General Mathematics
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Friday 11 September
The conformal geometric algebra Cl(4,2) ~ R(8) carries a natural Z_2-grading defined by the dilatation bivector D=\e_4e_0, which satisfies $D^2=1$ and splits the eight-dimensional spinor module into two four-dimensional eigenspaces $S=S_+\oplus S_-$. We prove that the ad_D-eigenvectors of Cl(4,2), those elements with eigenvalue~$\pm1$ under the Lie bracket $[D,\cdot]$, are exhausted by 16 doublets of the form $T^\pm_J=n_\bullet e_J$, where $n_\bullet \in {n_\infty,n_o}$ is a null vector and $e_J$ is a product of Lorentz-sector basis vectors indexed by $J\subseteq\{1,2,3,5\}$. These doublets satisfy universal algebraic identities: $(T^\pm_J)^2=0$ (nilpotency) and $T^+_J\,T^-_J=(-1)^{k(k+1)/2+1}\,\eta_J 2\Pi_+$ (single-projector proportionality), where $k=|J|$, $\eta_J=\prod_{j\in J}\eta_{jj}$ is the spectator metric, and $\Pi_+=\tfrac12(1+D)$. The contracted sums $\sum_{|J|=k}T^+_J T^-_J$ are shown to be controlled by the elementary symmetric polynomials $S_k$ of the Lorentz signature values $(+1,+1,+1,-1)$, with the vanishing of $S_2=0$ reflecting the $(3,1)$ signature of physical spacetime. The even-grade doublets (belonging to Cl^+(4,2) ~ Cl(4,1)) contribute $8\Pi_+$ to the unweighted grand total, while the odd-grade doublets cancel exactly. The 32 doublet eigenvectors generate, under the commutator bracket, the 63-dimensional Lie algebra sl(8,R) in which the conformal algebra so(4,2) embeds as a 15-dimensional subalgebra. All identities, including a closed form for the commutators across the Levi interface, are proved in every conformal extension Cl(p+1,q+1); there the commutator span misses precisely the center of the algebra, a parity dichotomy invisible at the Lorentz signature. Two applications are developed: the top rung of the tower computes the Dirac versus Majorana dichotomy of rank-five pseudoscalars ... [shortened]
Robotics
CHOREO: Every Humanoid Skill as a Trajectory
Recent advances in humanoid robotics have produced diverse skills through reinforcement learning, motion imitation, and generative modeling. Yet these capabilities remain siloed because they are built around incompatible representations, interfaces, and controllers. We present CHOREO, a framework for training-free composition of heterogeneous humanoid skills. Our key observation is that, regardless of how a skill is learned, it can ultimately be expressed as an executable motion trajectory. Based on this observation, CHOREO converts each capability into SkillMotion, a unified representation that combines motion states, contacts, semantics, and boundary conditions. Skills are composed through direct continuation, cubic Hermite blending, or validated bridge motions, without retraining source models or updating models at test time. On Unitree G1 in MuJoCo, CHOREO organizes 2,950 admitted SkillMotion assets derived from heterogeneous sources and achieves 95.4% sequence success across 130 multi-action tasks, including 93.8% success on eight-action sequences. These results demonstrate that executable trajectories provide a scalable interface for accumulating and composing pretrained humanoid capabilities.
Read the full abstract
Combinatorics
Partial-Twuality Polynomials of Paired Matrices
Gross, Mansour, and Tucker~[European Journal of Combinatorics, 95 (2021): 103329] introduced the partial-twuality polynomials of ribbon graphs. Recently, Deng, Jin, and Yan generalized the partial-twuality polynomials to the framework of matrix algebra and investigated several of their basic properties. They asked whether there exist matrix operations, called partial duality $\delta$ and partial Petrie duality $\tau$, on pairs $(M,A)$, where $M$ is a square matrix whose rows and columns are indexed by a finite set $V$ and $A\subseteq V$, such that $\delta^2=\tau^2=(\delta\tau)^3=id$ and the exponent of the partial-twuality polynomials coincides with some parameter of the matrix obtained by applying \(\bullet\) to \((M, A)\). In this paper, we introduce a paired-matrix framework for partial-twuality polynomials over the binary field $\mathbb{GF}(2)$. We prove that there exist two local operations \(\delta\) and \(\tau\) on \(\bigl((M,I_{|V|}),A\bigr)\) satisfying $\delta^2=\tau^2=(\delta\tau)^3=id$ and $P_{\langle \bullet \rangle}((M,I_{|V|}),z)=P_{\langle \bullet \rangle}(M,z)$ for $\bullet \in \{\delta, \tau, \delta\tau, \tau\delta, \delta\tau\delta\}$, thereby answering their question affirmatively. Finally, we establish a recurrence relation for the partial \(\langle\delta\tau\delta\rangle\)-polynomial with respect to an edge. This recurrence enables the computation of the partial \(\langle\delta\tau\delta\rangle\)-polynomial for certain bouquets, simple graphs, and simple signed graphs.
