Papers

We ask how an expert panel should disclose auditable conflicts of interest when reports are named and experts care about their reputations for ability. Before sealed binary assessments are filed, an auditor commits to disclosing no orientation information, each named expert's orientation, or only the number of experts oriented in each direction; disclosure occurs after reports are locked. Holding reporting behaviour fixed, composition disclosure and individual attribution are exactly Blackwell-equivalent for learning the state at every finite panel size. Conditional on named reports remaining public, composition is minimal among state-lossless disclosures of orientations: every lossless public message must make the composition recoverable together with the reports. Incentives nevertheless differ sharply. In the large-panel limit near pooling, the local career returns to following evidence against one's orientation under attribution, secrecy, and composition disclosure stand in the ratio 1:4:5. On an open set of parameters, the two extreme regimes have only pooling within symmetric cutoff strategies, while composition disclosure admits a tremble-robust informative equilibrium. Near the lower boundary of this region, a sequence of composition equilibria is sparse—the evidence-following rate is of order 1/N—yet induces a nondegenerate state experiment. Anonymous disclosure can therefore preserve the information available under attribution while creating stronger incentives to produce it.

All-atom structure predictors model diverse molecular interactions, but using their learned structural priors for binder design remains challenging. Here we present TorchCraft, a unified binder-design framework that optimizes sequence logits through a frozen all-atom predictor. Implemented in TorchFold, TorchCraft combines confidence, contact, geometric, and sequence-prior objectives within a shared optimization procedure for minibinders, framework-conditioned VHHs, cyclic peptides, and ligand-binding proteins. Using pretrained AlphaFold 3 weights, TorchCraft generated representative minibinders and VHHs with experimentally measured binding across four targets in each format, without post hoc sequence redesign. Computational benchmarks further demonstrated the framework's applicability to cyclic peptides and ligand-conditioned pocket design. TorchCraft extends predictor inversion to multiple binder formats and molecular contexts, providing a common framework for reusing all-atom structural priors in design.

Convergence of the transition density of the Euler-Maruyama scheme for a Brownian-driven McKean-Vlasov stochastic differential equation is established, the coefficients being uniformly elliptic, spatially smooth, and Lipschitz with respect to the law in the 2-Wasserstein distance. The density p_n of the n-step scheme is shown to converge to the density p of the McKean-Vlasov equation in a Gaussian-weighted supremum norm at rate O(n^{-1/2}), that is, of order one half in the step size, and hence also uniformly in space. The proof develops a discrete parametrix (Levi-Hadamard) expansion for the Euler chain and compares it with the continuous parametrix of the limiting equation along a hierarchy of measure-flow, time-discretisation and lattice corrections; the discrepancy between the discrete and the continuous measure flows is of order O(n^{-1/2}) in the 2-Wasserstein distance and does not degrade the final rate.

Multivariate ocean forecasting must exploit shared evolution in a coupled ocean system while adapting to the heterogeneous statistical and dynamical characteristics of different prediction variables and locations. Fully shared models may lack the flexibility to handle this heterogeneity, whereas fully independent models discard the common ocean context shared across variables. The key question is how to retain shared context in a unified model while allowing computation to specialize according to the prediction target and local state. We propose OceanMoE, a structured conditional sparse Mixture-of-Experts framework that combines sharing and specialization for multivariate ocean forecasting. OceanMoE fuses cross-variable information to construct target-specific local representations and uses them to perform content-conditioned sparse routing at each spatial location, with the number of active experts adapted to router confidence. In the decoder, routing is augmented with a learned geographic bias parameterized by spherical-harmonic spatial bases, while shared residual and seasonal pathways provide common cross-variable and month-dependent context. Experiments on long-horizon autoregressive ORAS5 forecasting show that OceanMoE lowers aggregate forecasting error in both evaluated settings and maintains lower geometric-mean normalized RMSE than the corresponding baselines over most later rollout months. Routing analyses further show that expert allocation varies with prediction targets and spatial locations. These results support structured conditional computation as a modeling strategy for balancing shared ocean context with adaptive specialization.

