV cache management, and compute placement across end devices, edge resources, and cloud model services. At request time, the gateway jointly selects a target model, an execution site, and a KV cache action under task-quality, latency, cost, and resource constraints. In parallel, background cache-management actions optimize KV cache placement, replication, retrieval, and lifecycle decisions for subsequent requests. We synthesize existing evidence on KV cache reuse, compression, cross-model mapping, distributed storage, and transfer, and discuss the remaining challenges of integrating these capabilities into one system. Across eight typical workload profiles, our workload-level analytical simulation reports TTFT speedups of 1.25$\times$–13.28$\times$ and input-cost benefits of 1.20$\times$–6.16$\times$.
Read the full abstract
Dynamical Systems
Balanced truncation for bilinear time-delay systems with approximate Gramians
A balanced truncation (BT) procedure is presented to achieve a structure-preserving model order reduction (MOR) for bilinear time-delay systems in this paper. We attempt to define Gramians for bilinear time-delay systems based on Volterra series theory. The controllability and observability Gramians in the frequency domain are given explicitly, which align with the ones of bilinear systems and time-delay systems. Based on the derived Gramians, a BT procedure is provided to produce reduced order models with the same structure. We also propose a numerical quadrature rule based on the truncated Laguerrre expansion to give an accurate approximation to Gramians. The resulting low-rank structure of approximate Gramians benefits a lot the efficient execution of the whole MOR procedure. Two numerical examples are simulated to showcase the efficiency of our approach.
Read the full abstract
High Energy Physics (Phenomenology)
Electromagnetic structure of strange vector mesons in nuclear medium
We investigate the in-medium modifications of the charge (electric) $G_C^{*}(Q^2)$, magnetic $G_M^{*}(Q^2)$, and quadrupole $G_Q^{*}(Q^2)$ form factors of the positively charged vector meson $K^{*+}$ in symmetric nuclear matter at zero temperature within the Schwinger proper-time Nambu-Jona-Lasinio (NJL) model. In this framework, both the nuclear medium effects and the electromagnetic structure of the $K^{*+}$ meson are described consistently in the NJL model at the quark level. We find that the charge, magnetic, and quadrupole form factors are suppressed with increasing nuclear density, indicating substantial in-medium modifications of the strange vector meson's internal structure. We further obtain a charge radius of $r_{K^{*+}}^{*}=0.74~\mathrm{fm}$ at normal nuclear density, which is slightly smaller than the corresponding value for the $\rho^{+}$ meson, $r_{\rho^{+}}^{*}\simeq0.75~\mathrm{fm}$.
Read the full abstract
Geometric Topology
Bounded cohomology and optimal separating constant for representations
Let $\Sigma$ be a closed oriented surface of genus $>1$ and $M$ a complete hyperbolic 3-manifold with a marking $i:\Sigma\longrightarrow M$. We consider the case that $M$ has no parabolic cusps and at least one of the two ends is simply degenerate. For $\varGamma=\pi_1(\Sigma)$, let $\rho_M:\varGamma\longrightarrow \mathrm{PSL}_2(\mathbb{C})$ be the holonomy of $M$ and $\rho:\varGamma\longrightarrow \mathrm{PSL}_2(\mathbb{C})$ any representation in $\mathrm{PSL}_2(\mathbb{C})$. We will show that, if $\rho$ is discrete and non-faithful, then \[ \|[\mathrm{Vol}(\rho)]-[\mathrm{Vol}(\rho_M)]\|_\infty\geq \boldsymbol{v}_3 \] holds, where $[\mathrm{Vol}(\rho)]$ denotes the bounded fundamental class of $\rho$ in the bounded cohomology $H_b^3(\varGamma,\mathbb{R})$ of $\varGamma$ and $\boldsymbol{v}_3$ is the volume of a regular ideal 3-simplex in $\mathbb{H}^3$. As an application, we present a rigidity theorem for $\rho_M$ in the set of representations $\rho$ of $\varGamma$ in $\mathrm{PSL}_2(\mathbb{C})$ in terms of $[\mathrm{Vol}(\rho)]$. The rigidity theorem implies that $\boldsymbol{v}_3$ is the optimal separating constant.
