Peer Revenue Now accepting positions

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

All of arXiv

Monday 9 November

Dynamical Systems

Periodic vortex dipoles near closed geodesics on Riemannian surfaces

We study the motion of a pair $(p_1,p_2)\in\Sigma\times\Sigma$ of point vortices with equal magnitude and opposite signs on a closed oriented Riemannian surface $(\Sigma,g)$. Given a closed geodesic that satisfies a non-degeneracy condition we construct a one-parameter family of periodic solutions $(p_1^\varepsilon,p_2^\varepsilon)$, $0<\varepsilon<\varepsilon_0$, of the Hamiltonian dynamical system for point vortex pairs whose trajectories converge towards the geodesic curve. The two vortices lie on opposite sides of the geodesic, are separated of order $\varepsilon$ and have period $2\pi\varepsilon$. The proof is based on a differentiated expansion of the Riemannian distance in Fermi coordinates, an augmented Lyapunov–Schmidt reduction, and an intricate contraction argument in scaled variables. The reduced normal equation is governed by the Jacobi operator of the closed geodesic, and the first order correction to the normal center of the dipole is characterized by an inhomogeneous Jacobi equation involving the curvature and the regular part of the Green function of the Laplace-Beltrami operator.
Read the full abstract

Audio and Speech Processing

Full-sphere emission and reception directivities of seven echolocating bat species

Echolocating bats shape their calls and the returning echoes with their noseleaf or mouth, head and pinnae. Understanding these spectrospatial directivity patterns is central to bat echolocation. We describe a dataset of simulated far-field emission and reception directivities of seven echolocating bat species from three families (Phyllostomidae, Rhinolophidae and Vespertilionidae). For each species, the head of one preserved specimen was scanned with micro-computed tomography and converted into a closed surface mesh, on which the responses of both ears and of the emitter were computed with a boundary element method or, for the two mouth-emitting species, with an analytical piston model. The complex-valued responses cover the full sphere at 2.5 deg resolution from 20 to 160 kHz in 0.5 kHz steps. They come with the meshes, the landmarks of ears, nostrils and mouth, alternative source positions and skeletons for posing the pinnae and noseleaves, all in one frame, and with Python and MATLAB code. The technical validation covers solver verification, mesh convergence, comparisons with earlier simulations and measurements, and sensitivity analyses.
Read the full abstract

Optimization and Control

KL-Regularized Policy Iteration for Adaptive LQR: Gain Updates and Probing from One Policy Improvement

Policy-gradient methods for adaptive linear quadratic control improve the feedback gain of an unknown plant online by gradient-type steps computed from closed-loop data. A recent framework of such methods, policy-gradient adaptive control (PGAC), takes one gradient-type step per sample, which its indirect form computes from an estimated model and its direct form directly from the data. Because data generated by a fixed linear feedback cannot identify the plant, PGAC adds a probing perturbation to the input and leaves its covariance as a design parameter. We study the adaptive LQR problem through a KL-regularized formulation, in which the probing is the randomness of a Gaussian policy kept close to a Gaussian reference policy, so that the cost of the probing is part of the objective. One policy improvement of this formulation gives a gain update together with a probing covariance. We show that, for suitable references centred at the current gain, this gain update is exactly PGAC's Gauss-Newton or natural-gradient update with any step size below a bound. We repeat this policy improvement, each time against a tempered copy of the current policy, and refer to this scheme as KL-regularized policy iteration (KLPI). On a known plant and with tempering, the cost of KLPI's gains converges linearly to the optimal cost, and the probing covariance keeps a floor proportional to the regularization weight. On an estimated plant, under conditions like PGAC's (small updates and sufficiently informative data), the closed loop stays stable, the cost of the gains converges linearly up to a bias, and the probing covariance again keeps this floor. In simulations at equal probing cost, the shape of KLPI's probing covariance, which PGAC can adopt, lowers the extra cost caused by suboptimal gains on nearly all tested plants, whereas KLPI's use of past iterations in its gain update changes this cost little.
Read the full abstract

All authors Ajinkya Bhole, Mohammad Mahmoudi Filabadi, Guillaume Crevecoeur, Tom Lefebvre

Game Theory

Computing EF1 Allocations under Budget Constraints in Polynomial Time

We study the fair allocation of indivisible items to agents with budget constraints and additive valuations. Each agent compares its own bundle with the subsets of another bundle that fit its budget. An allocation is envy-free up to one item (EF1) if, for every agent and every nonempty subset under comparison, removing an item of maximum value to that agent eliminates its envy. We require EF1 both between agents and towards the charity, which contains all unallocated items. EF1 allocations exist, but computing them in polynomial time has remained open even for homogeneous valuations. We resolve this problem with a polynomial-time algorithm for EF1 allocations under heterogeneous valuations. We also prove that detecting an EF1 violation is NP-complete. Equivalently, checking whether a given allocation is EF1 is coNP-complete. Finally, we give polynomial-time algorithms for envy-freeness up to any item (EFX) both when each agent's valuation is proportional to item size and when each agent has a fixed number of distinct positive item values. As an application, for homogeneous valuations and any $\varepsilon>0$, we obtain an EF1 allocation with at least a $(1/2-\varepsilon)$ fraction of the maximum social welfare in time polynomial in the input length and $1/\varepsilon$.
Read the full abstract