Read the full abstract
Computer Vision
Moonworks Lunara: Modeling Artistic Intelligence
We formulate Artistic Intelligence as exploration driven world realization, leaving space for creative possibility while preserving the semantic, artistic, and compositional structure that must remain true. Moonworks Lunara, a text-to-image model, implements this framework with a novel Diffusion Mixture Transformer architecture. A new training algorithm iteratively evolves the data distribution through informative sample acquisition and targeted injection of human-created art. We benchmark Lunara against seven image-generation models, including FLUX.2-Klein-4B, Qwen-Image (20B), and GPT-Image-1-Mini. With GPT-5.6 Sol as evaluator, Lunara ranks first in Aesthetic Quality (8.473 vs. 8.457 GPT-Image-1-mini), second in Emotional Resonance, and remains competitive in Content Integrity. A blind human evaluation over the same evaluation set corroborates the automated metrics, ranking Lunara first. It also stays among the strongest models under conventional measures including CLIPScore and LAION Aesthetic Predictor. On GenEval, Lunara achieves competitive performance against a broader set of 16 models, including GPT Image 2 and Seedream 4.0. These results place Lunara at the frontier with Artistic Intelligence while maintaining a sub-10B active-parameter footprint and sub-10-second inference latency. Lunara advances the general visual intelligence frontier by shifting the question from whether models can get images right to how deeply they can interpret meaning and realize it as imaginative, expressive worlds.
Read the full abstract
Computation and Language
To Memories and Beyond: From Remembering to Knowing You across Long-Term Multimodal Personal Archives
As AI systems evolve into personalized digital companions, a central capability is reasoning over a user's long-term personal history: not merely storing past events, but tracking longitudinal experiences and evolving preferences. Progress here is bottlenecked by evaluation, existing long-term memory benchmarks are largely synthetic and text-only, they overlook the visual records that anchor everyday human memory, lack the authentic and causally connected longitudinal data that real personalization demands, and consequently remain confined to shallow factual recall. We introduce ReaLMem (Real-world Long-term Multimodal Memory), the first benchmark built from authentic multi-year personal visual archives, paired with first-person subjective annotations. ReaLMem evaluates models across three cognitive tiers of increasing difficulty: factual recall, persona inference, and predictive personalization. We further propose ChronoProfiler, a temporal-weighting profiling module that computes temporal stability scores for user attributes and applies them as a salience prior, resolving conflicts among temporally inconsistent preferences and helping models compound multiple co-active preferences in complex personalized decisions. Extensive evaluation of frontier multimodal large language models (MLLMs) and memory systems on ReaLMem reveals predictive personalization as a consistent ceiling, exposes clear performance gaps and bottlenecks between MLLMs and memory systems, and shows that high-quality, temporally informed representations substantially improve personalization. Together, ReaLMem and ChronoProfiler provide an authentic testbed and a simple, effective mechanism for long-term personalization, laying a foundation for future research on lifelong AI companions.
Read the full abstract
Metric Geometry
Monge's transport problem in synthetic Lorentzian spacetimes
We prove the existence of solutions to the Lorentzian Monge problem on synthetic Lorentzian spacetimes that satisfy the forward timelike measure contraction property $TMCP^+(K,N)$. The cost for this problem is the time separation function $\ell(x,y)$ (the Lorentz distance), which represents the maximum amount a particle can age when traveling from $x$ to $y$. The solution is based on the needle decomposition technique, and amounts to a reduction of the full problem to its one-dimensional counterparts. As an application of needle decomposition, we introduce the blended $TMCP$ condition, and obtain a quantitative refinement of the synthetic Hawking singularity theorem. We additionally prove a timelike nonbranching reversibility theorem for $TCD_p(K,N)$ spacetimes, showing that forward $p$-essentially nonbranching spacetimes are backward $p$-essentially nonbranching.
Read the full abstract
General Relativity and Quantum Cosmology
From Geometry to Gravity: A Friendly Introduction to the Schwarzschild Metric
The first exact solution to the field equations of general relativity was found by the German physicist and astronomer Karl Schwarzschild in 1916. Due to its simplicity and its various astronomical applications, the Schwarzschild solution is the most important, which gives it great educational potential. However, this potential is not usually exploited, since the Schwarzschild solution is generally taught in advanced undergraduate physics courses or in specialized graduate courses. The objective of this article is to offer a friendly introduction to the Schwarzschild solution, and at the same time analyze some concepts of Riemannian geometry and general relativity, employing only simple mathematical tools that do not go beyond the rudiments of differential and integral calculus.
