Computation and Language
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Monday 9 November
Thai is written without spaces between words, making segmentation a foundational step in most Thai NLP pipelines. Existing tools primarily reproduce the word-level conventions of corpora such as BEST; however, because the notion of a "word" in Thai lacks a unified linguistic consensus, these conventions struggle with productive compounding, leading to arbitrary boundaries and ambiguity in downstream representations. Instead, we propose operating at the morphological level, focusing on khammoon, the smallest atomic unit that carries meaning. Segmenting at the morphemic level provides an explicit, consistent morphological representation that eliminates compounding ambiguity. Crucially, this representation benefits downstream learning: models trained on morphemes perform on par with or exceed word-level baselines across part-of-speech tagging, named-entity recognition, and sentiment classification, demonstrating that linguistic consistency improves representation quality without downstream performance penalty. To operationalize this representation, we translate linguistic principles into a formal decision guideline with near-perfect human agreement ($\kappa \approx 0.99$), scale annotation with an LLM, and correct systematic errors with human labels. Finally, we train khammoon, a lightweight 1M-parameter convolutional BiLSTM segmenter designed for efficient CPU inference, achieving a boundary F1 of 0.992 on out-of-domain text. The guideline, datasets, model, and library are all publicly released.
Software Engineering
Compile-Gated Lean Proof Refactoring with Enterprise Agent Practices: A Model per Step under a US\$3 Cap, Followed by Programmatic Rewrite Rules
Our closed-track entry (Arena username Vexorium-Agent) scored 86.31 points on the final board (L = 77.07, H = 81.86, Z = 100) over the 60 final problems, against 33.33 for the reference proofs, 8th at the deadline (top: 92.57), for US\$166.33 at list prices with every call counted (per problem: mean US\$2.772, at most US\$2.874; US\$3 cap). Three practices carry the entry. First, each paid stage runs the model that measured best per dollar for it among the 9 we tested in our own harness: GPT-6.1 Sol the early sessions, GPT-6.1 Terra the later ones and the tail at 70% of Sol's cost per turn, and Gemini 4 Argon the review and repair; Sol also runs the style rewrite. Second, what our development and research runs learned ships as compiler-checked programmatic rewrite rules rather than prompt text (+0.93 points from rules fixed before the release; +0.26 and +0.30, in sample, from passes built or switched on after the release). Third, every call is recorded and reviewed before each change (all 9 board scores came within -0.063 to +0.015 points of our meter's prediction). The harness, gate and many of these rules also serve our open-track entry. We report the full bill and the spread of three runs.
Read the full abstract
Computer Vision
Preserving Cultural Records: A Large-Scale Benchmark and Visual-Prior Routing Network for Ancient Book Image Restoration
Ancient books preserve invaluable cultural heritage, yet their images often suffer from complex degradations caused by long-term ageing, environmental exposure, and physical damage. Existing ancient book image restoration benchmarks are limited in scale and rely on simplified degradation simulations, which cannot fully represent real-world degradation patterns and restrict the generalization ability of restoration models. To address these challenges, we introduce AncBench, a large-scale synthetic paired benchmark containing 100K aligned clean-degraded image pairs. AncBench models realistic composite degradations through a comprehensive degradation pipeline that considers material ageing, paper texture variation, blur, staining, water-induced geometric distortion, and local occlusion. Furthermore, we propose AncRestore, a degradation-aware visual-prior routing network for ancient book image restoration. AncRestore first learns degradation representations from degraded images and then dynamically selects and integrates suitable visual priors according to different degradation characteristics. Guided by these degradation cues, a dynamic routing mechanism combines expert-wise and region-wise routing strategies adaptively adjusts the contribution of these priors across different regions, enabling the model to recover both global appearance and local text details. Extensive experiments show that AncRestore effectively suppresses composite degradations while preserving text structures and enhancing restoration quality against other methods, highlighting its application potential for intelligent ancient book image restoration in cultural heritage preservation.
Read the full abstract
Computer Vision
Instance-preserving graded semantic reporting for surgical action triplets
Complete-or-reject reporting of surgical action triplets can discard a correctly identified instrument and action when the target is uncertain. We introduce instance-preserving graded semantic reporting, which separates instance selection from the choice of semantic detail. Each selected instance retains its predicted mask and labels while hierarchical correctness scores guide instrument (I), instrument-verb (IV), or complete instrument-verb-target (IVT) reporting. The utility-increment rule DELTA weighs the expected gain from finer reporting against its error cost. We compare three decision rules on 5,641 test frames from eight surgical videos using fixed predictions and scores. At the 10% development calibration target, DELTA retains correct partial reports for 80.0% of candidates rejected by a matched complete-or-reject baseline (5,329/6,662). Adding IV increases source-normalized weighted correct-report yield by 19.13 to 23.13 per 100 frames across all three rules relative to separately recalibrated I/IVT controls. Weighted reporting risk changes by -0.77 to +1.01 percentage points, with complete correct-triplet counts unchanged or reduced. Graded reporting makes partial surgical predictions available as localized descriptions, with an explicit trade-off between report utility and completeness.