True machine intelligence requires transcending passive pixel registration to master top-down functional reasoning over absent information via visual negation understanding. However, unconstrained visual negation paradigms remain overly open-ended, and pervasive affirmation bias causes both existing Multi-Modal Large Language Models (MLLMs) and evaluation metrics to fail under negative semantics. To solve these intertwined challenges systematically, we first anchor the boundaries of negation reasoning within specific cognitive goals. Specifically, by focusing on safety as a highly pragmatic and critical cognitive dimension, we define the task of Scene Negation Understanding under Safety Cognition (SNUS). Under this framework, we construct a high-fidelity negative caption dataset mapping dense assertions of localized hazards. Concurrently, we propose the Cognitive Expected Scene Graph (CESG) Score, a structure-grounded, polarity-aware evaluation metric. Extensive experiments demonstrate that while current models struggle on the task, traditional metrics completely collapse under semantic reversals. Conversely, our framework delivers a solid benchmark for SNUS, providing a rigorous foundation to advance risk-aware situational comprehension and counterfactual cognition.

Coronoid systems, nanotubes and toroidal polyhexes (or fullerenes) can all be regarded as carbon networks composed of carbon atoms linked in hexagonal shapes. The resonance graphs of coronoid systems and nanotubes are not necessarily connected. For coronoid systems and elementary nanotubes, by using flow across cuts the present authors gave criteria for two perfect matchings lying in the same connected component of the resonance graph (Discrete Appl. Math. 395 (2026) 443-455). However, the sufficiency of such criterion does not hold for general nanotubes and toroidal polyhexes. In this paper we strengthen this requirement to obtain valid criteria for two perfect matchings of a nanotube (resp. toroidal polyhex) to lie in the same connected component of its resonance graph: they have the same flows across cuts along the $x$-axis (resp. longitude and latitude) and the same ladders. For toroidal polyhexes, our method uses homotopic classes of simple loops on the torus, and the above criterion can be simplified by using only simple flows, for the case in which two perfect matchings have alternating hexagons.

Conversational context provides semantic and acoustic cues across turns for automatic speech recognition (ASR), but relying on historical transcripts can propagate recognition errors and discard pronunciation and speaker information. We present a multimodal conversational-context framework for LLM-based ASR that integrates a scenario-controlled data pipeline, scalable multimodal context training, and systematic evaluation. We construct dialogues around entities and their confusable forms and interleave historical user speech with assistant text responses for supervised fine-tuning. We also introduce MM-ContextASR Bench, which evaluates contextual understanding and entity error correction across five scenarios. Experiments with Qwen3-Omni and Step-Audio-2-mini reveal limitations in handling irrelevant and erroneous history and show that our data construction and training improve context utilization, with multimodal context achieving the highest overall entity recall on both models. Further experiments on accent, dialect, and target-speaker ASR demonstrate the value of historical speech. The benchmark data and evaluation code are publicly available at https://github.com/llh666521/MM-ContextASR.

This paper studies the composition of taxation when the government cannot fully commit to respecting private returns after investment. A fiscal authority must finance a given expenditure from labor and capital income. Ordinary tax receipts are protected, but an opportunistic executive can seize part of the remaining capital payment. We show that a revenue-neutral increase in the capital tax raises the probability of compliance whenever the labor tax is below its local, fixed-wage revenue peak, provided the equilibrium remains on a regular mixing branch. This result does not depend on the elasticity of substitution between capital and labor. The same reform can raise investment: the gain in expected retention must outweigh the decline in the opportunist's continuation gain as compliance becomes less informative. We also characterize the Ramsey allocation conditional on full compliance. Limited commitment then imposes a ceiling on sustainable capital payments, rather than a general lower bound on the statutory capital-tax rate. Finally, we distinguish a change in the authority's concern for the future from an increase in both actors' patience. The former favors more informative policies when continuation welfare is convex; the latter has no unconditional direction in the reduced-form policy problem.