Computer Vision

Plant Orbiter: 3D Plant Reconstruction and Generation with Video Diffusion Orbit Cameras and 3D Gaussian Splatting

Modern digital phenotyping pipelines require the precise measurement of a range of 3D morphological plant traits. The accurate, high throughput acquisition of these traits helps breeders and researchers accelerate the development of new varieties, techniques and products. Recent advances in 3D technologies have enabled a rapid transition from 2D to 3D pipelines, but high throughput acquisition of digital 3D twins of plants remains a significant challenge due to complex morphology, thin structures and occlusions. In this paper we present Plant Orbiter, a new approach to generating 3D plants that leverages recent advances in video diffusion to produce high quality 3D Gaussian Splats of plants from a single image. We show that our method more accurately recreates plant morphology than other state-of-the-art methods, achieving the highest 3D F-score against all baselines, while being more practical for high throughput pipelines due to easier alignment and reduced cost of camera placement. Combined with a 2D diffusion model, we show that our method also achieves state-of-the-art results in the generation of 3D plant models from a text prompt.
Read the full abstract

All authors Zane K. J. Hartley, Yong En Kok, Lewis A. G. Stuart, Andrew P. French, Michael P. Pound

Quantum Physics

Quantum Robustness via Sandwiched Rényi Divergences

If a quantum state is known only approximately, how much can the expectation of an observable change? We answer with the worst-case expectation over all states within a sandwiched Rényi divergence budget of a reference state, for every order $\alpha \in [1/2,\infty]$. Convex duality gives this value as a one-dimensional optimization over a sandwiched Schatten norm, running from a fidelity risk at $\alpha = 1/2$, through the entropic risk at $\alpha = 1$, to a quantum conditional value-at-risk at $\alpha = \infty$. The main result is a classicalization theorem: the quantum problem coincides with the classical one for a probability measure $\omega_\alpha$ on the spectral interval of the observable, constructed from Heinävaara's tracial joint spectral measure and given by Stahl's theorem at $\alpha = 1$. The Hoeffding, Bernstein and Markov–Krein bounds of classical probability thereby apply to the quantum problem, and yield Pinsker-type inequalities with Bernstein-type refinements. The measure $\omega_\alpha$ is the Born measure only at $\alpha = 1/2$; for $\alpha > 1/2$, noncommutativity of the observable with the state strictly lowers its variance below the Born variance, and with it every variance-sensitive bound: a quantum deflation.
Read the full abstract

Software Engineering

Inference Validation for Regulated AI: A Governed Decision Pipeline for LLM Serving

Deploying large language models in regulated industries requires more than picking the fastest inference engine: it requires evidence that a configuration meets contractual and legal obligations before it serves a single customer request. We describe Runway, an internal validation harness built at MasterControl that builds a governed decision pipeline around NVIDIA AIPerf. The key insight is separating concerns: AIPerf provides the trusted measurement, while Runway provides the experiment design that comes before it and the compliance, cost, and audit decisions that come after. We describe the architecture and the storage model that makes caching and audit one mechanism. We report three findings from Runway sweeps: TensorRT-LLM and vLLM rankings invert with concurrency; a newer, pricier GPU instance can be the cheaper one per request; and speculative decoding behaves differently across engines under load, so engine configuration can outweigh engine choice.
Read the full abstract

Materials Science

Finite temperature and pressure structural relaxations including zero-point motion and anharmonicity

At finite temperature and pressure, the equilibrium structure of a crystal minimizes the Gibbs free energy rather than the static Born-Oppenheimer energy. We present WARP (Wyckoff-constrained Anharmonic Relaxations at Pressure and temperature), a symmetry-constrained, variable-cell relaxation based on the stochastic temperature-dependent effective potential method, including anharmonicity and zero-point motion. Every position update is projected onto Wyckoff positions and every strain onto symmetric strains, so a relaxation can raise the symmetry but never lower it. For aluminum, the thermal expansion follows measurements at cryogenic temperatures, including the quantum plateau below 50 K. For lanthanum hydrides at megabar pressures, equations of state agree with phonon-based and path-integral calculations, and LaH$_{10}$ relaxes from its rhombohedral static minimum to the observed cubic phase, also with a machine-learned potential fine-tuned on 150 supercells. For brucite, Mg(OD)$_2$, lattice parameters, bond distances and thermal expansion agree with neutron diffraction, and the O-D isotope effect is captured.
Read the full abstract