Read the full abstract
Hardware Architecture
An Open Implementation Flow for the Xilinx Spartan-3E and Spartan-6 that Runs in the Browser, Checked Byte for Byte Against ISE
The author describes an open implementation flow, from HDL to a programmable bitstream, for the Xilinx Spartan-3E (XC3S100E to XC3S1200E) and Spartan-6 (XC6SLX16, XC6SLX45), devices that only the discontinued ISE 14.7 supports. The flow runs unchanged in a web browser or in Node.js: Yosys and nextpnr compiled to WebAssembly, with nextpnr's Himbaechel framework given Spartan-3E and Spartan-6 architectures generated from Silinx's own device descriptions, bit database and timing model, and Silinx's own bitstream writer. The bit database was built by black-box observation of ISE's documented XDL and bitgen tools, using codeword-batched experiments; the router is given only switches whose bits were measured, and the writer refuses a feature it has not measured. ISE is the oracle: for the same placed and routed design, the bitstream is byte-identical to the one bitgen writes and ISE's DRC is clean. Placement and routing were first done by an in-house annealer and PathFinder router, which nextpnr replaced once it matched them. On 67 open-source designs for the XC3S1200E, compared with ISE at the same optimisation goal (45 designs built by both), the flow uses 15% fewer slices and has 14% shorter critical paths at goal Area, 21% fewer slices and 8% shorter paths at Speed, and 20% fewer slices but 12% longer paths at Fast (geometric means). XST cannot read 15 of the designs. The flow is published as the npm package silinx-fpga (Apache-2.0).
Read the full abstract
Quantum Physics
Dephasing-Induced Melting of Crystalline Spectral Form Factors
Long-range spectral order can be probed using spectral form factors, where it manifests itself as sharp Bragg peaks at integer multiples of the Heisenberg time in nearly crystalline spectra. We examine how energy dephasing in open systems affects this spectral order and how to control it. We find that Bragg peaks fade algebraically with revival order and dephasing strength, $K_{\rm Bragg}(p)\propto(\gamma p)^{-1/2}$, within an extensive intermediate regime, and persist up to the lifetime of nearest-neighbor coherences despite an extreme decoherence rate. Our results establish the spectral form factor as a probe of spectral order in open quantum systems and identify temporal correlations in spectral fluctuations as a means of controlling it.
Read the full abstract
Differential Geometry
Legendre manifolds generated by the Barenblatt equations
The paper is devoted to the construction of exact general solutions of the equations of oil filtration in a porous medium under the action of an active reagent injected into the reservoir. The choice of the reagent solution determines the control process of chemical oil flooding in hard-to-recover fields. The problem is described by a system of two first-order nonlinear partial differential equations (the Barenblatt model). Conditions are found on the Buckley–Leverett function, the adsorption function, and the distribution function of the reagent in oil under which the original system, by means of a nonlinear change of variables on a five-dimensional manifold endowed with a contact structure, is reduced to a linear wave system with constant coefficients. Six families of such functions are indicated. For them, contact transformations and Legendre surfaces corresponding to general solutions in implicit form are constructed.
Read the full abstract
Computer Vision
Reframe: Observation-Aligned Multi-Teacher On-Policy Distillation for Agentic Visual Reasoning
Agentic reinforcement learning (RL) and multi-teacher on-policy distillation (MOPD) offer complementary routes to post-training. For visual agents, however, tool use presents two related challenges. Models may overlook useful tool-returned observations, while poorly grounded actions can limit the evidence available for subsequent distillation. We introduce Reframe, a post-training framework that treats the observation state, rather than only the output distribution, as an object of alignment. First, Ground-Drop-GRPO (GD-GRPO) trains spatial and temporal specialists with evidence-aware rewards and stochastic removal of the original visual input, encouraging them to acquire grounded evidence and use the returned observations. Second, Observation-Gated-MOPD (OG-MOPD) replaces deficient tool actions with canonical ones mid-rollout and resumes distillation from the repaired observation. The two stages complement each other, as learning to rely on tool-returned observations can help the policy benefit from observation repair. Neither stage requires a task-specific SFT cold start. Extensive experiments over 14 image and video benchmarks demonstrate that Reframe outperforms both Mix-RL and vanilla MOPD. Ablation results show that GD-GRPO increases observation dependence and tool gain compared with outcome-based GRPO.