Let $f_m(x_0,\ldots,x_{2m-2}) =\det(x_{i+j})_{0\leq i,j\leq m-1}$ be the determinant of the generic symmetric Hankel matrix of size $m$. We determine its global Bernstein–Sato polynomial. The proof combines the nearby cycles of the corresponding secant hypersurface with an explicit Archimedean zeta integral over the cone of positive definite Hankel matrices. A Gauss–Radau parametrization reduces this integral to a Morris integral and hence to a quotient of Gamma functions. Its primitive poles are detected by test functions supported arbitrarily close to the vertex of the cone. Lichtin's theorem, applied to a secant-stratum log resolution, fixes the integer shift between these poles and the roots of the Bernstein–Sato polynomial. A relative $D[s]$-lattice realization of nearby cycles, together with the simplicity of the corresponding intersection complexes, gives uniqueness and multiplicity one in every monodromy class. The induction in the matrix size follows from the local product structure of secant varieties.

For a graph $G$ and a set $B\subseteq V(G)$, the spread $\mathrm{sp}(B)$ of $B$ is the difference between the largest and the smallest degree in $G$ of a vertex of $B$, and for an integer $k\geq0$ the parameter $\mathrm{sp}(G,k)$ is the largest cardinality of a set $B$ with $\mathrm{sp}(B)\leq k$. Caro, Lauri and Zarb derived a lower bound for $\mathrm{sp}(G,k)$ and, among several families of graphs, considered \[ \mathrm{MOP}(n,k)=\min \{\mathrm{sp}(G,k):G\text{ is a maximal outerplanar graph of order }n\} \] and determined $\mathrm{MOP}(n,k)$ up to an additive constant for every $k\not =2,$ leaving the case $k=2$ open, with the bounds $4n/9\leq \mathrm{MOP}(n,2)\leq (5n+19)/11$. We first prove a lower bound on $\mathrm{sp}(G,k)$ for an arbitrary graph $G$ in terms of its order $n$, its number of edges $m$ and its minimum degree $\delta $. This lower bound contains the bounds of Caro, Lauri and Zarb and, for $k=0$, the bound $\mathrm{rep}(G)\geq \left\lceil n/(2d-2\delta +1)\right\rceil $ of Caro and West, where $d=2m/n$. We determine when this lower bound is attained, exhibit explicit graphs attaining it, and show that it is exact for all graphs once $n\geq n_{0}(\delta ,k,d)$. We then apply the bound to maximal outerplanar graphs: adjusting the count to this class we prove \[ \mathrm{MOP}(n,2)\geq \left\lceil \frac{4n+10}{9}\right\rceil \qquad \text{for every }n\geq 14, \] with equality for $n\equiv 2\ (\mathrm{mod}\ 18)$, and $\mathrm{MOP}(n,2)=4n/9+O(1)$ for every $n$.

Statistical inference is a crucial part of HEP analyses. Historically based on RooFit and RooStats, the statistical tools used by the experiments are now facing unprecedented challenges, such as the rapidly growing complexity of statistical models (involving hundreds of parameters of interest and thousands of nuisance parameters), the need for scalable performance in large likelihood minimizations, and the demand for interoperability across an increasingly diverse ecosystem of tools, computational hardware, and frameworks and libraries, especially the ones developed and used within the machine learning world. This contribution summarizes status and future plans for the statistical tools used by some of the main LHC experiments (CMS, ATLAS), with a focus on improvements coming from the ROOT world (RooFit automatic differentiation), interoperability with modern libraries (JAX) and communication across frameworks (HS3).

SabreAgent uses a language model at design time to construct two components for lost-sales inventory control: a product-specific seasonal prior and a validation-selected capped base-stock policy family. During operation, statistical forecasting and inventory optimization use these frozen artifacts to determine orders, with zero language-model calls. We evaluate the approach on the $1{,}320$ instances of InventoryBench. Under the benchmark's cost assumptions, the operations-research core draws on a zero-lead-time optimality result and a projected-inventory rule for positive deterministic lead times. The latter computes replenishment shortfalls by propagating inventory using sales along simulated demand paths. The seasonal prior adds forecast variants alongside the original forecaster, and the selected policy family handles stochastic lead times with order destruction. SabreAgent scores $0.6311$, compared with $0.5380$ for the strongest published baseline, and ranks first in all six benchmark cells. Ablations attribute most of the gain to the OR core. In the paired analysis, the seasonal component adds $1.79\%$ across the three real-data cells, and the search component adds $2.3\%$ across the two stochastic-lead-time cells. These results demonstrate how model-generated priors and policy structure can improve an OR controller through design-time use.