Soft Condensed Matter

Catalysis-induced phase separation of enzymes via diffusivity contrast

Catalysis can drive phase separation by coupling local chemical composition to molecular transport, even in the absence of attractive interactions. Here, we demonstrate such catalysis-induced phase separation in a microscopic particle-based model of enzymes catalyzing the interconversion of substrate and product molecules with different diffusivities. We derive an analytical instability criterion in the dilute limit, showing that a homogeneous mixture becomes unstable when the nonequilibrium chemical drive and the diffusivity contrast are sufficiently large, and the substrate and product concentration exceeds a threshold set by the enzyme volume. Brownian-dynamics simulations confirm the predicted instability and reveal macroscopic coexistence of enzyme-rich and enzyme-poor phases, including in regimes where equilibrium depletion interactions cannot account for demixing. Finally, we show that, upon eliminating the fast chemical degrees of freedom, the reactive ternary mixture maps onto an effective binary mixture nonreciprocal interactions. Our results establish an explicit microscopic mechanism for catalysis-induced phase separation and connect chemically driven phase separation to the physics of motility-regulated and nonreciprocal matter.
Read the full abstract

All authors Marie Jardat, Jacques Fries, Vincent Dahirel, Jaime Agudo-Canalejo, Pierre Illien

Astrophysics of Galaxies

The Cool Circumgalactic Echo of the Stellar-to-Halo Mass Relation

The stellar-to-halo mass relation (SHMR), a key prediction of galaxy formation models, peaks at $M_* \sim 3 \times 10^{10}\,M_\odot$. This scale is often attributed to circumgalactic baryonic feedback processes, however, direct observational evidence linking the two remains scarce. Using Dark Energy Spectroscopic Instrument (DESI) galaxy–quasar pairs, we map the cool circumgalactic medium (CGM) using Mg II $\lambda\lambda2796, 2803$ absorption with stacking technique across nearly three decades in stellar mass. The radial Mg II absorption equivalent-width profile, $W_{\rm MgII} \propto D^k$, becomes progressively shallower with increasing stellar mass, following $k = 0.70\log(M_*/M_\odot)-9.12$. Crucially, we find that the $W_{\rm MgII}$ in the CGM scales tightly with the SHMR. The $W_{\rm MgII}$ peaks at the same characteristic mass as the SHMR, coinciding with a transition of the cool CGM from super-virial to sub-virial kinematics. Low-mass halos show excess non-gravitational motions consistent with recycled accretion and feedback-driven circulation, whereas massive halos show halo-centric heating that may maintain star formation quenching. This study bridges the gap between cool CGM and SHMR, showing that peak star formation efficiency coincides with the most abundant cool CGM reservoirs.
Read the full abstract

All authors Zeyu Chen, Enci Wang, Huiyuan Wang, Houjun Mo, Luis C. Ho, Haoran Yu, Zhijie Qu, Hu Zou, Siwei Zou, Zhicheng He, Cheqiu Lyu, Chengyu Ma, Weiwen Kong

Materials Science

Out-of-plane interference effects in the angle-resolved photoemission of epitaxial graphene

The epitaxial growth of monolayer graphene by Si sublimation of 6H-SiC(0001) is one of the most effective ways of producing large-scale monocrystalline graphene samples of high quality. However, graphene is left interacting with a reconstructed carbon buffer layer known as zerolayer graphene (ZLG) and the SiC substrate underneath. Beyond the mere creation of Dirac cone replicas, the effect of this interaction on photoemission measurements is poorly understood. We utilized synchrotron radiation for photon-energy-dependent angle-resolved photoemission spectroscopy (ARPES) to investigate the complex final-state interactions of photoelectrons within the heterostructure of monolayer graphene, ZLG, and SiC. We identify multiple resonances in the photoemission intensity of the Dirac cone as a function of photon energy that originate from interlayer interference, confirmed with a multiple scattering model. We also demonstrate for the first time that the relative intensity of the replica bands induced by the $(6\sqrt{3} \times 6\sqrt{3})\text{R}30^\circ$ reconstruction scales with the electron kinetic energy. Finally, we resolved four nondispersing electronic states, which originate from the ZLG. Our findings show that the three-dimensional (3D) structure of SiC and ZLG manifests itself in the photoemission from two-dimensional (2D) epitaxial graphene. By revealing how 3D substrate architecture dictates the photoemission of a 2D layer, this work establishes a crucial framework for investigating the electronic structure of complex heterostructures via ARPES.
Read the full abstract

All authors E. Favorite, P. Krüger, Ph. Rosenzweig, R. Joshi, S. Williams, B. Matta, K. Küster, M. Hashimoto, U. Starke, H. Nakamura