Read the full abstract
Astrophysics of Galaxies
Catching Up with the Fastest Star in the Galaxy: A Two-Passage GRAVITY+ Campaign to Detect S301's Schwarzschild Precession
S301 is the fastest known star in the Galaxy. It orbits Sgr A* every 8.68 yr on an e = 0.9832 orbit that carries it within 136 Schwarzschild radii at 25,600 km/s (GRAVITY Collaboration 2026). An orbit like that must precess, by the same Schwarzschild effect that turns Mercury's perihelion, whatever the spin of the black hole. We ask whether an astrometry-only GRAVITY+ campaign covering the 2031.8 and 2040.5 periapsis passages can catch it. We validate a relativistic light-curve model, design a 131-epoch Fisher-optimal cadence with a periapsis floor, fit a seven-parameter orbit including the light-travel-time delay, and build a systematic error budget. With white noise at the 207 microarcseconds achieved on S301, the fit recovers the injected rate, 0.2307 deg/yr, as 0.2324 +/- 0.0023 deg/yr (+/-0.0012 at the forecast 100 microarcseconds). Real astrometric noise is not white. We add per-run calibration offsets and a wandering Sgr A* photocentre at GRAVITY's published amplitudes, marginalize over the reference zero point, mass and distance, add the bounded systematics in quadrature, and inflate by the excess scatter of real GRAVITY orbit fits. What survives is 0.0074 deg/yr, a 31 sigma detection. The Sun hides the first periapsis for 153 d, at a cost to the period and periapsis epoch but not to the precession rate. Radial velocities are out of reach at S301's magnitude; photometry only validates the light-curve model. The spin of Sgr A* could shift the rate by 5.17-5.74 per cent and cannot be bounded, so the campaign is a spin-agnostic test of general relativity, and a forecast until the star comes round.
Read the full abstract
Category Theory
Models for dagger $(\infty,1)$-categories II: Dagger complete Segal spaces and the Joyal–Tierney equivalence
Building on the dagger Bergner and dagger Joyal models constructed in Part I, we develop simplicial-space models for dagger $(\infty,1)$-categories. We first construct a homotopically constant resolution model and a model structure for unitarily complete dagger Segal spaces. A change-of-index adjunction compares the two presentations of dagger simplicial spaces. We then prove a dagger analogue of the Joyal–Tierney Quillen equivalence. Together with Part I, these results give a chain of Quillen equivalences between the dagger Bergner, dagger Joyal, and dagger Rezk models.
Read the full abstract
Machine Learning
The Geometry of Refusal: Why Post-Hoc Refusal Is Fragile and Pretraining-Time Refusal Persists
Post-hoc safety training (RLHF, DPO) is the dominant way to align large language models, yet jailbreaks (Zou et al., 2023), fine-tuning attacks (Qi et al., 2024), and activation-space edits (Arditi et al., 2024) keep recovering the behaviors it was meant to remove. We give this fragility one geometric explanation and follow it into pretraining. We measure the safety update $\Delta = W_{safe} - W_{base}$ against the curvature of the model's capabilities (the empirical Fisher of a capability loss). Across five model families, post-hoc safety lands in a suppression regime: $\Delta$ is nearly orthogonal to the capability directions, and its small in-subspace part concentrates on a few high-curvature ones. The update is thin but sharp, a refusal gate laid over intact capabilities rather than erasure of them. A kernel-immobility lemma explains why such an update can only mask a capability, not remove it, and a little benign fine-tuning switches the gate off: 100 benign examples cut the AdvBench refusal of Qwen-2.5-7B-Instruct and Llama-3-8B-Instruct by 35 to 38 pp. Following the account into pretraining, a pretraining-checkpoint sweep of OLMo-2-1B (Team OLMo et al., 2024) shows the features that refusal attaches to emerging in a sharp transition between 1B and 63B pretraining tokens. We then use the account constructively: models trained from scratch with safety co-training spread continuously across pretraining reach 87 to 98% AdvBench refusal that the same attack erodes by only 2 to 14 pp at every scale from 410M to 6.9B, at a small cost on short-answer capability probes; a windowed schedule of equal total safety weight installs no refusal. Persistence, not timing, decides whether refusal survives. It does not keep output safer: by a harmful-output classifier, the attack raises harmful completions by 8 to 12 pp after post-hoc safety and by 11 to 27 pp after co-training.
Read the full abstract
Algebraic Geometry
Cyclic Spectral Data
In this paper, we study cyclic spectral data on smooth varieties over an algebraically closed field in characteristic zero. We show that the Hitchin fiber of any nonzero cyclic spectral data is non-empty. We also construct sections of the Hitchin morphism for cyclic spectral data and tautological (flat) families of cyclic spectral covers.
Read the full abstract
Mathematical Physics
Commutative Factorization of Nonlinear Second-Order Differential Equations: Theory and Applications
This work presents a generalization of the commutative factorization framework for a broad class of second-order nonlinear ordinary differential equations. The main objective was to extend the commutative factorization procedure in order to use the Riccati-Bernoulli equation and the Bäcklund transformation. This provides a systematic route for constructing both particular and general solutions of the original second-order nonlinear equation. The theoretical development is illustrated through several nonlinear models of physical and mathematical interest, including equations arising in classical mechanics and nonlinear oscillatory systems. The results demonstrate that commutative factorization constitutes an effective analytical tool for solving nonlinear differential equations.