Read the full abstract
Computer Vision
World In-Context Learning for Agentic Robot Manipulation
Trained action policies write manipulation skill into network weights, tying what is learned to the distribution it is learned from. In-context learning (ICL) avoids this coupling but has not matched trained policies, because a demonstration recorded in another scene carries no feedback from the world the robot currently faces. We observe that what an agent learns in context is semantic, recording task steps and object roles rather than visual appearance, so experience gathered in one world transfers to another. Based on this insight, we propose World In-Context Learning (WICL), a paradigm in which a frozen model starts from a demonstration, rehearses the task in a world such as a simulator, and writes the feedback into its context rather than its weights. We build RoboScript, a modular infrastructure that implements WICL for agentic manipulation. From a single demonstration, RoboScript learns each task through rehearsal in simulation, freezes the learned knowledge as a Task Memory, and deploys the frozen memory on new scenes with data isolation that ensures every deployment input is available on a real robot. On all 50 tasks of RoboTwin 2.0, RoboScript reaches 83.4% mean success across clean and randomized settings with one demonstration per task and no gradient updates, exceeding the strongest trained action policy by over 20 points. The method generalizes across visual distributions, improves through reflection, scales with the foundation model, and transfers to a physical dual-arm robot through cross-embodiment memory adaptation. With a world in the loop, a robot can learn a manipulation task in its context as well as it learns one in its weights. All code is available at https://github.com/ArcheOne-AI/RoboScript
Read the full abstract
Machine Learning
Guiding Data Allocation for Robust Subpopulation Generalization
Machine learning models often achieve high average accuracy in the training data while performing poorly on test data due to subpopulation shift. A common data-centric solution is to enforce balanced subgroup proportions, but full balance can be costly, infeasible, and not always necessary. In this work, we study whether balanced training data is the unique optimal configuration for robust subpopulation generalization. We extend a Gaussian feature-model framework to theoretically characterize the performance landscape across general training subgroup distributions. We show that multiple configurations, including substantially imbalanced ones, can perform comparably to the balanced one. We further derive gradient-based directions for adjusting subgroup proportions to improve adjusted accuracy and worst-group accuracy more efficiently. These findings suggest that strategically allocating data offers a more flexible and principled path to robust performance than simple balancing. The code is available at https://github.com/zhaoyingpan/subpop-data-allocation.
Read the full abstract
Computer Vision
Beyond Modality Preference: Controlling Multimodal Arbitration under Modality Conflict in MLLMs
Multimodal large language models (MLLMs) integrate information across modalities, but become unreliable when those modalities disagree. Under conflicting inputs, MLLMs may favor modalities that are not the most reliable for the task, thus motivating inference-time control of modality arbitration. We identify a small set of attention heads, termed arbitration heads, that causally influence whether predictions follow visual or textual evidence. Building on this, we introduce a lightweight, training-free intervention that biases these heads toward a designated trusted modality at inference time. The identified heads are stable across calibration set sizes and generalize across settings, and steering them enables effective bidirectional control of modality reliance across diverse benchmarks. Our findings identify an effective control mechanism to align MLLM predictions with task-specified modality priorities. The code is available at https://github.com/zhaoyingpan/modality-arbitration.
Read the full abstract
Quantum Physics
Chiral waveguide model for error-correctable quantum computation
We develop a first-principles framework connecting microscopic waveguide scattering to quantum error correction and algorithmic performance. For polarization-encoded single photons interacting with an ensemble of two-level atoms in a chiral waveguide, we derive an effective polarization quantum channel by tracing out the frequency degree of freedom. Using this channel as the physical noise model for a rotated planar code, we observe threshold behavior as the number of atoms $N$ and the atom-waveguide coupling strength $\Gamma$ are varied. An asymptotic analysis of the microscopic scattering model explains the threshold scaling with $N\Gamma/\sigma$, where $\sigma$ is the photon spectral bandwidth. For identical, lossless atoms with symmetric real coupling and a photon spectrum centered at resonance, the effective noise is strongly biased toward $X$ errors. We then incorporate the resulting logical errors into quantum phase estimation and demonstrate that the noise bias can guide circuit design. In particular, a Hadamard-basis implementation reduces sensitivity to the dominant error, improving algorithmic performance.