Machine Learning

$\Sigma$-LeWM: Learning World Models via S-Information for Robot Control

World models have increasingly been applied as a model-based control method for robotics tasks such as manipulation and locomotion. While early world model architectures relied on decoders to prevent mode collapse and learn a predictable latent space, recent works on world models, such as Joint Embedding Predictive Architecture or LeWorldModel, increasingly aim to learn embeddings without a grounding decoder, utilising only a latent forward dynamics predictor and a stabilising entropy term. In this paper, we set out to demonstrate that the further introduction of two additional latent losses to LeWorldModel, a backward dynamics predictor and an inverse dynamics action predictor, enforces further structure on the latent embedding, leading to further gains in both performance and planning efficiency. These are critical properties for improving the applicability of such methods to real-time model-based control of real-world robots. We demonstrate these gains on standard world model benchmarks and two simulated robot manipulation tasks.
Read the full abstract

All authors Xinrui Zu, Michael Przystupa, Zhao Yang, Shujian Yu, Kevin Sebastian Luck

Information Retrieval

NIFT-SQL: Customer-Supervised Table Scoring for Enterprise Text-to-SQL Retrieval

Enterprise text-to-SQL starts with finding the tables a question needs, which is hard when a schema spans several databases and hundreds of cryptic tables. We ask whether models trained separately on each customer's example questions, reading only table names, column names, and declared keys, improve table retrieval. NIFT-SQL (Necessity-Informed Fusion for Text-to-SQL) fuses hybrid search over a retriever fine-tuned on the customer's questions with a fine-tuned 560M cross-encoder that scores every table for whether the query needs it, through a logistic regression fitted on held-out questions. On BEAVER, LiveSQLBench, and LiveSQLBench-Large (244 to 971 tables; four splits, three trainings), NIFT-SQL places every reference table in the top 15 for 69.2%, 89.5%, and 85.6% of test questions, against 30.4%, 41.0%, and 21.8% for hybrid search and 61.1%, 81.7%, and 70.4% for the fine-tuned retriever with the same labels. With off-the-shelf search, the cross-encoder without fine-tuning is 21 points lower. A zero-shot 8B reranker needs no labels and, routed by database, shows no resolved complete@15 difference on two collections; adding its scores as two fusion features gives a high-accuracy mode that reaches 72.4%, 92.5%, and 88.6%, ahead of the validation-selected zero-shot reranker variant on two collections, with no resolved difference on the third. Rank fusion of the two trained parts, grouped by database, shows no resolved difference from the learned fusion. Scored once on a newer BEAVER release, NIFT-SQL beats search and leads the label-matched retriever at top 8 on the synthetic test, but trails it at top 15 on real questions.
Read the full abstract

Lattice Field Theory

Unpolarized proton PDF with physical quark masses and large momenta

We present a lattice quantum chromodynamics (QCD) calculation of the proton isovector ($u-d$) unpolarized parton distribution function (PDF) at physical sea and valence quark masses, using Coulomb-gauge-fixed quark bilinears within the large-momentum effective theory framework. The calculation is performed on a (2+1)-flavor ensemble with a lattice spacing of $a=0.076$ fm and proton momenta up to $2.3$ GeV. We combine kinematically enhanced nucleon interpolating operators, the Coulomb-gauge method, and high-statistics measurements to improve the signal quality at large momenta and quark separations. We use the static regularization-independent momentum subtraction and hybrid schemes to renormalize the bare matrix elements. The light-cone PDF in the $\overline{\rm MS}$ scheme is obtained from the quasi-PDF through next-to-leading-order (NLO) perturbative matching supplemented by next-to-leading-logarithmic renormalization-group resummation. With reduced uncertainties, the resulting PDF is consistent with our previous calculation on a similar lattice ensemble with $a=0.06$~fm and a valence pion mass of 300 MeV. Notably, despite the reduced uncertainties, the agreement with the phenomenological PDF persists over the longitudinal momentum fraction range $0.25\lesssim|x|\lesssim0.75$, where the lattice calculation is expected to be reliable. This agreement is further supported by a comparison in coordinate space, where our short-distance lattice matrix elements are consistent with those obtained from the phenomenological PDF through NLO matching. In particular, we observe close agreement in the imaginary part for the first time. Together, these results represent an encouraging step toward precision benchmarking of lattice QCD calculations of PDFs.
Read the full abstract

All authors Xiang Gao, Andrew Hanlon, Jinchen He, Xiangyu Jiang, Swagato Mukherjee, Peter Petreczky, Qi Shi, Yong Zhao

Machine Learning

KiteNorm: Variance Regularisation for Stable and Scalable Post-LN Transformers

Post-LayerNorm Transformers have seen limited practical adoption due to enduring difficulties in scaling them to large depths. While prior research has focused on stabilising Post-LayerNorm by improving conditioning at initialisation, stability often deteriorates when models are trained with large learning rates, forcing additional compromises. In the decoder-only settings we study, existing Post-LayerNorm methods fail to outperform strong Pre-LayerNorm baselines. We propose KiteNorm, a post-norm training method designed to break this trend by constraining hidden-state variance. This approach is motivated by a bound on the LayerNorm Jacobian suggesting that variance growth contributes to gradient vanishing, alongside empirical evidence linking it to rank collapse. KiteNorm combines residual scaling and variance regularisation to maintain stability throughout training, supported by scalarised double-normalisation that provides separately learnable scales for the skip and residual branches. Across decoder-only Transformers up to 1.3B parameters, KiteNorm improves training stability and outperforms leading baselines at their optimal learning rates, with scaling laws favouring KiteNorm across depth, width, batch size, and training budget. Ablations show that each component is necessary for the full stability and performance gains.
Read the full abstract

