arXiv:2606.07383v4 Announce Type: replace
Abstract: Vision-Language-Action (VLA) models have shown strong potential for robotic manipulation, but real-time deployment on edge hardware remains challenging. In this work, we identify VLM visual and context tokens as a major source of deployment latency: for GEMM-dominated projection operators, computation grows linearly with the number of input tokens when model dimensions are fixed. Motivated by this observation, we propose RhinoVLA, a deployment-oriented VLA model co-designed with the Huixi R1 edge SoC. RhinoVLA adopts a token-efficient Qwen3-VL backbone and a continuous Action Expert, reducing the VLM-side token and computation burden while preserving pretrained multimodal capability. To support cross-robot learning, RhinoVLA further introduces a unified interface that combines View Registry, 72D physical state-action slot space, and robotinstance LoRA, allowing heterogeneous robot observations and action schemas to be aligned under a shared policy. On the deployment side, RhinoVLA is optimized through hardware-aware compilation, mixed-precision execution, and parallel visual encoding. Experiments show that RhinoVLA achieves downstream performance comparable to {\pi}0.5 at a similar parameter scale, while reaching 11.69 Hz end-to-end inference on Huixi R1, meeting the 10 Hz real-time closedloop control target. The project will be open-sourced at https://github.com/HuixiAI/RhinoVLA.
Science Journals
arXiv:2606.07636v2 Announce Type: replace
Abstract: Long-form video editing over heterogeneous footage requires agents to coordinate source selection, multimodal analysis, timeline construction, narration and subtitle alignment, rendering, and revision while exposing intermediate state for inspection and repair. We present Crayotter, an open-source multimodal multi-agent demo system for prompt-driven long-form video editing. Crayotter organizes production around coverage-aware material preparation, artifact-grounded editing research, and tool-grounded timeline execution. Across these stages, retrieval reports, video analyses, editing blueprints, scheduler events, tool calls, intermediate renders, and final exports are treated as first-class artifacts rather than hidden transient state. The workbench supports local assets, agent-assisted retrieval, progress monitoring, artifact preview, failure diagnosis, interrupted-job resumption, and resource-aware asynchronous execution for long-running workflows. In a 23-theme evaluation, Crayotter achieves the highest human overall score (3.40/5) among the compared systems, with its largest margins in theme alignment, narrative coherence, and editing smoothness. These results show that long-horizon video editing agents can be made traceable, inspectable, and practically controllable through observable production artifacts. Code, traces, and examples are publicly available at https://github.com/idwts/Crayotter.
arXiv:2607.15315v1 Announce Type: new
Abstract: Dependency resolution at a specified point in time in the past can provide insight into software evolution in software ecosystems and facilitate the design of dynamic metrics (e.g., dependency freshness, dependency update rhythm). However, dependency resolution at specified points in time is not possible in major software ecosystems due to a lack of support from package management tools. The goal of this paper is to aid practitioners and researchers in analyzing the state of the ecosystem dependency graph at release points using an enriched dataset with dependency resolution at release points for npm, PyPI, and crates.io packages. In this paper, we present a methodology to construct dependency resolution at release points of packages in software ecosystems, which we enrich with vulnerability data from OSV. We apply our methodology to construct CHRONO-RESOLUTION, a dataset of dependency resolution at release points for npm, PyPI, and crates.io packages, and release it for future research.
arXiv:2607.15613v1 Announce Type: new
Abstract: We study the problem of dividing homogeneous divisible goods among agents with non-linear valuations. Specifically, the value that an agent gains from a given good depends only on the amount of the good they receive, and is not necessarily linear with respect to the amount. For instance, under one-breakpoint piecewise-constant valuations, each agent specifies a threshold for each good such that this agent receives utility zero (resp., full utility of the good) when getting an amount below (resp., at least) the threshold. Given non-linear valuations that are additive across the goods, we focus on designing fair allocation algorithms and consider two well-known fairness properties: the maximin share (MMS) guarantee and envy-freeness (EF). For MMS, we devise an algorithm which always produces a $\frac{1}{2n-1}$-MMS allocation for $n$ agents with arbitrary non-decreasing valuations. It is worth noting that this algorithmic result is almost tight as we give an impossibility of guaranteeing more than $1/n$ approximation to MMS, even when agents have one-breakpoint piecewise-constant valuations. For $n \leq 3$ agents, we show the ratio $1/n$ is tight. Regarding envy-freeness, we show it is NP-hard to check the existence of an EF and Pareto optimal (PO) allocation for $n$ agents and at least three goods, even when agents have one-breakpoint piecewise-constant valuations. We complement the hardness result by considering the case with a single divisible good, and devising a polynomial-time algorithm to check whether an EF and PO allocation exists or not for agents with piecewise-linear valuations.