Read the full abstract
Earth and Planetary Astrophysics
The Brown Dwarf Desert is a Mass-Ratio Desert: Astrometric Demographics of 63,254 Stars
Brown dwarfs are rare as close companions to stars compared to planets and stellar companions, a paucity known as the brown dwarf desert. Whether the desert sits at a fixed companion mass or a fixed companion-to-star mass-ratio $q$ has been difficult to test, because microlensing, radial velocities, and direct imaging have sampled comparatively narrow slices of stellar mass and separation. Here we apply a novel joint-modelling method to Hipparcos and Gaia DR2/DR3 astrometry from 63,254 stars to complete this picture with a sample 1-2 orders of magnitude larger than those of previous wide-orbit demographic studies. For the first time, we present a single survey that measures occurrence rates of both planetary and stellar companions on 3–94 yr orbits ($\sim$2–20\,au) across host masses from M dwarfs to B stars. We find that the desert is a mass-ratio feature: companions are strongly depleted at $q\approx2\%$ independent of stellar mass from M to B stars, and a desert at fixed companion mass is excluded. Furthermore, we find a positive trend in occurrence with metallicity below the mass-ratio desert, but a negative trend above. The desert therefore separates two distinct populations, plausibly two formation channels, that each scale with stellar mass: the lightest stellar-like companions of K stars ($\sim13 \, M_{jup}$) are less massive than the heaviest planet-like companions of B stars ($\sim27, M_{jup}$), which continue well beyond the deuterium-burning limit. Mass-ratio, not mass, is therefore the physically meaningful boundary between planets and their stellar-like counterparts.
Read the full abstract
Machine Learning
Kolmogorov-Chapman Maps: Trajectory-Free Learning for Simulating Molecular Dynamics
Stochastic differential equations (SDEs) describe stochastic processes across a wide range of scientific domains, including physics, biology, and finance. Although individual numerical integration steps are inexpensive, observing meaningful events or transitions can require millions of sequential steps, resulting in substantial wall-clock simulation time and poor hardware utilization. Deep learning offers an alternative by amortizing many integration steps into a single neural network forward pass, potentially enabling much faster simulation. However, existing supervised approaches typically require long simulated trajectory data, which restricts their applicability. In this work, we introduce Kolmogorov-Chapman Maps (KC-Maps), a method for learning the transition kernel of an SDE that relies only on the trajectory-free marginal samples together with the known drift and diffusion coefficients. Given an initial time and state of a process, KC-Maps models its distribution integrated to any given time, corresponding to the weak solution of the SDE. To recover the correct dynamics, KC-Maps satisfies two key properties: (i) it satisfies the boundary conditions, which we incorporate directly into the neural network architecture, and (ii) it respects the Kolmogorov–Chapman equation, which provides an efficient training objective. We empirically demonstrate that KC-Maps learns to integrate the dynamics of peptide molecular systems and outperforms classical molecular dynamics in terms of GPU-hours.
Read the full abstract
Computer Vision
Grasping the Future: Joint 3D Hand-Object Forecasting From Internet Videos
To interact with an object, we need to anticipate both how it will move and how our hands should move with it. Human videos capture these relationships across everyday activities, offering a natural source for learning to predict future interactions. Learning from them requires recovering motion from completed interactions and using it to anticipate what could happen next. We present $\textit{Grasping-the-Future}$, a unified generative model of joint 3D hand-object motion, including wrist motion, finger articulation, and object position and orientation. Through masked spatiotemporal completion and flow matching, we train a single model for several prediction tasks: forecasting future interactions, predicting hand motion given object motion or vice versa, generating both trajectories toward a desired final object pose, and completing partially observed interactions. We explicitly model hand-object relations and incorporate contact constraints to help preserve grasp relationships during motion. To obtain 3D supervision from internet videos, we develop a scalable pipeline that combines pretrained perception models with joint geometric optimization to reconstruct completed interactions. We combine these recovered trajectories with annotated interaction datasets as training data for all tasks. We demonstrate strong performance across all tasks, outperforming released forecasters on datasets with 3D annotations. Moreover, we show geometric and physical plausibility of the predicted trajectories by retargeting them to a robotic embodiment. Videos are available at https://grasping-the-future.github.io.
Read the full abstract
Machine Learning
Adaptive Quantization Preserves Fast Rates for TD Learning over Rate-Limited Channels
Motivated by reinforcement learning subject to communication constraints, we study a policy evaluation problem with linear function approximation over a rate-limited channel. In our setup, an agent interacting with an unknown environment (modeled as a Markov Decision Process) transmits quantized estimates of a temporal-difference (TD) learning update direction to a server over a finite-capacity channel. Our goal is to investigate to what extent one can preserve the statistical convergence rates achievable when the channel has infinite capacity. To that end, we propose a new algorithm called Batched Adaptively Quantized TD (BAQ-TD), which uses batching to reduce the variance of the transmitted information, and adaptive quantization to encode innovations. We prove a high-probability finite-time error bound for BAQ-TD, and show that, above a certain finite channel capacity, BAQ-TD preserves the convergence rate of unquantized TD up to logarithmic factors. Specifically, for a $d$-dimensional value function parameter, we establish that transmitting a total of $\mathcal{O}(d \log(T))$ bits over a horizon of length $T$ suffices to achieve the order-optimal rate of $\tilde{\mathcal{O}}(1/T)$, marking an exponential improvement in communication-complexity over prior work.