All authors Leon A. Trochelmann, Sajad Movahedi, Shiwei Liu, Antonio Orvieto

Machine Learning

Safe Reinforcement Learning for Multi-Turn Clinical Agents: When Dense Rewards Beat Outcome Rewards

Reinforcement learning with verifiable rewards is increasingly used to improve language model reasoning, yet its effects on sequential decision procedures remain poorly understood. We study this problem in clinical multi-turn agents that must decide what information to request, when to stop gathering evidence, and which action to take. We introduce three controlled environments: medication refill triage, acute diagnostic work-up, and dosing titration, each with hidden clinical states, query costs, exact adaptive oracles, and deterministic outcome verifiers. Across these tasks, outcome-only group relative policy optimization often improves aggregate reward while producing undesirable procedures, including premature commitment, one-question collapse, and failures on rare high-risk cases. In diagnostics, assigning dense credit to questions that reduce uncertainty improves held-out value and moderate-case support across three training seeds relative to outcome-only reinforcement learning, without the high-risk seed collapse observed under outcome-only training. Separating exploration from exploitation further increases high-risk support at its best checkpoint, but at the cost of excessive information gathering on low-acuity cases. Additional experiments show that the optimal strength of process credit is task-dependent; that after a supervised warmup that already gathers carefully, reinforcement learning raises aggregate reward by trimming questions while degrading high-risk performance; and that reinforcement learning does not reliably repair rare behavior gaps inherited from biased supervised data. Verifiable final outcomes are therefore insufficient for training reliable multi-turn agents: reward design must account for the procedure used to reach an answer, and evaluation must distinguish adaptive information gathering from reward-exploiting shortcuts.
Read the full abstract

Computer Vision

OptiCover: A Scalable and Multi-Surface Framework for Cold-Start Visual Optimization under Semantic Constraints

Selecting an effective cover photo is a critical yet understudied problem in visual merchandising for e-commerce marketplaces. On Airbnb, hosts upload many photos depicting different rooms, amenities, and points of interest, but must manually choose a single static cover photo that is reused across heterogeneous product surfaces. This creates substantial challenges: over 90% of listing photos have no engagement labels, leading to a severe cold-start regime; guest preferences vary across surfaces and search intents, and visually captivating photos may be semantically misaligned with the underlying listing. At Airbnb scale, this corresponds to hundreds of millions of photos and hundreds of thousands of new uploads per day. We propose OptiCover, a scalable framework for cold-start visual optimization under semantic constraints and multi-surface engagement signals. The framework combines (i) a photo candidate generator that identifies candidate photos through various proxies, (ii) lightweight vision heads applied to frozen CLIP embeddings for large-scale semantic scoring, (iii) a cross-modal grounding validator based on a vision-language model (VLM) that enforces semantic representativeness, and (iv) an Exploration Logging Module (ELM) that orchestrates the collection of randomized engagement data across heterogeneous product surfaces and contextual search intents. We evaluate OptiCover through extensive offline studies and multiple online experiments across HomePage, and Search surfaces. OptiCover is now deployed in production and performs tens of millions of photo swaps daily. Online results show a 1.5% engagement lift on the HomePage and a 0.35% increase in bookers on Search, generating millions of dollars in incremental revenue. Our findings suggest that action-space shaping and semantic constraints are primary drivers of performance.
Read the full abstract

All authors Guillaume Guy, Mihajlo Grbovic, Joe Mills, Selen Uguroglu, Liwei He, Pratiksha Kadam, Nalini Kartha, Qian Aria Li

Dynamical Systems

Residual-Based Spectral Submanifold Reduction of Constrained Multibody Systems on Matrix Lie Groups