arXiv:2607.15291v1 Announce Type: new
Abstract: We develop a physics-informed neural network (PINN) framework for nonlocal partial differential equations arising in dynamic density functional theory (DDFT). Such equations are challenging for standard PINN methods because they involve nonlinearities, nonlocal interaction terms, and an underlying gradient-flow structure, often leading to slow convergence and difficult optimization. We adapt the PINN methodology to DDFT gradient-flow equations and introduce two computational components: a modified Lorentzian activation function that behaves approximately linearly for small inputs and decays toward zero as the input magnitude increases, and a precomputed discrete operator for evaluating the nonlocal convolution term efficiently during training. The method is tested on four examples in one and two space dimensions. In the first example, the exact stationary solution is known, while in the remaining cases the neural-network approximations are validated against reference solutions computed using continuous and discontinuous Galerkin finite element discretizations. Accuracy and physical consistency are assessed through $L^1$, $L^2$, and $L^\infty$ errors, together with mass conservation and free-energy dissipation. The results show that the proposed activation function accelerates convergence relative to the standard $\tanh$ function, while the overall framework maintains good agreement with the reference solutions and captures the expected gradient-flow behaviour. These findings demonstrate the potential of the proposed PINN framework for solving nonlocal gradient-flow equations arising in DDFT.
arXiv:2607.15780v1 Announce Type: new
Abstract: Repository-level documentation, such as READMEs, is often the first point of contact between users and a repository. When this documentation is incorrect, users may encounter runtime errors or waste their time debugging. We call such mistakes in repository-level documentation README bugs. Addressing README bugs is challenging because documentation mixes prose with code, its connection to the source of truth is loose, and finding a bug still leaves developers to craft a repair. This paper presents READU, an inconsistency-driven technique for just-in-time detection and repair of README bugs. The key insight behind READU is that README bugs often manifest as inconsistencies between documentation and another source of truth: either repository-internal facts, such as source code, or repository-external facts, such as external dependencies. READU applies a high-recall commit filter, runs internal and external consistency checkers in parallel, uses an alert judge to remove false positives, and automatically synthesizes documentation patches. On 6,000 recent commits from six popular repositories including Linux and Spring Boot, READU detects 244 true positives with 75% precision, while consuming less than $0.01 and less than one minute per commit, on average. Of these true positives, READU correctly repairs 217. We report 66 found README bugs, of which (so far) 44 are confirmed and 26 are fixed.
arXiv:2607.15293v1 Announce Type: new
Abstract: Dyadic and circular convolution can both be computed in $O(N\log N)$ time using the Hadamard transform and the FFT-computed discrete Fourier transform (DFT), respectively. The Hadamard transform is preferable for its real-valued sign flips, yet its substitution for the DFT introduces algebraic error. We present three complementary results that characterize this error. First, we identify exact error cancellation: two input and two output positions are universally error-free, and no reordering of the output can eliminate this error. Second, the error operator is nearly full rank, while its null space has only logarithmic dimension. Third, the expected error is governed by a single alignment scalar, with a closed-form expression obtained by averaging over random filters. In general, the substitution error asymptotically doubles the output energy, except for filters in the universal zero-error subspace, which incur no error. Collectively, these results show that the substitution error is structured, predictable, and governed by alignment.
arXiv:2607.15658v1 Announce Type: cross
Abstract: For a fixed $r$, let $f_r(n)$ denote the minimum number of complete $r$-partite $r$-uniform hypergraphs required to partition the edge set of the complete $r$-uniform hypergraph on $n$ vertices. The Graham-Pollak theorem states that $f_2(n)=n-1$. It was known that $f_r(n) \leq (1+o(1)){n \choose \lfloor{\frac{r}{2}}\rfloor}$, which was subsequently improved to $f_r(n)\le \left[ \frac{r}{2} \left(\frac{14}{15}\right)^{r/4} +o(1) \right] \binom{n}{\lfloor r/2\rfloor}$. Let $c_r$ be $\displaystyle \lim_{n \to \infty}\frac{f_r(n)}{\binom{n}{\lfloor r/2 \rfloor}}$.