Read the full abstract
Neural and Evolutionary Computing
Evolutionary Ensemble Search: Council-Guided Program Evolution with Persistent Memory
Evolutionary Ensemble Search (EES) constructs machine-learning procedures through expert-guided program evolution. A role-specialized council turns task evidence and experimental results into structured search directions. An orchestrator allocates these directions to execution specialists and an evolutionary engine. The engine selects measured parents, diagnoses their errors, and produces descendants through code mutation, structured pipeline edits, and crossover. Each child must execute and acquire its own validation evidence. Population archives retain useful alternatives, while compatible predictions compete in a validation-gated ensemble stage. Search adapts through parent-relative operator credit, session memory, and problem-indexed lessons retrieved across runs. We specify these mechanisms, distinguish their execution profiles, and define the contracts required to compare candidates as their computations change. A public MLE-bench Lite development ledger records medal-threshold artifacts on 19 of 22 tasks (86.36%), with best outcomes of 11 gold, five silver, and three bronze. The procedures span text, images, tables, audio, scientific geometry, and deterministic transformations. The campaign includes grade feedback between runs, external-source routes, and mixed confirmation procedures; its aggregate is an achieved development result, not a blind autonomous-agent success rate. The report contributes a concrete architecture for cumulative executable search and a versioned account of its cross-modal development outcomes.
Read the full abstract
High Energy Physics (Phenomenology)
Astrophysical Signatures of Fermionic Dark Matter
Fermionic dark matter particles remain one of the most compelling candidates for the dark matter content of the Universe, yet no positive results have been obtained from direct detection experiments. In this work, we investigate the possibility that such particles form compact gravitationally bound objects supported by degeneracy pressure. Through an extensive review of microlensing surveys, we derive constraints on the masses of these compact objects. We further analyze the accretion of baryonic matter onto these objects and evaluate their thermal and radiative properties. The estimated burst emission is found to be well below the energies associated with the Galactic Center GeV excess, suggesting that these compact objects are unlikely to be the source of the observed signal. Our analysis suggests that admixed dark matter and baryonic matter objects could potentially account for a fraction of the presently unobserved baryonic matter, thereby providing a possible explanation for a fraction of the missing baryons in the Universe.
Read the full abstract
General Relativity and Quantum Cosmology
First-Law Entropy and a Degenerate Extremal Remnant in a Minimal-Length Simpson–Visser-Type Regular Black Hole: Geometrothermodynamics, Phase Structure, and Observational Discriminants
We construct the first-law-consistent entropy of a geometrically minimal-length-deformed Schwarzschild spacetime, obtained via the areal-radius substitution $R(r)=\sqrt{r^{2}+\ell^{2}}$ on $r\in[0,+\infty)$ with $f(R)=1-2M/R$, whose nonvanishing Einstein tensor sources an effective geometric fluid with no classical matter counterpart. Integrating the first law gives $S=\pi[r_{h}R_{h}+\ell^{2}\ln((r_{h}+R_{h})/\ell)]$. This coincides in functional form with the semiclassical term found independently by Joshi and Joshi, but we fix its boundary condition $S(r_h=0)=0$ on independent physical grounds and adopt it, rather than the Bekenstein–Hawking area law, as the complete entropy of the model, building the free energy, geometrothermodynamics, and mode-stability analysis on it. Evaporation, governed by the Helmholtz free energy $F(M)$, terminates at $M_{\min}=\ell/2$ in a previously unrecognised endpoint: a degenerate extremal regular black hole, where the regular centre coincides with a degenerate Killing horizon of quadratic order, $f\approx r^{2}/(2\ell^{2})$, at areal radius $\ell$, with $T_{H}\to 0$, $S\to 0$, $C\to 0^{+}$, and finite curvature everywhere; we display its Penrose–Carter structure for the first time. The same entropy, with $\ell$, defines the equilibrium state space of a Legendre-invariant geometrothermodynamic (GTD) description whose curvature scalar diverges independently at the Davies-type transition $M^{*}=\ell/\sqrt{2}$ and at $M_{\min}$, a divergence with no counterpart in the minimal-length black hole literature, specific to a genuine horizon at $S=0$. We embed this remnant within the observational discriminant noted qualitatively by Tsukamoto: exact shadow degeneracy combined with a measurable photon-ring flux enhancement $r_{n}=e^{-2\pi/a}>e^{-2\pi}$, accessible to next-generation very-long-baseline interferometry.