The parameterization method yields efficient and accurate reduced-order models for nonlinear dynamical systems, but existing procedures that compute them assume polynomial vector fields on vector spaces. Constrained multibody systems instead obey differential-algebraic equations on matrix Lie groups such as $\mathbb{R}^3 \times SO(3)$ and $SE(3)$. Minimal parameterizations of these groups lead to non-polynomial equations that must be approximated or recast before reduction, and current implementations require the nonlinearities as explicit multilinear maps, which are cumbersome to construct for assembled multibody systems. This work develops a parameterization framework for constrained multibody systems on matrix Lie groups. Rigid and floating frame of reference bodies use rotation-matrix components or Euler parameters, while geometrically exact beam elements use an Euler-Rodrigues representation that eliminates the logarithmic map and expresses the exponential map polynomially, retaining screw-motion interpolation. The reduced-order models satisfy linear joint constraints exactly, and a rotation-preserving parameterization keeps the configuration on the Lie group to machine precision. The parameterization method is formulated directly in terms of the assembled residual, so that only residual evaluations and their derivatives are required. Truncated Taylor arithmetic delivers all Taylor coefficients of a given order from a single residual evaluation, and a non-intrusive coupling to codes with real-valued residuals is demonstrated with the multibody code Exudyn. Numerical examples show accurate prediction of backbone curves, forced response curves, and internal resonances at substantially reduced computational cost.
Read the full abstract

Machine Learning

Scaling Taxonomy-Free Image Labeling with Generative AI: A Multimodal

Providing real-time multimodal insights into images is challenging because it requires low latency, high description accuracy, and scalable solutions that are both cost-effective and efficient. For example, given a new image of an outfit, we would like to generate insights that the image depicts "earth tones" and "linen pants" and allow users to explore similar ideas along these themes. Traditional strategies to generate such insights would typically involve building taxonomies producing a golden set labeled by humans, training models on these datasets, and deploying these models for online inference services. Such workflows are constrained by fixed taxonomy, high labeling cost, limited labeling coverage in sample space, and human errors and expertise gaps in labelers. In addition, these methods are expensive and cumbersome to iterate over time. Recent trends have started exploring generative AI models for online insights, but this method suffers from long latency and prohibitive costs to support a high volume of requests or large amounts of offline labeling needs. We present Generative Labeling On Clusters (GLOC) as a solution to provide highly scalable, cost-effective, low latency, and high accuracy online image insight services. GLOC leverages highly optimized image embeddings to form image clusters that are semantically close to each other for the purpose of the type of insights, and summarizes the insights at the cluster level, which effectively generalizes to members in the cluster. At Pinterest, GLOC powers image descriptors and pivot query suggestions for Hybrid Search, which is a multimodal search service enabling users to search with both an image and a text query. Our offline and online A/B tests demonstrate that GLOC significantly outperforms previous methods, delivering improvements in quality and efficiency.
Read the full abstract

All authors Zhenjie Zhang, Sainan Chen, Kshiteesh Hegde, Mengze Li, Pouya Rezazadeh Kalehbasti, Munachim Ezema, Karina Sobhani

Number Theory

A relative trace formula for the orthogonal Gan-Gross-Prasad model and the proof of its fundamental lemma

In this paper, we propose a comparison of relative trace formulas to study the Gan–Gross–Prasad conjecture. We unfold the relative trace formulas, match their regular orbits, and formulate the corresponding fundamental lemma and smooth transfer conjecture. Finally, we prove the fundamental lemma for the unit element of the Hecke algebra.
Read the full abstract

Representation Theory

Classification of quasi-lisse simple affine vertex algebras

We classify all noncritical levels $k$ for which the simple affine vertex algebra $L_k(\mathfrak{g})$ is quasi-lisse, for every finite-dimensional complex simple Lie algebra $\mathfrak{g}$. We prove that Fiebig's equivalence between the corresponding noncritical blocks of category $\mathcal{O}$ identifies the relevant parabolic Wakimoto modules and the homomorphisms from the corresponding universal affine vertex algebras. Combining this compatibility with a characterization of semisimple points in associated varieties, we establish that for coprime positive integers $p,q$, $L_{-h^\vee+p/q}(\mathfrak{g})$ is quasi-lisse if and only if $L_{-h^\vee+p/\gcd(q,r)}(\mathfrak{g})$ is quasi-lisse, where $r$ is the lacing number of $\mathfrak{g}$. The classification at the resulting levels is obtained using Drinfeld–Sokolov reduction, singular vectors and relations in the reductions of the maximal ideals of $V^k(\mathfrak{g})$. We also determine the associated varieties at the quasi-lisse levels for all $\mathfrak{g}$ of rank at most six and for $E_7$ and $E_8$.
Read the full abstract

Dynamical Systems

Hidden frequencies in topological dynamical systems

We show for certain skew products built over any infinite minimal nilsystem, the frequencies present in the nilsystem are not always enough to govern recurrence for the whole system. This strengthens and generalizes a theorem of Glasscock, Koutsogiannis, and Richter.