It was known that $c_r<1$ for every even $r \geq 4$, while for odd $r$ the smallest known value satisfying $c_r<1$ was $113$. In this note we lower this to $85$ and also provide a constant-factor improvement in the known bounds for $f_r(n)$.
arXiv:2607.16067v1 Announce Type: new
Abstract: Chirality Chirality underpins a wide range of light-matter interactions, yet methods for its dynamic control in photonic systems remain limited. Here, we demonstrate ultrafast all-optical control of chirality in silicon metasurfaces through transient symmetry breaking. Our approach exploits photonic eigenstates of opposite spatial parity engineered to be highly susceptible to symmetry perturbations. Optical excitation generates free carriers that establish a transient refractive-index gradient across the membrane thickness, breaking out-of-plane mirror symmetry and facilitating hybridization of the parity-opposite modes into chiral photonic states. This enables the reversible creation and modulation of chirality on a 10-100 ps timescale, manifested by pronounced changes in the metasurface circular dichroism. By dynamically reconfiguring a fundamental symmetry property of the photonic structure rather than merely its optical response, our work establishes a route towards ultrafast control of chiral light-matter interactions and opens opportunities for active nanophotonic and information processing technologies.
arXiv:2607.15323v1 Announce Type: new
Abstract: An AC and DC hybrid microgrid, which inherits advantages of AC and DC microgrids and discards some disadvantages, is considered to be the most promising power network structure and gradually applied in the community. Usually, the AC subgrid and the DC subgrid are interconnected by Bidirectional Interlink Power Converters (BILPCs). Besides, in view of the different droop characteristics of AC subgrid and the DC subgrid, it is necessary to design a suitable distributed secondary controller for an AC and DC hybrid microgrid. Accordingly, this paper proposes a flexible and scalable distributed control framework for an AC and DC hybrid battery energy storage system (ADHB) with BILPCs in an Energy Internet (EI) paradigm. An ADHB governed by multi agent systems via a cloud server can reach the State-of-Charge balance, proportional power sharing, frequency and voltage restoration. The proposed control framework provides the group play-and-plug by adding or removing an inter-MASs interaction link. For a single BILPC in an ADHB, active/reactive power, frequency and voltage are adjusted by an AC BESS. For the parallel BILPCs in EI, a decentralized secondary control scheme is proposed. Communication delay issues and stability are analysed. Then, the relevant simulation results verify the correctness of the proposed scheme.
arXiv:2607.15416v1 Announce Type: new
Abstract: Reliable AI for screening mammography requires training data representative of the low cancer prevalence and subtle abnormalities found in screening populations. We examined whether supplementing such data with biopsy-confirmed cases from abnormal-enriched external datasets improves performance. Using the Newfoundland and Labrador Breast Screening Dataset (NLBSD) alongside CBIS-DDSM and CMMD, we evaluated an EfficientNet-B5 encoder initialized with Mammo-CLIP weights as a frozen linear probe under consistent preprocessing and patient-level splits.
The NLBSD-only model achieved an AUC-ROC of 0.737 (95% CI [0.686, 0.785]). Adding external positive cases reduced performance in every configuration (AUC-ROC = 0.620--0.644; DeLong test, Holm-corrected $p < 0.05$), with degradation increasing as additional sources were introduced. Domain-matched evaluation produced modest gains only when the training and test domains coincided, and no configuration surpassed the NLBSD-only model.