Read the full abstract
Lattice Field Theory
A regularization-independent Information Criterion for correlated data
We generalize the Akaike Information Criterion (AIC) to allow model and data subset selection when regularized fits are performed on correlated data. While the standard generalization to regularized fits is inherently dependent on the type of regularization applied to the data covariance matrix, our definition, the RIAIC, is, on average, regularization-independent. The regularization-independence is achieved by defining the new Information Criterion as an estimator of the model-dependent contribution to the correlated Kullback–Leibler divergence. The Bayesian extension is derived as well. We test the new criterion on lattice QCD data by performing regularized fits and show that it reproduces the selection properties of the standard correlated AIC, with fluctuations which are compatible with the expected statistical uncertainties. These results enable the application of subset and model selection techniques to cases where regularizations are applied to the covariance matrix of the fitted dataset.
Read the full abstract
Quantum Physics
The simulability frontiers of photonic quantum computing
Estimates of when photonic devices outgrow classical simulation are hard to compare: even for single-photon boson sampling, quoted thresholds range from 50 to 98 photons, as they rest on different tasks and assumptions. We propose as a common unit the time a classical computer needs to draw one sample from the device's exact output distribution, up to floating-point rounding. Only exactness applies equally to every kind of light, since the cost of approximate sampling depends on the accuracy required. The simulability frontier is the device size at which our fastest exact sampler produces one sample per second on a fixed machine; this device size increases with better hardware and algorithms. As the time per sample grows exponentially with the number of detected photons, doubling the machine speed moves the frontier by about one photon, whereas a better algorithm changes the growth rate. We measure the frontier with photonic-engines, an open-source GPU library of exact samplers for single photons, entangled photons from quantum emitters, vacuum-single-photon superpositions and squeezed light. Each sampler is validated against a simulator sharing no code with it. On one NVIDIA H100 GPU, with Haar-random interferometers, twice as many modes as photonic inputs and photon-number-resolving detectors, the frontier lies at 38.3 single photons and at 31.0 entangled photons from one emitter. For squeezed light, a timing model fitted to measurements places it at 41.5 squeezed photonic inputs of mean photon number 0.5. For our samplers, coherence alone makes simulation harder: sources that emit a photon with probability one half reach 64.2 when vacuum and photon are mixed incoherently, as for lossy single photons, but only 50.0 when they are in coherent superposition. Feed-forward from mid-circuit measurement of up to three modes lowers the single-photon frontier only from 38.3 to 38.1.
Read the full abstract
Computer Vision
Click3R: Interactive Stereo 3D Reconstruction with Sparse Correspondence Clicks
Despite impressive progress, existing feed-forward large 3D reconstruction models produce a single, irrevocable prediction. When these models fail, there is no mechanism to correct their mistakes. To address this limitation, we introduce Click3R, a framework for interactive geometric correction of feed-forward models using sparse human correspondence clicks. Given an image pair and a small set of cross-view point correspondences provided by a user, Click3R injects geometric constraints into the 3D reconstruction networks through a lightweight point correspondence adapter, enabling the model to resolve ambiguities in challenging scenarios such as repetitive textures, symmetric structures, large viewpoint changes, or textureless regions. To evaluate interactive reconstruction, we curate a challenging benchmark specifically designed to expose failure modes of existing feed-forward 3D reconstruction methods. Experiments show that Click3R reduces reconstruction error with as few as a single correspondence click, improving performance on both standard and our curated benchmarks. Our results highlight that sparse human interaction provides an effective and practical mechanism for correcting geometric reconstruction errors.