Functional Analysis

On Solvability of Irregular Infinite-Dimensional Differential-Algebraic Equations

This paper investigates the solvability of infinite-dimensional differential-algebraic equations (DAEs) on Hilbert spaces. We focus on irregular systems for which the operator pencil is not surjective onto the full state space, but only onto a subspace determined by the algebraic constraints. We establish a Lumer-Phillips-type generation result for this class of systems, showing that dissipativity together with a partial surjectivity condition yields a contraction semigroup on the dynamic subspace. The resulting wellposedness conditions are compared with existing solvability results. We further extend the analysis to systems in which hidden algebraic constraints restrict the dynamics to a proper closed subspace of the dynamical subspace. Under an additional structural condition, we show that the resulting constrained system is $E$-radial on suitably chosen spaces. We also demonstrate that semigroup generation does not, in general, imply $E$-radiality on the original state spaces. This provides a Weierstraß-type decomposition useful for controller design. The framework is illustrated by some examples.
Read the full abstract

All authors Ala' Alalabi, Lorena Bociu, Birgit Jacob, Kirsten Morris

Mesoscale and Nanoscale Physics

Spin Shuttling Through Quantum-Dot Based Chiral Molecular Bridges: Gauge Constraints, Holonomy, and Quantum Gates

Chiral molecular bridges between silicon quantum dots can provide novel two-qubit entangling resources via the chirality-induced spin-selectivity (CISS) effect. We evaluate this using Makhlin invariants. Modeling a phase-coherent bridge as an $\mathrm{SU}(2)$-valued hopping between two spins, a Schrieffer-Wolff reduction yields a twisted exchange interaction $J \mathbf{S}_1 \cdot \mathbf{R} \mathbf{S}_2$, whose Dzyaloshinskii-Moriya and symmetric anisotropic parts are fixed by a single bridge angle $\alpha$. Using the Kaplan-Shekhtman-Entin-Wohlman-Aharony (KSEA) gauge argument, we prove this interaction is locally equivalent to isotropic exchange; thus, a single chiral bond does not constitute a new entangling class. Breaking this gauge symmetry reveals genuine chiral utility: chiral exchange converts a uniform transverse field into a synthetic gradient, bringing the two-qubit propagator within $7\times10^{-4}$ of the CNOT class in invariant space 4.7 times faster than an equivalent orthogonal gradient. Shuttling a spin through bridges with non-collinear axes implements a non-Abelian $\mathrm{SU}(2)$ holonomy, yielding rotationally non-reciprocal transport and closed-loop geometric gates that suppress dynamical-phase noise to second order. To address the unsettled microscopic origin of CISS, we introduce a unified transport channel combining coherent rotation, a direction-reversing spin filter, and pointer-basis dephasing. This model obeys an exact $\tanh$ composition law for pumped polarization and a closed-form transverse decay law, yielding four observables that experimentally over-determine the bridge's coherent fraction $F_c$. Finally, we quantify silicon-specific hardware constraints, establishing the Landau-Zener valley-adiabaticity window ($E_v \geq 2.1\sqrt{\hbar v}$) and charge-noise limit ($\sigma_J/J \leq 3\%$) required for $99\%$ gate fidelity.
Read the full abstract

All authors Amitoj S. Miglani, Rajdeep R. Dwivedi, Balsant S. Tiwari, Vishvendra S. Poonia

Algebraic Geometry

Perfect loops of proper schemes

We study the perfect loop space $LX$ of a proper scheme $X$ over a local field $F$. We show that if $X$ contains a rational curve, then $LX$ is not ind-representable, even in a very weak sense. We also show that if $X$ admits a proper model $\mathfrak{X}$ over the integers $\mathcal{O}$ of $F$, whose geometric special fiber does not contain rational curves, then $L^+\mathfrak{X} \stackrel{\sim}{\to} LX$ is an isomorphism via the natural map, where $L^+\mathfrak{X}$ denotes the perfect positive loop space (jet scheme) of $\mathfrak{X}$. In particular, $LX$ is a scheme in this case.
Read the full abstract

Quantum Physics

No Exponential Quantum Advantage for Decoded Quantum Interferometry with Local Quantum Decoders

Decoded quantum interferometry (DQI) reduces optimization of sparse linear constraint problems, and sampling from their Gibbs measures, to decoding a linear code from a quantum state. We study DQI on random $D$-regular $k$-sparse instances with $k\ge3$, for which local decoders such as belief propagation with quantum messages (BPQM) are commonly used. Such decoders estimate a typical codeword symbol using only the qudits within logarithmic distance of it. Our main insight is that the thresholds of the decoding problem induced by DQI correspond to phase transitions of the Gibbs measure that are well studied in statistical physics: tree reconstruction for local decoders and, for Max-$k$-XORSAT, condensation for the optimal decoder. This insight allows us to prove two main results. First, we prove an information-theoretic obstruction: no local decoder can recover the codeword with high probability above the tree reconstruction threshold. Conversely, we show that a decoder based on BPQM reaches this threshold. Since the cavity method predicts that local Markov chains equilibrate rapidly up to the same threshold, we expect DQI with local decoders to give no exponential speedup for sampling and no advantage in approximate optimization with amplitude functions that depend only on the number of violated constraints. Second, we consider the optimal decoder (the pretty-good measurement, for which no efficient implementation is known) and show that it succeeds exactly below the condensation threshold. The condensation threshold thus provides a tight upper bound on the performance of DQI with these amplitude functions, providing a strict refinement of the limit that Jordan et al.\ derived from the Holevo bound. Our results point to efficient nonlocal decoders that succeed between the two thresholds as the only viable path to a quantum advantage with DQI on these instances.
Read the full abstract