As a diagnostic, we reframed the task as predicting each examination's dataset of origin. The datasets were separated almost perfectly despite identical preprocessing, indicating that dataset-specific characteristics strongly influence the learned representation. These findings show that na\"ively pooling abnormal-enriched mammography datasets can introduce domain shift that outweighs the benefit of additional positive cases. Differences in acquisition, intensity mapping, and dataset construction persist after normalization, motivating domain-aware strategies for combining heterogeneous mammography datasets.
arXiv:2606.09421v3 Announce Type: replace
Abstract: Large language model agents increasingly rely on skills: reusable procedural documents encoding workflows, tool use, implementation patterns, validation checks, and domain rules. Skill rewriting is often treated as prompt compression, but shorter skills can make agents more expensive by removing sparse operational anchors that prevent exploration, debugging, and recovery. We study skill rewriting through this economic lens. Our controlled framework profiles skill structure, rewrites skills using information-preservation strategies, and evaluates the rewrites under fixed task instructions, environments, and verifiers. Experiments on SkillsBench reveal distinct quality--cost trade-offs across strategies: API/code anchoring, workflow guarding, and rule/formula anchoring benefit different task families, with no universally dominant template. In the main held-out evaluation, the learned policy reduces total cost by 7.0% and downstream agent-token cost by 6.0%; in frozen cross-model transfer, the corresponding reductions average 14.7% and 13.7%, while verifier quality is preserved. These results position skill design as cost-aware operational knowledge engineering rather than prompt compression. Resources: https://github.com/1Reminding/Skill_EE.
arXiv:2606.11042v4 Announce Type: replace
Abstract: Recent years have witnessed the rapid evolution of AI agents toward handling increasingly complex, real-world tasks. However, existing benchmarks rarely evaluate whether agents can operate graphical user interfaces to complete long-horizon, high-value professional workflows across diverse domains. Current GUI benchmarks still predominantly focus on general-purpose software, relatively simple applications, and short-horizon tasks, leaving it largely unknown whether modern agents can follow user instructions to autonomously operate domain-specific professional software and accomplish economically valuable work in an end-to-end manner. To bridge this gap, we introduce Workflow-GYM, a benchmark for long-horizon GUI tasks centered on professional domains and specialized software environments. Through extensive experiments on state-of-the-art models, we find that even the strongest models achieve only slightly above 30% success rates, highlighting that professional long-horizon GUI workflows remain highly challenging for current GUI agents. Further analysis reveals that current agents struggle to maintain long-horizon workflow consistency, frequently exhibiting workflow stage omission, error propagation, objective drift, and insufficient understanding of professional software environments. Our findings provide important insights into the limitations of current agent systems and suggest key directions for the next generation of GUI-agent research.
arXiv:2606.11781v3 Announce Type: replace
Abstract: We present a rotation-free magnetohydrodynamic dynamo driven by laminar thermal convection in a regular tetrahedral cavity. The tetrahedral boundaries organize the convective flow into a robust pattern of helical convection cells without global rotation or turbulence. Direct numerical simulations demonstrate exponential amplification of a weak seed magnetic field followed by nonlinear saturation, with the magnetic energy exceeding the kinetic energy. The velocity field develops $D_4$ dihedral symmetry, while the self-generated magnetic field exhibits a corresponding signed $D_4$ symmetry, including antisymmetry under $\pi$ rotations about the two horizontal axes. Analysis of the velocity and magnetic-field structures reveals a closed induction cycle sustained by geometry-induced helical convection. This system provides a conceptually simple setting for isolating and understanding the fundamental physical processes underlying magnetohydrodynamic dynamo action.
arXiv:2606.13669v3 Announce Type: replace
Abstract: Current LLM-based research agents have advanced through agent orchestration, yet largely overlook scientific knowledge orchestration. Existing works often reduce papers to abstracts, surface mentions, and flat \texttt{cites} edges, omitting key entities, claims, evidence, mechanisms, and method lineages essential for scientific reasoning. To this end, we introduce \textbf{Agents-K1}, an end-to-end knowledge orchestration pipeline that converts raw documents into agent-native scientific knowledge graphs. Agents-K1 integrates three components under a unifying theoretical foundation: a multimodal parser whose five-module schema captures entities, multimodal evidence, citations, and typed inter-entity relations across the full paper rather than abstracts alone; a 4B information-extraction backbone trained with GRPO under a rule-based reward; and a graphanything CLI, a tri-source agent interface that unifies web search, multimodal graph retrieval, and cross-document traversal. On top of this, we process 2.46 million scientific papers across six subjects to produce \textbf{Scholar-KG}, of which we release a one-million-paper subset, and the full Scholar-KG is accessible via the SCP link below. The same pipeline can be extended to general-domain corpora and to schema-conformant data synthesis. Extensive experiments demonstrate that Agents-K1 achieves superior performance in scientific information extraction, knowledge graph construction, and multi-hop scientific reasoning.