Read the full abstract
Robotics
MR-GLi: Mixed Reality-Based Gripper-Linked Overlays for Underwater Robot Arm Teleoperation via Bilateral Control
Visual torque feedback supports underwater bilateral teleoperation, but the benefit of mixed reality (MR) over conventional monitor presentation remains unclear. We present MR-GLi, an MR interface that spatially registers a reaction torque indicator and wrist-camera image to the robot gripper. Twenty participants performed lift and pick-and-place tasks with rigid and compliant objects in a counterbalanced within-subject comparison with a 2D monitor, using identical visual-feedback content and four-channel bilateral control. MR-GLi provided gripper-linked access to visual feedback while maintaining a similar level of torque-regulation performance to the 2D monitor. Subjective evaluation further indicated reduced perceived burden associated with shifting attention between the workspace and visual feedback. These results demonstrate the feasibility of gripper-linked MR overlays for underwater bilateral teleoperation and highlight the importance of considering information access in addition to task performance. Additional material: https://mertcookimg.github.io/mr-gli/
Read the full abstract
Robotics
Bi-MoDe: Bilateral Control-based Imitation Learning via Modifier-Conditioned Decoding for Modulation of Execution Speed and Contact Intensity
Bilateral control-based imitation learning captures both position and force information, making it well suited to contact-rich manipulation. However, existing approaches provide limited means for an operator to specify how a learned task should be executed at inference time, such as slowly or quickly, gently or firmly. We propose Bi-MoDe, a modifier-conditioned decoding framework that injects a constrained latent into every layer of the Transformer action decoder via adaLN-Zero, allowing behavioral directives to directly influence action-chunk generation. We evaluate the method on a real-world whiteboard wiping task with combinations of temporal and physical modifiers. Bi-MoDe improves physical directive following over the action-chunking baseline while maintaining comparable temporal control. An ablation further shows that decoder conditioning and latent-space composition interact, and that their combination is important for accurate physical directive following. Additional material is available at the https://mertcookimg.github.io/bi-mode/
Read the full abstract
History and Overview of Mathematics
On the Reconstruction of SAS from Other Triangle Congruence Criteria
Starting from a Hilbert plane and removing the Side-Angle-Side (SAS) congruence axiom, we investigate to what extent SAS can be recovered synthetically from the remaining classical triangle congruence criteria. We show that the Angle-Side-Angle criterion, together with a ray correspondence principle corresponding to Theorem 13 of Hilbert's Grundlagen der Geometrie, suffices to reconstruct SAS. We further show that both the Side-Side-Side and the Side-Angle-Angle criteria also suffice, once combined with the ray correspondence principle and suitable auxiliary principles – the existence of midpoints and a hypotenuse-angle criterion for right triangles in the first case, and the existence of angle bisectors, the congruence of supplements of congruent angles, and the Pons Asinorum in the second. Although the two routes rely on auxiliary principles of different character, we show that they converge on a single final argument once a common hypotenuse-angle criterion is established. A metamathematical analysis, based on an explicit model adapted from Hilbert's own independence construction, complements these reconstructions: it shows that the ray correspondence principle alone cannot reconstruct any of the classical criteria, and that the Pons Asinorum and the hypotenuse-angle criterion are each independent of the remaining auxiliary principles used in their respective reconstructions. The resulting picture is not a formal hierarchy of the congruence criteria, but it does show that the Angle-Side-Angle reconstruction rests on a provably more economical basis than those obtained from Side-Side-Side or Side-Angle-Angle.
Read the full abstract
General Mathematics
Study of the algebra of smooth integro-differential operators with applications
We study the algebra of integro-differential operators with smooth coefficients and kernels on a subspace of $C^\infty[a,b]$. We find a normal form for elements of this algebra and determine its unit group. The formulation of inverses gives explicit solutions of inhomogeneous linear Volterra integro-differential equations and Volterra integral equations of first kind with smooth kernels.
Read the full abstract
General Physics
Observational Constraints and Cosmic Growth Index of Realistic $f(G)$ Gravity Frameworks using MCMC Analysis
We present a comprehensive observational analysis of modified Gauss–Bonnet, or $f(G)$, gravity by investigating both the background cosmological expansion and sub-horizon linear matter perturbations. We consider two viable functional forms: an arctangent parameterization (Model~I) and a generalized polynomial power-law model (Model~II). Using a Markov Chain Monte Carlo (MCMC) ensemble sampler, we constrain their parameter spaces through background and structure-growth observations. Our analysis employs cumulative combinations of Cosmic Chronometers (CC), the Pantheon+ Type Ia Supernovae compilation (PP), Redshift-Space Distortions (RSD), and the Year-1 Baryon Acoustic Oscillation measurements from the Dark Energy Spectroscopic Instrument (DESI BAO). The inclusion of DESI BAO data produces a noticeable downward shift in the preferred values of the Hubble constant, $H_0$, and present matter density parameter, $\Omega_{m0}$. For Model~II, the full analysis yields $H_0=64.02^{+4.13}*{-3.08},\mathrm{km,s^{-1},Mpc^{-1}}$ and $\Omega*{m0}=0.249^{+0.047}_{-0.038}$. We further assess the statistical performance of the models relative to $\Lambda$CDM using $\Delta\mathrm{AIC}_c$, $\Delta\mathrm{BIC}$, and $\Delta\mathrm{DIC}$. For Model~II, the full dataset gives $\Delta\mathrm{AIC}_c=4.233$ and $\Delta\mathrm{DIC}=5.971$, favoring the $\Lambda$CDM baseline. At the perturbation level, both models exhibit stable growth histories compatible with large-scale structure observations. The models predict transitions in the late-time expansion dynamics at $z\approx0.5005$ and $z\approx0.6086$ for Models~I and II, respectively, highlighting differences in their cosmological evolution.