Read the full abstract
Computer Vision
CARDEA: Auditable Reasoning Grounded in Spatial Evidence for End-to-End Coronary Angiography Interpretation
Invasive coronary angiography (CAG) is the gold standard for diagnosing coronary artery disease, but interpretation varies substantially among observers. Existing AI systems can improve consistency but lack auditable decision processes and are limited in comprehensive open-ended assessment, undermining clinician trust and clinical adoption readiness. We developed CARDEA, a unified large vision-language model that serves as the inference core of a CAG pipeline. It was trained solely on public datasets and closed-ended tasks in three stages: visual feature alignment, a self-distilled Chain-of-Box (CoB) cold start, and reinforcement learning with verifiable rewards (RLVR) with a CoB reward encouraging bounding-box use in the reasoning trace. We assessed its two study-level diagnoses, dominance classification and complexity assessment, against a dedicated classifier and two interventional cardiologists. Report generation was excluded from training and evaluated zero-shot across stages on an external cohort using vessel-severity macro-$F_1$. CARDEA trailed the classifier on in-distribution dominance but drew level under domain shift (accuracy, 0.91 [95% confidence interval (CI), 0.86 to 0.95]) and was comparable to the cardiologists on complexity assessment (accuracy, 0.90 [CI, 0.82 to 0.97]). Only RLVR improved zero-shot report generation, raising its vessel-severity macro-$F_1$ (0.686 [CI, 0.664 to 0.707]) above the untuned base model (0.513) and over twice the always-normal floor (0.312). CARDEA runs an end-to-end CAG pipeline from raw multi-view videos through keyframe selection to study-level diagnosis while exposing auditable spatial evidence behind its conclusions. RLVR on verifiable closed-ended tasks surfaced open-ended reporting ability that supervised imitation did not. Clinical use requires prospective validation against expert cardiologists.
Read the full abstract
Robotics
Distributed Dexterous Manipulation with Spatially Conditioned Multi-Agent Transformers
Distributed Dexterous Manipulation (DDM) is a novel paradigm that presents significant control challenges due to high action-space redundancy, inter-robot cooperation, and dynamic object-robot interactions. This paper introduces a framework based on spatially conditioned Multi-Agent Transformers (MATs) to efficiently learn robust control policies for a DDM system grounded in an array of 64 soft delta robots arranged in an 8x8 grid. Our three core contributions are: (i) an MAT with adaptive layer norm for compute efficiency, (ii) spatial contrastive embeddings to ground transformer embeddings in the spatial configuration of the robots, and (iii) an MAT-based behavior cloning method fine-tuned using Soft Actor Critic. We also propose an action selection formulation to analyze the trade-off between task performance and the number of robots utilized. Our experiments show that MATs iteratively refine their actions through the stacked attention blocks. This further informs the benefit of spatial conditioning in transformers to learn DDM policies. We demonstrate long-horizon planar manipulation tasks with objects of various geometries in simulation and real-world. Finally, we show how action selection mitigates robot maintenance by reducing wear and tear due to inter-robot collisions while maintaining the ability to manipulate objects along various trajectories in the real-world, achieving an average error of ~1.5 cm, while using ~65% fewer robots.
Read the full abstract
Computer Vision
Sub-Pixel Affine Registration of Space Debris Images via the Radon Point Spread Function
Inter-frame affine misalignment caused by platform jitter and attitude adjustments poses a fundamental challenge for multi-frame analysis of point targets in optical surveillance. Conventional registration methods rely on spatial intensity correlations or distinctive image features, both of which are largely absent in low-signal-to-noise-ratio point target imagery. We introduce the Radon Point Spread Function (RPSF) to characterize point targets in the Radon-transformed domain, and derive a closed-form framework that jointly estimates inter-frame translation and rotation from as few as four scalar RPSF samples per frame pair. The method requires no iterative optimization, feature extraction or interpolation, which is suitable for resource-constrained onboard processing. Simulation results confirm sub-pixel translation accuracy and a mean rotation error of 0.2556{\deg} at 1{\deg} Radon angular resolution. Validation on five real space debris datasets including both ground-based and in-orbit observations yields a mean calibration error below 0.5 pixels, substantially exceeding the precision required for reliable multi-frame processing.