arXiv:2606.15254v2 Announce Type: replace
Abstract: Biofilms are spatially structured microbial communities whose architecture, chemistry, mechanics, and cellular states evolve over time. Bulk assays and two-dimensional projections remain useful, but cannot alone resolve how these properties vary with depth or change during growth, treatment, dispersal, and regrowth. Imaging provides complementary routes to three-dimensional measurement: fluorescence microscopy supplies molecular, taxonomic, and functional specificity; optical coherence tomography resolves mesoscale architecture and dynamics; quantitative phase imaging and holotomography report refractive index and biomass-related changes; Raman methods provide chemical and metabolic contrast; and Brillouin microscopy probes mechanical response. We compare these modalities using four independent descriptors-contrast provenance, live volumetric capability, perturbation, and demonstrated biofilm use-and connect their signals to quantitative biological readouts. No single modality simultaneously maximizes spatial coverage, resolution, acquisition speed, molecular specificity, and low perturbation. Implementations from any contrast class can serve as a longitudinal backbone when perturbation is empirically controlled at the relevant spatial and temporal scale, while molecularly specific measurements remain indispensable for identifying species, molecules, and functional states. We therefore frame four-dimensional biofilm measurement as a validated measurement architecture that integrates a low-perturbation volumetric backbone with spatially registered, molecularly specific measurements acquired continuously or at predefined validation points. Achieving this integration will require compatible cultivation formats, controlled imaging dose, shared quantitative parameters, and robust cross-modality registration.
arXiv:2606.16048v2 Announce Type: replace
Abstract: Reconstructing dense 3D scenes from sparse LiDAR point clouds (LiDAR scene completion) is a fundamental challenge in autonomous driving, where diffusion models offer a promising solution. However, existing approaches rely on object-level autoencoders that collapse into unstable global representations at outdoor scale, and suffer from ground truth data corrupted by odometry drift that systematically degrades supervision quality. Furthermore, multi-step diffusion inference incurs prohibitive latency for real-time deployment. We present CloudDiffusion, addressing these issues with three independent components. First, a multi-token Gaussian VAE with cross-attention pooling provides stable scene-scale LiDAR compression as a standalone reconstruction module, avoiding the global-pooling and codebook-collapse failure modes of prior point-cloud autoencoders. Second, an anchor-based ICP ground truth refinement pipeline eliminates drift-induced noise from training supervision, reducing our single-step x0 diffusion teacher's squared Chamfer distance by approximately 16x on SemanticKITTI seq. 08 (0.396 to 0.024 m^2) with no model change (partly aided by the denser, more compact refined references). Third, the same teacher completes scenes in a single x0 step, operating directly in coordinate space, not in the VAE latent. It runs in near real time at 209ms/frame, 65-138x lower inference latency than iterative diffusion baselines. Our results indicate that data quality dominates model design in this regime, and suggest that multi-token latent spaces could serve as a stable first stage for future latent diffusion-based scene completion.
arXiv:2606.16073v2 Announce Type: replace
Abstract: Sampling from complex, unnormalized probability densities is a fundamental challenge in Bayesian inference and probabilistic modeling. While Markov chain Monte Carlo (MCMC) methods provide asymptotic guarantees, they often suffer from slow mixing and high computational costs due to fixed or manually tuned trajectory lengths. In this work, we propose a novel framework that treats trajectory termination as a learnable component of the sampling dynamics. By framing MCMC within the theory of non-acyclic generative flow networks (GFlowNets), we train state-dependent neural classifiers to decide when a trajectory has reached a high-density region and should terminate. We theoretically establish the connection between optimal classifiers and the target density via detailed balance conditions and introduce a multilevel training scheme to facilitate exploration in complex geometries. Experimental results across various benchmark densities demonstrate that our approach significantly reduces average trajectory lengths while improving mode coverage and mixing compared to standard MCMC baselines.