Read the full abstract
Image and Video Processing
Anatomy-Change-Aware Bidirectional Selective State-Space Memory for Clinically Deployed Thoracic Radiotherapy Auto-Contouring
We developed DAMM-Net++, a 2.5D architecture for thoracic OAR and target volume segmentation that addresses three persistent challenges in radiotherapy auto-contouring: inter-slice surface incoherence, systematic failure on small low-contrast targets, and the absence of per-case reliability signals. The central component is an anatomy-change-aware bidirectional selective state-space memory that models through-plane anatomical change and selectively propagates context along the axial slice sequence. A boundary-aware decoder sharpens near-surface predictions, and an uncertainty head provides calibrated per-voxel confidence for clinical triage. We evaluated 2,146 patients across four centers, an independent external cohort of 112 patients, and a multicenter reader study involving 17 radiation oncologists on 305 cases. The model achieves a mean Dice of 0.955 and HD95 of 3.78 mm, with the largest gains on low-contrast organs-at-risk (OARs) and target volumes where through-plane context is most critical. The uncertainty head is well-calibrated and supports case-level triage. In the reader study, AI assistance reduced contouring time by 75-80% across experience levels and raised junior-reader IoU from 0.861 to 0.925, matching the unedited model. External validation showed a modest internal-to-external drop (< 5%) with calibrated uncertainty transferring without recalibration. The complete deployment pipeline from DICOM ingestion to TPS-compatible RTSTRUCT export has been integrated into the clinical workflow at a partner hospital, where it is used to assist with contouring. These results suggest that anatomically motivated inter-slice memory, paired with uncertainty-guided review, offers a clinically viable path for thoracic auto-contouring.
Read the full abstract
Image and Video Processing
Automated Distinction of Intimal and Medial Intracranial Arterial Calcification from CT Head
Intracranial arterial calcifications (IACs) are a common finding on clinical non-contrast enhanced head CT scans and are associated with neurovascular disease. Calcifications can occur in the intimal or medial layer of the arterial wall, subtypes that differ in aetiology and may have distinct clinical relevance. These subtypes can be visually distinguished by radiologists based on the shape of the calcifications. We investigate three automated approaches for subtype classification of IAC from head CT-derived segmentation masks: (1) an automated adaptation of the established radiological visual score, (2) a sphericity-based method, and (3) a method based on shape embeddings extracted by a medical shape foundation model. All approaches use the same lightweight classification pipeline on top of the features they compute and are evaluated using 5-fold cross-validation. The three methods achieved comparable performance, with the embedding-based approach yielding the best overall results with a weighted F1 (mean $\pm$ SD) of up to 71.5 $\pm$ 3.7 for a single artery and 59.8 $\pm$ 1.7 for the joint artery classification. Performance was largely preserved when using automated instead of manual IAC segmentation masks, and we found the difference in weighted F1 not significant. Our results show that fully automated IAC subtype quantification from head CT is feasible and remains robust to the use of manual and automated IAC segmentation masks. Code at https://github.com/bjin96/iac-subtyping.
Read the full abstract
Audio and Speech Processing
StepAudio 3 Music Technical Report
We introduce StepAudio 3 Music, a large-scale, long-form music generation model that supports explicit musical planning and open-domain text-controlled generation. The StepAudio Music Tokenizer represents audio as a 50-Hz stream from a 65536-entry single codebook, using semantically informed self-supervised and multi-task training to preserve musical structure and reconstruction-relevant information. A flow-matching diffusion Transformer (DiT) predicts continuous StepAudio VAE latents, which our VAE decoder converts into 48-kHz audio. This discrete-continuous design is guided by comparisons of single-codebook VQ, Semantic and Acoustic RVQ, and different DiT configurations. For explicit planning, a Mixture-of-Experts autoregressive model uses ABC notation to produce an intermediate arrangement plan (ABC-CoT) before predicting music tokens, making harmony, rhythm, and melodic structure part of the generation context. A progressive training curriculum and supervised fine-tuning support song and instrumental generation, accompaniment generation from dry vocals, and cover-song synthesis for up to 5 minutes and 30 seconds. With reinforcement learning via direct preference optimization (DPO), the final model achieves the highest AudioBox Content Enjoyment, Content Usefulness, and Production Quality scores and the highest MuQ-MuLan similarity among the evaluated systems, with competitive SongBench results. On the preliminary Artificial Analysis Music Arena Vocals leaderboard, it obtains a Quality Elo of 1105, behind only Suno V5.5 and Mureka and ahead of Suno V5, MiniMax models, and other systems. Audio demonstrations are available at https://stepaudiollm.github.io/step-audio-3-music.