The rapid advancement of large language models and single-agent harnesses has reshaped the landscape of autonomous systems, raising a critical question of when multi-agent collaboration offers genuine value. As individual agent capabilities continue to scale, multi-agent collaboration faces diminishing returns while incurring growing context overhead. Through systematic analysis, we delineate the capability boundaries of multi-agent collaboration relative to single-agent alternatives, showing that it confers systematic benefits specifically in long-horizon tasks with sparse dependencies, while single-agent harnesses remain superior in tightly coupled, sequential workflows. Building on these insights, we propose SAIGE, a lightweight multi-agent collaboration mechanism based on Semantic-Aware Incremental Graph Evolution. SAIGE models collaboration as a dynamically evolving graph, where nodes are agent instances spawned on demand and edges encode semantic dependencies established through content-based information retrieval. Experiments on long-horizon, complex task benchmarks show that SAIGE achieves a favorable trade-off between context efficiency and task performance, and that scaling the agent pool or deepening the recursion level does not consistently improve outcomes. Our findings suggest that multi-agent superiority is bounded by task structure rather than universal, and that more agents do not necessarily make a system more intelligent.

Silent data corruption (SDC) from defective accelerators now interrupts large scale training, yet deployed mitigations act on whole nodes. Algorithm based fault tolerance (ABFT) for a single GEMM has to be fused into the kernel or encode the operands, and it localizes at most one error per checksum. We present FP-Sketch, a verifier that runs after an unmodified tensor core GEMM whose half precision operands are accumulated and delivered at FP32. A sum sketch detects corruption on every call. Hashed first moment sketches, confirmed by independent recomputation, then localize several corrupted entries with no false positives by construction, and each fault yields a coordinate and a magnitude for fleet diagnosis. In floating point, sketch noise rather than bucket collisions limits localization. We measure that noise and find that its constant depends on the BLAS and the operand format and that the bucket count must grow as $n^{2.57}$ for a square product. Sizing the bucket count by measured noise rather than by a fitted power of $n$ raises recovery on eight transformer shapes from 0.402 to 1.000, and measuring the noise at run time adapts the bucket count to the kernel and the model. Instruction level injection with NVBit shows that upsets in a live accumulator are often only 2 to 9% of a typical entry, a population that output side injection cannot produce. Output side injection recovers every fault, while under NVBit the same engine sized for faults of typical magnitude recovers 0.550, and sizing for the measured magnitudes restores 1.000. On Llama-2-7B, guarding the MLP down projections removes 99.4% (BF16) and 99.9% (FP16) of the perplexity damage caused by 2048 bit flips, and the clean path probe costs 0.78 to 3.06 ms against GEMMs of 0.35 to 12.47 ms.

Population III (Pop III) stars are hypothetical zero-metallicity stellar structures formed from primordial hydrogen and helium. They are theorized to span a wide mass range, extending to several 100 solar masses, and may have played important roles in nucleosynthesis and reionization. Their expected fluxes are far below JWST NIRCam detection limits, making direct observation unlikely. However, extreme magnification near the caustics of massive foreground galaxy clusters may enable their detection. We search for overlooked high-redshift (7 <= z <= 17) caustic transits in the lensing cluster MACS J0416.1-2403. Using three observations spanning 126 days, we create difference images to identify potential candidates. Critical curves for sources at 7 <= z <= 17, derived from a strong-lensing model, guided visual inspection of three difference-images. No additional transits were found. The longer caustics in our model sweep a larger source-plane area, increasing the probability of detecting an event. Combining this increased statistical sensitivity with deeper imaging, we establish a fainter limit for the unresolved stellar population at z >= 7. From this null result, we constrain the 100 solar mass 2 micron Pop III sky surface brightness to >= 32.8 +/- 0.6 mag arcsec^-2. Modeling the non-detection as a Poisson process gives a posterior mean caustic-transit rate of 0.29 cluster^-1 yr^-1 and a 95 percent upper credible limit of lambda_95 = 0.86 cluster^-1 yr^-1. The inferred rate remains consistent with the adopted fiducial Pop III caustic-transit model and provides an empirical benchmark for future multi-epoch monitoring campaigns. This search exploits the fact that individual stars projected close to source-plane caustics can be briefly magnified far beyond their unlensed fluxes. These constraints provide a direct test of the abundance of luminous Pop III stars at early cosmic times.