All authors Brandon Augustino, Shouvanik Chakrabarti, Enrico Fontana, Dylan Herman, Guneykan Ozgul, Ruslan Shaydulin, Shreehari Sureshbabu, Jacob Watkins

Algebraic Geometry

Fourteenth-canonical birationality of threefolds with positive geometric genus

We prove that the pluricanonical map of a complex projective threefold of general type with canonical singularities and positive geometric genus is birational in every degree at least fourteen. The bound is sharp.

Computer Vision

Spaghetti Splatting: Differentiating Curvilinear Dirac Masses via Convolution

In production systems, 3D geometry is represented by primitives such as points, lines, and triangles, as rendering hardware is optimized to generate images from them very efficiently. Learning such primitives from images by volume rendering is challenging, as they are Dirac masses defined at infinitesimal locations in space. EWA splatting, Poisson reconstruction, and 3D Gaussian Splatting overcome this by convolving Dirac masses with a low-frequency kernel, a recipe that generalizes along two axes: the kernel and the primitive. As points and triangles have already received significant attention, we investigate this recipe for curvilinear primitives, and introduce Spaghetti Splatting. We approximate curves by polylines whose convolution converges to that of the continuous curve, and volume render each segment so that connected segments composite without overcounting at their joints. For fast training and inference, we linearize the camera projection per segment, which also accommodates kernels beyond Gaussians, such as Beta kernels. As a point is a curve of zero length, our representation is a strict superset of 3D Gaussian Splatting. It remains competitive with the state-of-the-art on general scenes, outperforms methods specialized to line-like geometry, natively supports anti-aliasing, and produces long connected strands that are directly amenable to physics simulation.
Read the full abstract

All authors George Shramko, Shrisudhan Govindarajan, Rudy Lam, Yizheng Xie, Daniel Rebain, David B. Lindell, Andrea Tagliasacchi

High Energy Physics (Theory)

Boundary Formula for Volume Response of Entanglement Wedge Polygons

The vacuum volume of an entanglement wedge polygon is topological. We determine its semiclassical quadratic response to time-symmetric coherent scalar matter in AdS$_3$. A contrast of six interval entropies measures a nonnegatively weighted energy integral, whereas volume depends on the unweighted integral. For negative mass squared above the stability bound, perturbations with the same quadratic coefficient of global energy can match all six vacuum entropies through second order yet have opposite volume responses. A continuum of narrowing quartets reconstructs the volume response with controlled noise. An independent CFT calculation for a dimension-$3/2$ primary reproduces an exact polygon edge formula.
Read the full abstract

Data Structures and Algorithms

Polynomial-Space Quantum Algorithms from Classical Branching: Colouring, Domination and Monotone Local Search

We give quantum algorithms for exponential-time problems on $n$ vertices or elements, with bounded error, polynomial space and no quantum random access memory. Our results are these. (1) 4-Colouring in quantum time $O(1.3032^n)$, from a polynomial-space branching algorithm with polynomial-space terminal algorithms; with the branching algorithm alone, $O(1.3100^n)$. The previous best bound in this model, to our knowledge, is $O(1.3231^n)$. (2) $k$-Colouring below the previous best bound for $k \in \{5, 7, 8, 9, 10, 11\}$ and $13 \le k \le 18$, from $O(1.4518^n)$ for $k = 5$ to $O(1.9233^n)$ for $k = 18$. (3) Dominating Set in $O(1.2235^n)$, against the previous $O(2^{n/2})$, Independent Dominating Set in $O(1.1583^n)$, and Edge Dominating Set and Minimum Maximal Matching in $O(1.1501^n)$. (4) A quantum form of monotone local search: if the extension problem has a backtracking tree with $c^k$ nodes up to polynomial factors, the problem is solved in quantum time $(2 - 1/c)^{n/2}$ up to polynomial factors. This gives Feedback Vertex Set in $O(1.3081^n)$, Subset and Group Feedback Vertex Set in $O(1.3229^n)$, Cluster Vertex Deletion in $O(1.2033^n)$, 3-Hitting Set in $O(1.2405^n)$, Proper Interval Vertex Deletion in $O(1.3541^n)$, Multicut on Trees in $O(1.1647^n)$ and Min-Ones $d$-SAT in $O^*((2-1/d)^{n/2})$. (5) Quantum exponential-time approximations below the quantum exact bounds, such as a 1.1-approximation for Vertex Cover in $O(1.0555^n)$ and for Feedback Vertex Set in $O(1.2161^n)$. The colouring exponents for $k \ge 6$ come from an evaluation of a known reduction recursion, computed in floating point with a margin.
Read the full abstract