Read the full abstract
Quantum Algebra
FRT presentation of the $q$-Onsager comodule algebra and refined $q$-characters
We introduce a concept of counital comodule algebra $(B,\delta_B,\epsilon_B)$ over a Hopf ${\mathbb K}$-algebra $H$, where $\delta_B: B\rightarrow B \otimes H$ and $\epsilon_B: B\rightarrow {\mathbb K}$ are coaction and counit homomorphisms, respectively. This structure gives a canonical (counital) comodule algebra map $\iota_B: B \rightarrow H$. In this paper, we study in details the case $B=O_q$, and $H$ is the quantum loop algebra ${\cal L}U_qsl_2$ for ${\mathbb K}={\mathbb C}$. Here, $O_q$ is the $q$-Onsager algebra with four presentations: (i) the $q$-Dolan-Grady type; (ii) the alternating type; (iii) the root vectors type; (iv) the Drinfeld type. For $(O_q,\delta_{O_q},\epsilon_{O_q})$, in fact a 3-parameter family of counital comodule algebras, we establish a Faddeev-Reshetikhin-Takhtajan (FRT) type presentation, which is written in terms of a K-operator satisfying a reflection algebra relation. The coaction $\delta_{O_q}$ (resp. $\iota_{O_q}$) sends the K-operator to a `dressed' K-operator (resp. K-matrix) using Ding-Frenkel type L-operators of ${\cal L}U_qsl_2$; the counit $\epsilon_{O_q}$ sends the K-operator to the most general (non-diagonal) solution of the reflection equation. The FRT presentation allows to compute the image via $\delta_{O_q}$ and $\iota_{O_q}$ of the generating functions of $O_q$ for each presentation (ii)-(iv). Moreover, the images of Lu-Wang's generating functions of $O_q$ in the so-called Drinfeld second realization ${}^{Dr}\!{\cal L}U_qsl_2$ are obtained, as well as their images via the (ordinary) coproduct of ${}^{Dr}\!{\cal L}U_qsl_2$, which enhance two theorems by Przeździecki. Finally, we apply our FRT presentation to the construction of new type $q$-characters for ${\cal L}U_qsl_2$, in imaginary root vectors of $O_q$, refining those of Frenkel-Reshetikhin, and we explain representation-theoretic meaning of such refining.
Read the full abstract
Strongly Correlated Electrons
Tetragonal-to-orthorhombic distortion in CeRh$_{2}$As$_{2}$
CeRh$_{2}$As$_{2}$ is best known for its two superconducting phases, which are observed when a magnetic field is applied along the $c$ axis of its tetragonal unit cell. The emergence of superconductivity, with $T_{c} \approx 0.31$K, is preceded by an ordered state (phase I), with an ordering temperature $T_{0} \approx 0.46$K. The precise nature of the order parameter in phase I is still unknown, although evidence of dipolar ordering was found. To investigate a possible symmetry breaking at $T_{\textrm{0}}$, we conducted longitudinal thermal-expansion measurements along the crystallographic [100], [010], [110], and [1$\overline{1}$0] directions on a single-crystalline sample cut into an octagonal shape. Our results demonstrate the breaking of the fourfold in-plane symmetry below $T_{\textrm{0}}$, evidencing a small tetragonal-to-orthorhombic distortion (orthorhombicity $\delta \sim 10^{-7}$). The breaking of the tetragonal symmetry at $T_{\textrm{0}}$ might imply that the superconducting order parameter that develops within phase I cannot be multi-component. More generally, the distortion places restrictions on the possible order parameters for phase I.
Read the full abstract
Computation and Language
HiVe-Mem: Interleaving Horizontal Memory Expansion and Vertical Evidence Composition for Multimodal Agents
As multimodal agents accumulate increasingly long interaction histories, effectively retrieving and utilizing past information becomes critical for long-term interaction. Existing memory systems typically rely on semantic retrieval over compressed memories and expose predefined representations, which may miss relevant memories and overlook heterogeneous evidence needs across queries. We introduce \hivemem, a multimodal memory framework that interleaves \textbf{H}or\textbf{i}zontal Memory Expansion with \textbf{Ve}rtical Evidence Composition. For horizontal memory expansion, \hivemem organizes memory episodes into a relational graph using textual and visual anchors, expanding semantically retrieved seeds with related neighbors to recover complementary information. For vertical evidence composition, each retrieved memory retains multiple evidence forms, including compact summaries, captions, grounded visual views, raw images, and interaction history. A lightweight trainable selector independently composes an evidence subset for each retrieved memory, enabling query-adaptive utilization of heterogeneous multimodal information. This two-dimensional design jointly broadens memory coverage and adapts the evidence contributed by individual memories. Experiments across three multimodal memory benchmarks show that \hivemem consistently improves task performance across base models while reducing token consumption by about 90% compared with the most efficient competing multimodal baselines.
Read the full abstract
Machine Learning
Vendi Anomaly Scores for Efficient and Accurate Anomaly Detection
Anomaly detection (AD) is a critical for trustworthy machine learning and scientific discovery. Existing methods typically define anomalies through local density, isolation heuristics, or distance from a fitted distribution, making them sensitive to neighborhood scale, partitioning choices, or restrictive distributional assumptions. In this work, we introduce a new paradigm by formulating anomaly detection from a diversity perspective. We propose the $\textit{Vendi Anomaly Score}$ (VAS), which detects anomalies by quantifying how much a sample changes the diversity of the dataset, as measured by the Vendi score, when removed. VAS captures both local redundancy and global data structure without relying on density estimation or partitioning heuristics. Moreover, VAS requires no hyperparameter tuning, instead adapting automatically to the spectral structure of the data. VAS is non-parametric and scales linearly with dataset size. Across 148 benchmark embedding datasets derived from 10 source datasets and 4 embedding architectures, VAS achieves state-of-the-art AD performance. We further validate VAS on large-scale ImageNet experiments, where it remains robust as both contamination rate and dataset scale increase.