Read the full abstract
Image and Video Processing
LM-PCVMNet: Pediatric Cervical Vertebral Maturation Analysis with Deep Fusion of Landmarks and Metadata
Cervical vertebral maturation (CVM) assessment plays a pivotal role in orthodontic diagnosis and determining the optimal timing of treatment, especially for pediatric patients. In this paper, we propose LM-PCVMNet, a novel deep learning framework for automatic pediatric CVM staging. Specifically, our method integrates vertebral anatomical landmark information, heatmap-guided feature modulation, and metadata-informed similarity modeling into a unified learning framework. We introduce a heatmap-guided feature modulation module that enhances feature extraction by leveraging landmark-centered heatmaps to highlight morphologically relevant vertebral regions. A vertebral landmark-prompting block is designed to incorporate anatomical geometry into the representation learning process. Furthermore, we develop a learnable metadata supervised contrastive loss that adaptively modulates positive-pair similarity based on metadata similarity, enabling the model to learn more biologically consistent and discriminative features. To facilitate further research in pediatric orthodontic treatment, we additionally release PCVM+. It contains 1800 lateral cephalometric radiographs from real-world patients aged 3-15 years, with expert-annotated CVM stages, 13 vertebral anatomical landmarks, and corresponding metadata. We perform comprehensive experiments on two datasets, and the results show that our method achieves state-of-the-art performance, effectively improving landmark localization and classification accuracy over existing models. Code and dataset will be available at github.com/ybupengwang/LM-PCVMNet.
Read the full abstract
Image and Video Processing
Conditioning noise is a free regularizer for LoRA fine-tuning: no pathology encoder required for diffusion-based artifact detection in histopathology
Diffusion-based artifact detectors score whole-slide image patches by reconstruction error under a model fine-tuned on clean tissue. We show that conditioning this fine-tuning on random Gaussian embeddings – resampled at every step from approx. 200 KB of precomputed embedding statistics, with no encoder, no cache, and no change to inference – consistently widens the clean/artifact separation. A four-step ablation chain shows the benefit requires neither content (shuffled real embeddings), provenance (synthetic Gaussians), a tuned intensity (flat across an 8x variance range), nor per-patch identity (fresh per-step noise); a LoRA-dropout control shows the conditioning pathway specifically, not generic weight perturbation, carries the effect. Patch-level gains of +0.25-0.48 Cohen's d replicate across nine trainings; honest leave-one-slide-out evaluation clears a pre-registered bar in 2/2 seeds; and two pre-registered external endpoints on a 281-case set confirm pooled Delta F1 = +0.0073 (95% CI) and +0.0129 (97.5% CI, two-look corrected). We release the full evaluation protocol, including measured seed noise and selection-optimism pricing.
Read the full abstract
Image and Video Processing
A deep dictionary network-based foundation model for ultra-low-dose CT denoising
Ultra-low-dose computed tomography (ULDCT) reduces radiation exposure but suffers from severe noise that degrades diagnostic image quality. Existing deep learning-based denoising methods are typically trained in an organ-specific fashion, resulting in limited generalization across heterogeneous multi?organ imaging scenarios. Foundation models present a promising all-in-one paradigm for unified multi-organ denoising. However, their architectures suffer from poor interpretability and rely on heuristic training strategies. To address these limitations, we propose an architecture?interpretable foundation model based on the deep dictionary network (DDN) for unified multi-organ ULDCT denoising. Inspired by multilayer sparse representation theory, DDN cascades convolutional sparse coding layers with iterative soft-thresholding, providing inherent architectural interpretability. Furthermore, a dynamic dictionary module and a threshold generation module are embedded within each layer to enhance representation ability. We conduct DDN pre-training on more than one million multi-organ normal-dose CT images by recovering clean images from Gaussian-noised inputs. Sparse regularization is additionally imposed on latent feature representations, guiding the network to learn compact and noise-robust priors. The complete architecture is jointly fine-tuned on multi-organ ULDCT datasets, enabling a single unified model to perform denoising across diverse anatomical regions. Extensive experiments validate that our proposed method achieves state-of-the?art performance and consistently surpasses competing ULDCT methods across all mul
Read the full abstract
General Mathematics
Sharp Total-Support Uncertainty for the Standard Complete Set of Mutually Unbiased Bases
We determine the sharp total-support uncertainty for the standard complete set of mutually unbiased bases in every prime dimension at least five, proving that the minimum is \(T_s(p)=p^2-p+2\). This resolves all prime dimensions left open by Fiorentino and Weigert, beginning with dimension eleven, and gives a uniform analytic determination of the optimal lower bound in dimension seven, where the earlier determination was partly numerical. The main new ingredient is a sharp zero-count theorem for arbitrary complex Fourier data supported on the finite parabola. A cyclotomic reduction and truncated local expansion produce a first nonzero finite-field jet that is affine in the transverse variable, reducing the planar zero count to a one-dimensional multiplicity problem. We then classify all extremizing projective states and derive an exact enumeration formula. In dimension seven, this gives analytic four-point extremizers and exactly \(12348\) such rays, completing the analytic description and count left open in the earlier work.