Carrier-envelope-phase (CEP) stabilized few-cycle pulses make the sub-cycle field structure a control parameter for laser-driven polarized electron sources, yet spin solvers for such pulses are rarely tested against exact results. We show that for an electron initially at rest in a linearly polarized plane-wave pulse the rest-frame polarization angle follows the instantaneous vector potential, $\Sigma = 2\arctan(a_x/2) + a_e a_x$ with $a_e$ the electron anomaly: the net rotation vanishes for every CEP, while the peak intra-pulse angle varies over the CEP by 3.02 degrees for a two-cycle and by 0.22 degrees for an eight-cycle pulse. The result follows from the Volkov orbit in longitudinal and transverse rapidities. We use it to validate a light-front reference integrator and a physics-informed neural network trained only on the light-front equations, which reaches $4\times10^{-5}$ in the spin sector and carries over to elliptical polarization, for which no closed-form solution is known. This gives validated tools for CEP-resolved spin dynamics in intense fields.

Accurate classification of brain disorders from neuroimaging data remains challenging because of substantial inter-subject heterogeneity and the complex multi-scale patterns present in functional connectivity and morphological representations. To address these challenges, we propose HyperAMS-Net, a deep learning framework for brain disorder classification using neuroimaging representations derived from resting-state functional MRI or structural MRI. HyperAMS-Net integrates adaptive multi-scale convolution, hypergraph attention, spatial-channel attention, and adaptive feature fusion. Specifically, adaptive multi-scale convolution learns data-driven weights over multiple receptive fields to capture complementary patterns at different scales. Hypergraph attention models higher-order dependencies among learned feature representations through node–hyperedge–node message passing, while spatial-channel attention enhances discriminative feature learning. Adaptive feature fusion further aggregates complementary information across parallel network branches. HyperAMS-Net is evaluated on three benchmark datasets spanning distinct brain disorders: ABIDE for autism spectrum disorder, REST-meta-MDD for major depressive disorder, and ADNI for Alzheimer's disease, using 5-fold stratified cross-validation. HyperAMS-Net achieves state-of-the-art performance across all evaluated datasets, attaining the highest accuracy and AUC among the compared methods. Ablation studies further demonstrate the contribution of each proposed component, with the largest performance degradation observed when hypergraph attention is removed.

LLM agents have recently shown promise in automating machine learning engineering by editing model and training code under execution feedback. Data, however, remains largely outside this agentic optimisation loop. We frame pre-training data selection as heuristic engineering over per-document features, i.e., lexical statistics, categorical labels, and perplexity. We introduce AutoData, an agent that searches directly over executable selection algorithms. Unlike prior data mixture methods that optimise weights over a fixed set of domains, AutoData searches a richer program space of scoring, stratification, and stochastic selection rules, discovering feature interactions automatically by iteratively refining algorithms with validation feedback from a proxy model. Within an overnight search, AutoData discovers a selection algorithm that outperforms existing human-designed curation pipelines. Despite being searched only on this small proxy, the discovered recipe transfers to larger scales and improves the downstream metric CORE. These results suggest that data engineering can be treated as an agentic machine learning problem, extending autonomous research from model and training-code optimization to the data.

The Analysis Grand Challenge (AGC) showcases an example of HEP analysis. Its reference implementation uses modern Python packages to realize the main steps, from data access to statistical model building and fitting. The packages used for data handling and processing (coffea, uproot, awkward-array) have recently undergone a series of performance optimizations. While not being part of the HEP Python (PyHEP) ecosystem, the Combine tool is a pillar of CMS analyses, covering more than 90% of the analyses published in the last few years. As such, it is necessary to have Combine integrated in the PyHEP ecosystem, using the AGC as example. This project also includes, in the long-term, providing support and integration for the High Energy Physics Statistics Serialization Standard (HS3), as a way to have a language-independent way of representing the likelihood and use different frameworks interchangeably. In these proceedings we cover part of the recent work performed on the AGC and Combine, including: performance benchmarks, covering benefits introduced by the recent improvements in the data processing packages; examples of how Combine can be integrated and run in a dedicated infrastructure (coffea-casa); and examples and plans to integrate HS3 in Combine.