arXiv:2607.15655v1 Announce Type: new
Abstract: Masked diffusion language models (DLMs) enable parallel text generation by iteratively refining masked tokens, offering a promising alternative to autoregressive decoding. Recent lookahead-based decoding methods improve the accuracy--efficiency trade-off by exploring future decoding states before committing token updates. However, existing approaches mainly rely on shallow one-step lookahead, which optimizes immediate information gain but can be suboptimal for longer-horizon decoding trajectories. Meanwhile, we find that a naive extension for deeper lookahead is also ineffective, as fixed-depth rollout introduces additional computation and cannot adapt to heterogeneous intermediate decoding states. Thus, in this work, we propose AdaLook, an adaptive lookahead framework for DLM decoding. AdaLook dynamically determines whether to continue rollout based on candidate-score variance and further enables branch expansion when intermediate rollout states require additional exploration. This design avoids unnecessary deep rollout while allowing the decoder to re-trigger lookahead from informative intermediate states. Experiments on various benchmarks and models demonstrate that AdaLook achieves a better accuracy--decoding steps trade-off than existing one-step lookahead decoding methods.
arXiv:2607.15652v1 Announce Type: new
Abstract: Bundle adjustment (BA) remains a critical refinement module for image-based 3D reconstruction and continues to improve geometric accuracy even in learning-based pipelines. However, in low-parallax and near-rotational regimes, classical Schur-based Levenberg--Marquardt (LM) often becomes ill-conditioned and yields unreliable pose and calibration estimates. We propose Gate-Guided CSS-BA, a solver-side modification of Schur-LM that preserves the classical BA objective and trust-region framework while constraining each update to a geometrically informed low-dimensional subspace. By integrating Column Space Search (CSS) with geometry-aware gating, the method stabilizes the Schur-LM update without altering the estimation problem. In contrast to keyframe or state-selection approaches, all camera and point parameters remain in the optimization problem; only the update direction is restricted. The method serves as a drop-in replacement for existing BA pipelines. Experiments on both generic and challenging weak-geometry scenarios show more stable optimization, improved relative pose accuracy, and competitive calibration behavior while maintaining reprojection quality.
arXiv:2606.16127v2 Announce Type: replace
Abstract: The worldwide rise of authoritarianism and the growing role of Large Language Models (LLMs) in users' everyday lives raise the question of whether specific models exhibit or promote authoritarian attitudes. We introduce AuAu, a comprehensive benchmark for assessing the risk of authoritarian tendencies in LLM responses. AuAu combines three evaluation approaches: (i) psychometric questions from 15 human-validated instruments, (ii) vignettes probing intended behavior in concrete situations, and (iii) responses to realistic user prompts. Unlike prior work, AuAu measures not only overall authoritarian alignment but also its established sub-concepts: Authoritarian Aggression, Authoritarian Submission, and Conventionalism. Evaluating 17 models from China, the EU, Russia, and the USA, we find substantial authoritarian response rates on psychometric instruments across all models, though rates drop significantly on more realistic downstream tasks. Moreover, a simple authoritarian system prompt manipulates 15 of 17 models into promoting increased authoritarianism. Our results underscore the need for continued, systematic auditing of LLM-based AI systems to detect and mitigate authoritarian tendencies in their outputs.
arXiv:2606.16624v2 Announce Type: replace
Abstract: Plate and shell structures are widely used in engineering fields. Rapid response prediction for such structures under complex geometries, heterogeneous materials, and varying loads is important for engineering design, but conventional numerical methods usually require repeated modeling and solution when the physical configuration changes. To address this issue, this study proposes a geometry-aware variational physics-informed neural operator (GA-VINO) for Mindlin-Reissner plates. GA-VINO represents the plate geometry using boundary point clouds and incorporates a material encoder, a load encoder, and a scalar-parameter branch to handle spatially random material fields, spatially varying pressure loads, and sample-level uniform parameters. Through multi-branch point cloud encoding and cross-attention, GA-VINO fuses geometric, material, loading, and query point information, and predicts the transverse deflection and rotations at arbitrary query locations. Unlike conventional data-driven neural operators, GA-VINO requires no labeled solution data during training. Instead, it minimizes a variational physics-informed loss constructed from the discretized total potential energy of the Mindlin-Reissner plate. Compared with grid-based neural operators, GA-VINO directly processes irregular point clouds and allows different physical fields to be discretized on different point sets, avoiding forced interpolation onto a common grid. The method is validated on multiple examples involving different geometries, material fields, and load distributions. The results show that GA-VINO achieves promising accuracy in deflection, rotation, gradient-sensitive, and energy-based metrics, completes full-field inference for new samples within milliseconds, and exhibits promising cross-geometry generalization capability.