Read the full abstract
Machine Learning
AttnFuse: A Composable DSL for Compiling Attentions to Fused GPU Kernels
Modern AI systems are built on the Transformer architecture, whose core operation, attention, accounts for the majority of computation and memory cost. Researchers continually propose new attention variants to improve quality, efficiency, or context length, but each variant currently requires expert-written GPU code to run at usable speeds. PyTorch's recent flex\_attention lets researchers describe custom attention patterns in Python and compile them to fused kernels, but its design is limited to modifications applied after the central matrix multiplication, excluding Rotary Position Embedding (RoPE), the positional encoding used by every major LLM. We introduce AttnFuse, a small DSL for attention that makes pre-multiplication transformations like RoPE first-class operations. Researchers compose ten high-level building blocks to describe a variant, and AttnFuse's compiler emits a single fused GPU kernel for the entire computation. On an RTX 3090, AttnFuse achieves a 2.10$\times$ speedup over flex\_attention on the RoPE+causal pattern. On an H100, it runs a full Llama-3-8B training step within 5% of PyTorch's hand-tuned backend. Our investigation reveals the Rotation Calculus: whether to fuse RoPE or apply it separately depends on the GPU's compute-to-bandwidth ratio, with a derived crossover that matches measurement. AttnFuse demonstrates that a small, attention-specific compiler can close the gap between flexible research code and production kernels.
Read the full abstract
Computation and Language
In the Blind: Building Pseudo-References for MT Evaluation
The WMT26 General MT task evaluates systems on 10 language pairs that have no human references (neither translated from scratch nor post-edited from MT output by humans). We describe how we built the pseudo-references for these pairs and six other language pairs (in which some forms of human references are available): seven models translate the 3,277 official documents under up to five prompt conditions, giving a total of 26 system-prompt combinations; then three reference-free quality estimation (QE) models score every candidate; and a per-document selector picks one translation, which GPT-5.5 post-edits where needed. Working without references exposed a failure mode of QE-guided selection: the metrics rank fluent output in the wrong language above correct translations. Adding a confidence-scaled language identification penalty to the score fusion drives the wrong-language count to zero, and the resulting selector still scores better on MetricX than the rank-fusion baseline it replaces. Since no references were available for these pairs while we were building them, we calibrate every selection decision on last year's WMT25 human judgments. The human evaluation, released after construction, shows the cost of getting selection wrong: our references stand with the strongest participating systems when the selector kept a frontier-model candidate, and fall up to 17 ESA points below them when it did not. We release the selection method and the provenance of every reference (https://github.com/surrey-nlp/PseudoRef)
Read the full abstract
Cosmology and Nongalactic Astrophysics
CHEX-MATE: X-ray surface brightness discontinuities across a representative cluster sample
We analyze the presence of X-ray surface brightness (SB) discontinuities, such as shocks and cold fronts, in the intracluster medium (ICM) of the 116 galaxy clusters from the CHEX-MATE sample. These features arise from accretion-related processes, including the merging of subclumps onto the main cluster halo and the sloshing of cold gas in a higher-entropy environment. We identify these structures in the XMM CHEX-MATE maps by examining both the SB residuals, obtained by subtracting a cluster model from the observations, and the gradient variations across X-ray images. We validate our method with an extensive analysis and comparison of the discontinuities described already in the literature. Our analysis reveals 66 discontinuities, 32 of which are newly detected, in 48 objects ($\sim$41%). We find that discontinuities in relaxed systems tend to be closer to the core, weaker, and less aligned with the global X-ray morphology compared to those in disturbed systems. Using wavelet-based temperature maps, we provide a preliminary classification of these discontinuities, proposing the occurrence of 8 putative shocks and 34 cold fronts. Furthermore, about 48% of the systems with extended radio emission show evidence of ICM discontinuities; conversely, $\sim$88% of the objects with detected discontinuities have an associated diffuse radio emission. This work also explores the limitations of using XMM data, characterized by relatively low spatial resolution, for detecting and analyzing such features, highlighting the challenges and potential biases introduced by instrumental constraints.