This paper presents the physics basis for commercially relevant laser indirect-drive (LID) (radiation-driven) inertial fusion energy (IFE) using a 10 MJ laser driver. To date, this approach, proven at the National Ignition Facility (NIF), remains the first and only controlled fusion method to demonstrate the key physics required for fusion energy production, including a self-sustained burning plasma, substantially de-risking the path to commercial fusion energy. Building directly on these results, we present scaled designs to larger target sizes and fusion gains relevant for commercial power generation ($G\sim 26$–$43$). The designs remain close to experimentally demonstrated ignition physics, modifying target components to improve scalability, manufacturability, and cost-effectiveness for fusion energy applications while preserving ignition-relevant implosion physics and fusion power plant compatibility. The baseline platform retains a high-density carbon ablator and clean cryogenic DT fuel layering while extending ignition platforms to substantially larger fuel masses (exceeding 10 times that of current ignition experiments at the NIF) and higher burn fractions ($\sim$40%) with total areal densities at stagnation of $\sim$3~g/cm$^2$. Benchmarked simulations anchored to NIF ignition experiments (using HYDRA and LASNEX) predict that these designs achieve robust ignition and propagating burn at substantially higher fusion yields (265–427~MJ) with significant ignition margin (2–4$\times$ relative to NIF) against hydrodynamic instabilities and representative power-plant non-idealities, including low-mode asymmetry, polycrystalline DT ice roughness, HDC ablator voids, and target-support and fill-hole perturbations. We also show that implosion symmetry and laser-plasma interactions (LPI) can be controlled with our novel multi-beam configuration using thousands of laser beam-lines.

We prove a quenched functional central limit theorem for the random conductance model with ergodic translation-invariant, strictly positive nearest-neighbor conductances, assuming only finite first moments of the conductances and their inverses. This settles an open problem by reaching the critical first-moment threshold, which is sharp for a certain class of integrability assumptions. A key ingredient of the proof is a new Sobolev inequality that seems to be missing from the literature.

Particle-based physics simulations, including fluids, smoke, and granular media, are fundamental to visual realism in immersive VR and AR. With the growing adoption of social VR and digital twins, demand is increasing for shared experiences in which multiple users interact with the same simulation in real time. Realizing such experiences requires low-latency streaming of large-scale particle data from a server to each client, yet existing point cloud compression methods such as G-PCC (TMC13) and Draco assume static geometric structures; when applied to dynamic particle streaming, their encoding latency exceeds the frame period, failing to meet real-time delivery requirements. We propose DELUGE, a streaming compression architecture that exploits the temporal coherence and velocity predictability inherent in physics simulation particles, achieving sub-frame-latency encoding and decoding through three complementary techniques. Evaluation on dynamic point cloud datasets demonstrates that DELUGE achieves approximately $20\times$ faster decoding than G-PCC (TMC13) and $6\times$ faster encoding than Draco. We further build end-to-end client implementations for both web browsers and Apple Vision Pro, and confirm through a within-participants perceptual quality evaluation and a two-person collaborative task study on Vision Pro that the proposed method supports real-time collaborative experiences with hand-tracked fluid interaction.

We carry out a unified study of the low-lying $1S$-wave compact states and hadronic molecules composed of hidden charm pentaquarks $qqqc\bar{c}$ ($q=u,d$) within a semirelativistic potential quark model. Apart from the linear confinement and one-gluon exchange potential between quarks and/or antiquarks, one-boson exchange potential is also included for baryon and meson clusters within the pentaquark system. We also evaluate the fall-apart decays by combining the obtained spectra within the quark exchange model. It is found that the $P_c (4312)^+$, $P_c (4440)^+$, and $P_c (4457)^+$ observed by the LHCb Collaboration in 2019 can be well explained by the hadronic molecules of $[\Sigma_c\bar{D}]_{1/2^-}^{1/2}(4318)$, $[\Sigma_c\bar{D}^*]_{1/2^-}^{1/2}(4437)$, and $[\Sigma_c\bar{D}^*]_{3/2^-}^{1/2}(4458)$, respectively. Meanwhile, $P_c(4380)^+$ reported by LHCb in 2015 may be assigned as the molecule $[\Sigma_c^*\bar{D}]_{3/2^-}^{1/2}(4382)$, except that it turns to be a narrow state other than a broad one shown by the experimental data. Our study shows that the $\sigma$- and $\rho$-meson exchanges are crucial for the formation of $\Sigma_c^{(*)}\bar{D}^{(*)}$ bound states with isospin $I=1/2$. Depending on the potential strength of the $\sigma$ exchange, there may exist very shallow bound states of $\Lambda_c\bar{D}^{(*)}$ with isospin $I=1/2$ and $\Sigma_c^{(*)}\bar{D}^{(*)}$ with isospin $I=3/2$. Our study may provide useful information for further exploring the hidden-charm pentaquarks in future experiments.