Read the full abstract
Computer Vision
Emo-DVS: A Multimodal Benchmark for Privacy-Aware Emotion Recognition with Event Cameras
Emotion analysis is a fundamental task in computer vision, but its practical deployment remains constrained by the privacy risks inherent to conventional RGB cameras. Bio-inspired event cameras present a promising hardware-level solution because they capture asynchronous brightness changes, thereby reducing exposure of facial identity details while leveraging high dynamic range for robust perception under challenging illumination conditions. Despite these advantages, existing event-based methods struggle in complex real-world settings due to limited dataset scales, simple acquisition conditions, and reliance on single-modality visual cues. To address these, we establish a challenging tri-modal benchmark with event, audio, and text modalities and propose the Information-Guided Gated Fusion (IGF) framework, which first pre-trains an event encoder on the FAU subset of Emo-DVS to capture fine-grained facial dynamics, then employs adaptive modality gating to suppress modality-specific noise, and finally leverages mutual information maximization to align robust cross-modal representations. To alleviate data scarcity, we introduce Emo-DVS, the first large-scale event-based emotion analysis dataset, which couples dynamic illumination with the Facial Action Unit (FAU) subset and emotion subset. Extensive experiments demonstrate that IGF achieves state-of-the-art performance.
Read the full abstract
Mesoscale and Nanoscale Physics
Self-Trapping Enabled Highly Bright Momentum-Indirect Interlayer Excitons
Interlayer excitons in two dimensional material heterostructures exhibit large exciton binding energies and long lifetimes, making them ideal platforms for studying excitonic devices and many body quantum phenomena. However, the spatially separated electron and hole nature of IXs reduces their oscillator strength by two orders of magnitude compared to intralayer excitons. Achieving high efficiency IX emission remains challenging and requires optimal material selection with appropriate momentum matching and meticulous device fabrication. Here we demonstrate a highly bright momentum indirect IX emission within heterostructures formed between 2D perovskites and monolayer transition metal dichalcogenides. The quantum yield of IX emission reaches 35.2% on average, over 50 times higher than that of the corresponding constituent TMD monolayer, with the highest value exceeding 60%. Notably, the radiative recombination efficiency of this momentum indirect IX exceeds that of momentum direct IXs in monolayer TMD-based heterostructures by two orders of magnitude. We suggest that the remarkably bright IX emission in our heterostructure originates from IX self trapping, induced by strong exciton phonon coupling arising from the soft lattice nature of the 2D perovskite. Our findings provide new insights into achieving high IX emission efficiency and open new avenues for exploring long lifetime excitonic devices.
Read the full abstract
Cosmology and Nongalactic Astrophysics
Little Red Dot Cosmology: A Matter-Era Baryon Acoustic Oscillations Probe of $\Lambda$CDM
The discovery of Little Red Dots (LRDs) with JWST provides a new source population for cosmology in a largely unmapped period of cosmic history. LRDs are most numerous at $4\lesssim z\lesssim9$, thereby bridging existing low-redshift large-scale-structure measurements and the cosmic microwave background. In this epoch, the Universe is deep in the matter-dominated era, where dark energy is dynamically negligible and the expansion history is tightly predicted in $\Lambda$CDM, making these redshifts a clean test of the standard cosmological model. Recent measurements indicate that LRDs have number densities of $\bar n\sim10^{-4}$ $h^3$ Mpc$^{-3}$, a bias rising from $b\sim 3$ to $\sim8$ over this interval, and distinctive spectrophotometric signatures – a remarkably favorable combination for baryon acoustic oscillation (BAO) measurements. Using a Fisher forecast with these fiducial inputs, we find that a wide-field spectroscopic survey of LRDs over a DESI-like (14,000 deg$^2$) footprint delivers a percent-level isotropic distance scale measurement of $D_V/r_d$ across four bins spanning $4\lesssim z\lesssim9$. LRDs are therefore a compelling target for future wide-field near- to mid-infrared spectroscopy aimed at mapping large-scale structure deep into the matter-dominated era.