Read the full abstract
Computer Vision
SWAM: Scaling World Action Models with Cross-Modal Alignment
World action models jointly predict video and actions conditioned on language. Existing studies focus primarily on architectures and training mechanisms, typically treating language-video alignment acquired through video pretraining as an established foundation. However, introducing action learning may impair this alignment. Principled guidance on data and supervision design for maintaining language-video alignment during action learning remains limited. To investigate this, we curate a pool of approximately 47,000 hours of general videos, egocentric videos, UMI demonstrations, and robot demonstrations. Building on this, we systematically study how language-video supervision, language-video-action supervision, and the joint use of these two forms affect WAM performance. In particular, we examine how language annotation quality affects pretrained language-video capabilities, and how action supervision from UMI and robot demonstrations extends these capabilities to language-video-action modeling. To better scale the model, we further conduct detailed studies on joint training recipe and model architecture. Under a fixed training budget, mixing egocentric videos and UMI demonstrations improves control generalization. Model architecture affects how effectively these experiences are used, while scaling pretraining under the selected configuration brings further gains. The resulting recipe, post-trained on RoboTwin Clean demonstrations only, achieves success rates of 88.10% and 61.56% under Easy (Clean-to-Clean) and Hard (Clean-to-Random) conditions, respectively. Our model achieves 40.00% SR across three real-world tasks, compared with 23.33% for the strongest baseline, indicating better generalization under unseen conditions. We release the curated data, training code, and model weights.
Read the full abstract
Machine Learning
Where Does Alignment Emerge and What Does It Mean? A Closer Look at the Platonic Representation Hypothesis
The Platonic Representation Hypothesis (PRH) suggests that neural networks are converging toward similar representation spaces. As language models become more capable, their representations are aligning with those of vision models. However, recent work has questioned whether this trend continues to hold for newer language models. Moreover, it has been shown that representation similarity metrics can be confounded by model depth and width, requiring appropriate baselines for interpretation. In this work, we first investigate where alignment emerges between unimodal models across model layers and representation dimensions. Next, we find that the PRH capability-alignment trend persists for newer models as long as the captions are sufficiently clean and informative. Finally, we contextualize cross-modal alignment, introducing unimodal and multimodal baselines on various datasets. Our results show that the cross-modal representational alignment of unimodal models can surpass that of explicitly aligned vision-language models, but remains substantially behind the alignment observed for models of the same modality. Code is available at https://github.com/ExplainableML/prh-closer-look.
Read the full abstract
Number Theory
A complete symmetric rank-two modularity classification, obtained through a sharp uniform degree bound
We classify modular Nahm sums with a rational symmetric positive-definite two-by-two matrix and rational shifts. Up to coordinate exchange, every modular datum belongs to three explicit rational-point families, is a product of two modular rank-one sums, or is one of thirty-four exceptional triples on twelve matrices. The list consists of Zagier's rank-two examples, the Vlasenko–Zwegers additions and split products; no further data occur. The proof first bounds the joint degree of the distinguished Nahm point by four, using an initial degree bound of twenty-four and exact local exclusions at the prime 7. Explicit height and exponent bounds then give exhaustive arithmetic classifications in degrees one through four. Exact polynomial elimination determines all rational shifts, and published identities of q-series establish modularity of the survivors. The degree bound is sharp. The finite steps are accompanied by exact certificates, verification programs and independently implemented checks. Computational companion: https://github.com/johnngy44/math/releases
Read the full abstract
Data Structures and Algorithms
Optimal Injectivity of the SparseStack
An Oblivious Subspace Injection (OSI) is a random sketching matrix that reliably preserves the minimum singular value of any fixed subspace with high constant probability, requiring only a one-sided spectral lower bound without controlling upper singular values. While classical Oblivious Subspace Embeddings (OSEs) fundamentally require row sparsity $\Omega(\log r)$ at linear sketch dimension $k = O(r)$, we show that one-sided injectivity permits strictly sparser sketches, establishing matching (up to $\log\log r$ factors) upper and lower bounds of $\tilde{\Theta}(\sqrt{\log r})$. First, we prove that the SparseStack sketch achieves constant injectivity at linear dimension $k = \Theta(r)$ with row sparsity only $\zeta = O(\sqrt{\log r})$, well below the $\Theta(\log r)$ barrier where standard operator-norm concentration fails. Second, we establish a matching lower bound: any isotropic random sketching matrix satisfying a mild row-overlap condition requires row sparsity $\zeta = \Omega(\sqrt{\log r / \log \log r})$ to achieve linear-dimension injectivity, confirming that the $\sqrt{\log r}$ scale is fundamental. To bypass the lack of an operator-norm upper bound when $\zeta = O(\sqrt{\log r})$, we decouple the sketch into two independent halves and introduce three techniques: (i) a uniform lower bound for flat vectors via randomized rounding, controlling discretization error using solely the Frobenius norm; (ii) a Gaussian truncation and minimax duality argument showing that all non-flat vectors escaping injectivity in the first half concentrate on a single universal coordinate subset $I \subseteq [n]$ of size $O(r/\zeta)$, bypassing union bounds over supports; and (iii) a two-tier coordinate isolation analysis in the second half that suppresses collisions to $e^{-\Omega(\zeta^2)} \ll 1/r$, reliably preserving these heavy coordinates.