In many application domains, such as student dropout, insurance fraud, loan approval, and machine failures, several labelled public datasets are available where (i) data is about the same type of objects but the set of actual underlying objects are disjoint; and (ii) the class labels are same; and (iii) the feature spaces of the datasets are largely distinct (heterogeneous), with a few shared features. We call such datasets as allied. A single classifier cannot be trained on both datasets together, and one classifier trained on one dataset cannot be tested on the other. In this paper, we propose a method to merge the feature-spaces into a single feature-space for a pair of given allied heterogeneous datasets. We then use a matrix completion method to create a unified dataset based on the merged feature-space. The hypothesis is that the merged representation facilitates the transfer of classification knowledge from one dataset to another. We conduct experiments on several pairs of allied, heterogeneous datasets and several classifiers to demonstrate that any classifier trained on the unified representation always outperforms classifiers separately trained on the constituent allied datasets on several pairs of allied datasets. This work provides an easy way to substantially improve classifier performance by unifying and using multiple allied datasets together.

Light field (LF) reconstruction from limited and noisy focal stack (FS) measurements is a highly ill-posed inverse problem. Although the LF-to-FS imaging geometry is fixed for a given optical setup, LF spatial-angular structure—including within-view spatial details, cross-view angular dependencies, and disparity across views—varies across scenes. Consequently, a fixed pre-trained prior may not optimally capture the spatial-angular structure of each test LF. We propose Structure-aware Test-time Adaptation for diffusion-based light field Reconstruction (STAR), the first test-time adaptation framework for reconstructing an LF from FS. For each test LF, STAR freezes a pre-trained diffusion prior and fits three lightweight adapters to the observed FS to jointly adapt the three components of the LF's spatial-angular structure. STAR outperforms existing state-of-the-art methods in both two- and three-focal-sheet settings, with shorter inference times than those with test-time parameter updates.

A splay tree is a self-adjusting binary search tree that allows access, insertion, and deletion to be performed in amortized $O(\log n)$ time, where $n$ is the number of stored elements. The sequential access theorem states that, when the elements of a splay tree are accessed in increasing order, the amortized cost per operation becomes a constant. In this paper, we show that the upper bound for this constant is at most $5.5$ by refining the existing analysis and introducing a new potential function. Furthermore, we complement our result by showing that there exists a splay tree for which the constant is lower-bounded by almost $4$.

Fine-grained visual perception in MLLMs is commonly improved by raising the resolution, but the added visual tokens inflate vision-encoding and language-model prefilling costs. We show that the two operations underlying fine-grained perception, localizing the region of interest (RoI) and recognizing its content, have different resolution requirements. In a controlled diagnostic, localization tolerates roughly 3 to 4 times stronger token compression than recognition, which motivates localizing from a coarse view and concentrating resolution on the selected evidence. Decoding coordinates with the MLLM can be trained end-to-end from answers, but costs a full model pass per query and depends on grounding ability. A lightweight proposal network distilled from the model's attention is fast, but inherits the noise of its attention targets. The RoI from the proposal network reaches the answer through a discrete region choice, so its faithfulness to the answer cannot supervise the network. We therefore optimize the proposal network with region-level reinforcement learning, which we call Vision-RL2. It treats coherent regions as actions, and a frozen MLLM reader scores each one by how its removal changes the answer likelihood. Complementary subtractive and additive objectives suppress distracting proposals and recover missing evidence, updating only the predictor without region annotations, response sampling, or reasoning trajectories. The refined proposal further enables a sparse encoding that magnifies evidence and excludes background tokens. Across six fine-grained benchmarks and four MLLM backbones, Vision-RL2 improves accuracy over the base model at every token budget and surpasses its largest-budget accuracy with about 4 times fewer visual tokens. Code is available at https://github.com/YuHengsss/VisionRL2 .