Read the full abstract
Data Structures and Algorithms
Frequency Moments Beyond Equality: Streaming Cosine Density Moments
For a stream of nonzero vectors $x_1,\ldots,x_n\in\mathbb{R}^d$, let $u_i=x_i/\|x_i\|_2$. We define the cosine density of the $i$-th stream element by $D_i:=\sum_{j\in[n]}\langle u_i,u_j\rangle$ and study the density moments $M_p:=\sum_{i\in[n]}D_i^p$ in both the signed- and nonnegative-cosine regimes. These quantities are similarity-aware analogues of classical frequency moments: replacing cosine similarity by equality (that is, $D_i = \sum_{j\in[n]} \mathbf{1}\{u_j = u_i\}$) gives $M_p=F_{p+1}$ and, in particular, $M_{-1}=F_0$, the number of distinct elements. We give one-pass streaming algorithms and lower bounds whose dependence on $d$ is tight up to logarithmic factors for $p=-1$, $p=1$, and even $p \ge 2$ (the latter under signed cosine), and whose exponents differ by less than $1/2$ for odd $p \ge 3$ under nonnegative cosine. Our results thus extend several fundamental statistics from the classical data stream literature to cosine similarity, a widely used measure for comparing vector embeddings in modern AI systems. The main challenge in proving a space lower bound for nonnegative cosine is to eliminate unwanted contributions without relying on pairs of opposite vectors. We address this through a construction that we call equal-sum moment isolation: two insertion-only prefixes have the same cardinality and vector sum, and a finite-difference comparison cancels the baseline shared by all their vectors while isolating the desired higher-order signal. This proof framework may be useful for other insertion-only streaming lower bounds, where direct cancellation is not possible.
Read the full abstract
Analysis of PDEs
Gradient estimates for the fractional $p$-Laplacian in the superquadratic regime
We establish gradient potential estimates for solutions obtained as limits of approximations (SOLA) to the fractional $p$-Laplace equation with finite signed Radon measure data. Under the assumptions $n\ge2$, $p>2$, $0<s<1$, and $sp>p-1$, every SOLA belongs to $W^{1,p-1}_{\mathrm{loc}}$, and its weak gradient satisfies a Wolff potential estimate at every Lebesgue point of the weak gradient. Under the additional condition $sp>n$, the solution has a continuous representative that is Fréchet differentiable at every point where the potential is finite. These results give a partial answer to questions raised by Diening and Nowak [Ann. PDE, 11(2025)] and by Diening, Kim, Lee and Nowak~[J. Eur. Math. Soc. (JEMS), 2025] concerning gradient potential estimates for the fractional $p$-Laplacian. The proof combines homogeneous affine decay with constants independent of the affine slope and comparison estimates to obtain an affine excess recurrence. Iteration then yields the Wolff potential bound.
Read the full abstract
Robotics
DriftParking: Trajectory Modeling via Drifting Field for End-to-End Automated Parking
Automated parking requires generating complete and executable trajectories in highly constrained spaces with low tolerance for goal pose error. Existing end-to-end parking methods struggle to jointly achieve inference efficiency, trajectory quality, and precise endpoint alignment, while conventional imitation objectives provide limited supervision on structured deviations from expert maneuver geometry. We propose DriftParking, a one-step trajectory generation framework that reconstructs the drifting-field paradigm for high-precision conditional trajectory generation. Specifically, we replace distribution-level attraction with conditional one-to-one attraction toward the paired expert trajectory, introduce expert-centered constructive repulsion, and adaptively attenuate repulsion near convergence. We further formulate trajectory generation in an endpoint-residual space by decomposing each trajectory into a start-to-goal baseline and a learnable residual, turning endpoint alignment into a representation-level structural constraint on the supervision target while providing a structured space for repulsive supervision. DriftParking achieves state-of-the-art performance across all evaluation metrics. Closed-loop on-vehicle experiments across diverse parking scenarios further show a 97% parking success rate, demonstrating strong zero-shot generalization.
Read the full abstract
Computer Vision
BEFORE THE FLIP: Measuring Hidden Score Shifts In Quantized Vision Language Models Before The Answer Changes for Visual Question Answering
Quantization makes vision language models (VLMs) cheaper to store and run by using fewer bits to represent their weights. While unchanged answers on visual question answering (VQA) after compression are an expected behavior, they can still hide changes in the underlying scores (log probabilities). For example, a model may still answer yes after compression, even as the score gap between yes and no shrinks. We introduce BEFORE THE FLIP to measure these hidden changes. Our method compares the score change caused by compression with the change caused by replacing the image's internal representations, or image tokens, with one fixed average token. We then increase the precision of one weight group at a time to identify where extra bits help, and test whether choosing different groups for each question offers benefits beyond shuffled controls. Among 8,277 LLaVA questions where image token replacement measurably affects the scores, 4-bit compression shifts the yes or no score gap farther toward the replacement output than 8-bit compression. Qwen shows the same pattern, but with a smaller difference. Yet only 265 of 9,000 LLaVA answers change at 4 bits. In a separate study of 1,024 calibration questions, choosing weight groups separately for each question does not outperform both shuffled controls at any tested storage budget. These findings show that compression can alter the scores behind unchanged answers, but do not establish a reliable benefit from adjusting precision for each question.