arXiv:2606.16900v2 Announce Type: replace
Abstract: Physical systems often exhibit heterogeneous mechanisms, where rapidly evolving dynamics coexist with persistent structures. Capturing such multiscale physical behavior remains challenging for existing neural operators, which typically rely on single dominant inductive bias and therefore couple distinct physical responses into a shared representation. We introduce the Unified Green's Function Framework across domains and propose the Factorized Neural Operators (FaNO), which decompose spectral representations into equivariant-inspired dynamic responses and invariant-inspired persistent responses, leading to better interpretability and generalization. Mechanistically, we show that the two operator branches spontaneously specialize into distinct physical roles that remain consistent across scales and domains: the equivariant-inspired branch captures rapidly varying transient dynamics, whereas the invariant-inspired branch extracts coherent persistent structures. This factorized mechanism of FaNO consistently improves prediction accuracy, parameter efficiency and cross-scale generalization across physical systems and domains. In particular, it maintains consistent predictions under long-horizon autoregressive rollout, cross-resolution extrapolation and physical-regime shifts. These findings suggest that scalable physical modeling may benefit from moving beyond single-inductive-bias formulations toward factorized operator representations that better reflect the heterogeneous organization of physical systems, accelerating the reliable deployment of machine learning for scientific computing and discovery.
arXiv:2606.17522v2 Announce Type: replace
Abstract: Deep neural networks are widely believed to derive their expressive power from their ability to form \textbf{hierarchical representations}, capturing progressively more abstract and compositional features across layers. In language modeling, \textbf{transformers} have emerged as the dominant architecture, with early layers capturing local syntactic patterns and later layers encoding more complex clause-level dependencies. While this intuition has shaped model design, there remains a lack of rigorous theoretical work demonstrating \textbf{how} deep transformers represent such hierarchical structures. In this work, we analyze the expressiveness of deep transformer models through the formal lens of bounded-depth, non-recursive context-free grammars. For this class of grammars, we explicitly construct transformers with positional attention whose depth grows linearly with grammar depth, while the neuron count scales with the number of derivation-tree shapes and quadratically with the number of production rules. Our theoretical results support the linear representation hypothesis by demonstrating that these architectures possess the structural capacity to encode abstract grammatical states into low-dimensional, linearly separable subspaces within the residual stream.
arXiv:2607.15556v1 Announce Type: new
Abstract: 3D CT vision-language models (VLMs) classify abnormalities from text prompts in a zero-shot manner, enabling cross-institution deployment where labels are scarce and clinical tasks shift faster than supervised models can be retrained. A real CT scan, however, typically contains several co-occurring abnormalities, and the reliability of zero-shot multi-label prediction under distribution shift remains poorly understood. Test-time adaptation (TTA) updates a model on unlabeled target scans without source data or target annotations, yet existing TTA methods target multi-class softmax prediction on natural images or 2D medical segmentation, and none addresses unsupervised multi-label adaptation for zero-shot 3D CT VLMs. We study when TTA helps zero-shot 3D CT VLMs. A controlled diagnostic analysis shows that TTA is conditional: the volumetric input must preserve the encoder's depth structure, and the base representation must transfer to the target cohort, with depth reduction alone lowering internal AUROC by more than 0.12. We then focus on the regime where the base model already separates present from absent abnormalities. We introduce CARVE (Cardinality-Aware Retained-View Entropy), the first TTA method for this setting. CARVE estimates a sample-specific positive-label cardinality $\hat{k}$, optimizes a top-$\hat{k}$ objective to preserve co-occurring abnormalities, and performs memory-efficient multi-view adaptation by scoring weak 3D views without gradients before updating on a retained subset. Across contrastive CT-CLIP and anatomy-aware fVLM, CARVE provides the most consistent improvements across multi-label, three-class, and binary CT tasks when the base model is already discriminative. These results establish multi-label TTA for zero-shot 3D CT VLMs as a distinct problem and CARVE as a cardinality-aware solution.