Read the full abstract
Statistical Methodology
Assumption-Lean Inference for Spectral Differential Network Analysis of High-Dimensional Time Series
Network analysis for multivariate time series is popular in many fields, from neuroscience to seismology. The inverse spectral density is a common choice for time series network analysis due to its representation of the frequency domain correlation between two variables after removing the best linear predictor of all other variables. In many applications, the goal is to study how these networks change across different conditions. For example, in neuroscience, one might be interested in how the brain connectivity network changes before and after stimulation. Towards this goal, we develop an inference framework based on a direct estimate of the difference in two high-dimensional inverse spectral densities. We develop a new Gaussian approximation error bound for any de-biased D-trace estimation procedure which is then leveraged to both inform optimal window sizes of Welch's estimators of the spectral density and establish asymptotic normality of our de-biased D-trace estimator. Moreover, we develop an efficient algorithm based on a generalized D-trace estimation procedure to overcome the computational complexity of high-dimensional inference. The method is illustrated on synthetic data experiments and on experiments with electroencephalography data.
Read the full abstract
Combinatorics
A lattice family with kissing numbers $\tau(\mathcal{L}_n) \ge e^{2 \sqrt{n}}$
For all prime powers $q\geq5$, we construct lattices $\mathcal{L}_q\subseteq\mathbb{Z}^q$ with kissing numbers \[ \tau(\mathcal{L}_q)\geq \left(\frac{1}{2\pi e^2}+o(1)\right)\sqrt{q}\,e^{2\sqrt{q}}. \] The same asymptotic bound holds on a set of integer dimensions of natural density $1$, and in every sufficiently large integer dimension $n$ with an additional factor $e^{-\tfrac{1}{2}n^{1/40}}$. The construction is an extension of a previous construction by Bennett-Peikert based on Reed-Solomon codes.
Read the full abstract
Geometric Topology
Stratification and realization of singularity data on the Loch Ness monster
The classical theory of compact translation surfaces is organized through the stratification of moduli spaces according to the orders of the zeros of the associated Abelian differentials. In the setting of tame translation surfaces of infinite topological type, no analogous notion of strata has been systematically developed. In this article, we introduce a stratification theory for tame translation surfaces based on their singularity data. To each non-compact tame translation surface, we associate a sequence recording the cardinalities of finite-angle cone singularities of each possible cone angle together with the cardinality of the set of infinite-angle singularities. We call this sequence the singularity data of the surface. We study the realization problem for singularity data on the Loch Ness monster. Our first main result shows that there is no tame translation structure on the Loch Ness monster having only a finite number of finite cone angle singularities. We then prove that this is the only obstruction: every singularity data not excluded by the previous result is realized by a tame translation structure on the Loch Ness monster. As a consequence, we obtain a complete characterization of the strata of tame translation structures on this surface in terms of their singularity data. These results provide the first realization theorem for strata of non-compact tame translation surfaces and reveal a strong interaction between singularity data and the ends space.
Read the full abstract
Robotics
From Vision to Harvest: Benchmarking Vision-Language Models for Multi-Arm Robotic Fruit Harvesting
Multi-arm robotic harvesting offers a promising path to improve harvesting efficiency and reduce reliance on manual labor. However, practical deployment remains challenging because the system must generalize across diverse environments while efficiently coordinating multiple arms in a shared workspace. Existing methods often require substantial data collection in target environments or rely on simplifying assumptions that limit planning quality. In this work, we introduce the first comprehensive benchmark for evaluating pretrained Vision-Language Models (VLMs) on zero-shot multi-arm fruit harvesting planning. Our benchmark uses real-world apple and citrus orchard images and compares a VLM-based planning pipeline with a traditional perception-and-planning pipeline. The VLM pipeline directly generates harvesting sequences and waypoints for each arm, while a lightweight trajectory verifier checks for collisions. Our results show that frontier VLMs can generate effective multi-arm harvesting plans zero-shot, but a practical deployment remains limited by accurate 3D waypoint generation and collision-aware coordination. These results highlight both the promise and current limitations of pretrained VLMs for multi-arm robotic harvesting.
Read the full abstract
Signal Processing
Learning-Free Hierarchical Joint Estimation of AoA, Pointing Error, Receiver Jitter, and Turbulence in Multi-Aperture FSO Systems
This paper proposes a learning free hierarchical estimator for jointly recovering the angle of arrival (AoA), transmitter pointing error, receiver induced jitter, and per aperture turbulence coefficients in a multi aperture free space optical system. The proposed method exploits the distinct spatial signatures contained in quad photodetector measurements. First, normalized quadrant imbalance ratios provide an approximate closed form AoA estimate. Next, AoA compensated lens powers are transformed into a log linear regression model for non iterative pointing error estimation. Finally, receiver jitter and turbulence coefficients are directly reconstructed after compensating for the estimated angular losses. The method requires neither neural network training nor exhaustive multidimensional search and has computational complexity linear in the number of lenses. Monte Carlo results demonstrate robust AoA estimation under Gamma Gamma turbulence, reveal turbulence induced error floors in OLS based pointing and jitter estimation, and show improved turbulence reconstruction with larger arrays and explicit pointing compensation.