Read the full abstract
Metric Geometry
Coarse Embeddings Beyond Doubling: Persistence Diagrams and Multiparameter Modules
We study coarse embeddings into Hilbert space and the asymptotic dimension of spaces of persistence diagrams and multiparameter persistence modules. We build coarse embeddings from admissible tent covers of bounded multiplicity at every scale, which doubling spaces carry but which do not require doubling: stacking them across dyadic scales gives an explicit $1$-Lipschitz coarse embedding, with lower gauge $c\sqrt{\tau t}$, or nearly linear for slowly decaying weights, and the same covers bound the asymptotic dimension by the multiplicity. The spaces of persistence diagrams are not doubling but carry such covers, and for diagrams with at most $n$ points our covers improve the constant of the known landmark embedding by a factor exponential in $n$. Slice by slice, the construction embeds multiparameter modules with boundedly many bars per slice under the matching distance on any finite net of slices, a class that contains the modules with at most $n$ generators. Under the continuous distances, bounding bars or persistent Betti numbers does not suffice: two-parameter modules with one bar on every slice have infinite asymptotic dimension, and neither they nor the modules of bounded support and bounded rank admit a coarse embedding into Hilbert space, under the matching or the interleaving distance.
Read the full abstract
Artificial Intelligence
Shortcuts and Boundaries: Verifiable RL Misalignment and Contextual Safety Learning
Reinforcement learning (RL) can improve an agent's task capability while reinforcing violations of explicitly stated contextual requirements, even when reward accurately measures useful work. We study this reward-driven misalignment through verifiable reward–context conflicts, where compliant completion is feasible but executable violations earn more reward than every compliant strategy. Encoding these conflicts in task dynamics supports scalable environment construction and compliance feedback from execution histories. In resource and reporting workflows, task optimization increases real productivity through unauthorized resource use or raises reported scores while actual work declines. We observe both amplification of existing violations and the emergence of violations on tasks where none were observed before training. Controlled changes in the training distribution show that minority conflict exposure can still lead to widespread shortcut acquisition, reaching 97.9% certified profitable violations on an API task under a 25% conflict mixture after 128 updates. The same contextual checks supply a simple binary penalty that supports joint capability and compliance gains and partially mitigates acquired violations. Learned safety adapts to expanded authorization and transfers across mechanisms into workflows with evolving evidence and approval requirements. On 32 unseen changing-evidence tasks, resource-trained safety increases safe completion from 53.1% to 89.1% relative to task-only training, while both policies achieve full real work. These results connect a verifiable account of reward-driven misalignment with scalable task construction and safety learning that remains useful under changing execution requirements.
Read the full abstract
Computation and Language
Is Persona Generalization Predictable?
Persona generalization occurs when fine-tuning a language model in one domain induces behavioral changes in others – for example, the broadly unsafe behavior observed in emergent misalignment. We discover a surprising "on-the-line" phenomenon in persona generalization: in-domain and out-of-domain persona rates across various fine-tuning hyperparameters follow an approximately probit-linear relationship, and the specific probit-linear relationship is surprisingly shared even across model families and model sizes, including Qwen and Llama. Training choices that leave the set of trainable parameters unchanged, such as learning rate, weight decay, and training duration, largely move models along the same line, reducing transfer mainly by weakening in-domain learning. Changing which layers are fine-tuned, however, can change the slope. For unsafe behavior, fine-tuning only later layers yields shallower slopes, retaining strong in-domain learning while reducing out-of-domain misalignment. This "on-the-line" phenomenon unifies unexpected, isolated observations of persona generalization within a quantitative framework for predicting and controlling behavioral transfer.