We prove that the integral Chow groups $CH_i$ and mixed Hodge degree $H^{2i, (i, i)}$ cohomology groups of a really full rank (RFR) sink-recurrent cluster variety vanish for $i > 0$. In particular this applies to braid varieties and open projected Richardson varieties in any Lie type. Our main technical tool is the construction of a stratification of any RFR sink-recurrent cluster variety $\mathcal{A}(\Sigma)$ over any field $k$ into (affine space times) RFR sink-recurrent cluster varieties of seeds with fewer mutable vertices than $\Sigma$, with which we show that RFR sink-recurrent cluster varieties are linear in the sense of Totaro. We also show that cycle class maps are isomorphisms onto the lowest-weight part of rational Borel-Moore homology for any mixed Tate variety over a number field. Finally, we use our results to deduce vanishing theorems for the Khovanov-Rozansky homology groups of closures of positive braids and generation properties for the Chow groups of flag Richardson, projected Richardson, and brick varieties: namely, that these groups are generated integrally by the fundamental classes of their closed Richardson, positroid, and braid strata, respectively.

Quantized neural network inference runs integer matrix multiplications on GPU tensor cores, and the INT32 accumulators inside those cores have neither parity nor ECC. A transient fault in this datapath returns a valid but wrong integer and raises no interrupt. Checksum based Algorithm Based Fault Tolerance (ABFT) can detect such silent data corruptions (SDCs), but its verdict is binary. It cannot identify the corrupted element or its magnitude, and unweighted row and column checksums are blind by construction to errors that cancel on both axes. We present SProbe, a trailing verification kernel that reads the output of an unmodified vendor GEMM. A randomized Freivalds gate with three independent evaluation points in a 61 bit prime field misses a nonzero error with probability at most $2^{-141}$. When the gate fires, per row power sum syndromes over three primes are decoded with the Reed Solomon chain of Berlekamp Massey, Chien search, and Forney, recovering the column and exact magnitude of up to four colliding errors per row. SProbe then repairs the accumulator in place or recomputes the GEMM. On an NVIDIA H100, SProbe detects every injected fault across seven fault classes and four matrix sizes, including constructed patterns that TR-ABFT never detects and patterns that a weighted grid code detects but cannot correct. The gate costs 49% of the cuBLASLt GEMM time at N=16384 and 11% at N=65536. In an INT8 medical LLM, protection eliminates all observed silent corruptions at a 30% throughput cost. Our measurements also show that recomputation is faster than in place recovery in every configuration we tested, that diagnosis rather than repair dominates recovery cost, and we report the defects we found while validating the verifier itself.

Autonomous Underwater Vehicles (AUVs) navigating without GPS typically fuse inertial measurements with acoustic Doppler Velocity Log (DVL) velocities and pressure-derived depth. Posing the navigation state on a Lie group improves accuracy and consistency. However, state-of-the-art filters based on the Invariant Extended Kalman Filter (IEKF) append the Inertial Measurement Unit (IMU) biases as a Euclidean extension, which breaks the group-affine structure required for exact log-linear error dynamics, causing the reported covariance to degrade alongside the estimate. We apply the Tangent-Group (TG) symmetry, which carries the biases within the geometry of the state space, to derive an Equivariant Filter (EqF) for this system, leaving zero linearization error in the navigation states and second-order error only in the biases. We develop an equivariant output model for the DVL, whose update incurs only third-order linearization error, together with a direct pressure output. Monte Carlo simulations benchmark the TG-EqF against a Two-Frame-Group IEKF and a Multiplicative EKF. The TG-EqF reduces error by 18–25% against both alternatives in each of attitude, velocity, and position. The main benefit is in the covariance it estimates: its Average Normalized Estimation Error Squared (ANEES) stays closer to its nominal value of one than that of the others. Offline analysis on AUV field data corroborates the findings of the simulations, demonstrating reduced position drift.