Computation and Language
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Monday 9 November
Does grouping translation updates before speech synthesis change the measured completion gap between two translators? We compare StreamAtt with its fixed-audio-history baseline on ten English-German talks. For each translation, we keep the intended text and its availability times fixed and vary how updates are grouped for synthesis. Four-second grouping increases StreamAtt's mean completion lead from 106.6 to 229.1 seconds with XTTS-v2, but reduces it from 169.5 to 146.5 seconds with Piper. Both produce shorter audio; different changes in playback idle account for the opposite effects. These directions recur on a separately selected twenty-talk panel. In continuing CosyVoice 3 sessions, grouping instead leaves paired audio unchanged and delays completion on nineteen of twenty blocks. A speech-timing comparison therefore needs to specify how text enters synthesis, not just which synthesizer is used. Recognition and automatic word timing describe the outputs, but do not establish translation accuracy or intelligibility.
Number Theory
Horizontal estimates for Kloosterman sums under Hecke eigenvalue constraints
We prove weighted cancellation under small-value conditions on Hecke eigenvalues of finitely many non-CM holomorphic newforms. We also obtain many squarefree integers of each sign with fixed normalized amplitude, simultaneously small Hecke eigenvalues, and a fixed bound on the number of prime factors, uniformly in growing reduced residue classes. Our questions are motivated by Xi's noncorrelation conjectures and the mixed-moment estimates of Jiang and Lau. Their estimates imply that small Hecke eigenvalues can occur together with non-small Kloosterman values. Our result retains this property with a fixed prime-factor bound, both signs, and the residue conditions. Using OpenAI's zero-free half-plane theorem for Dirichlet $L$-functions, we give an example with at most $83$ prime factors, normalized amplitude at least $10^{-21}$, and moduli $M\le C^{11/200}$.
Read the full abstract
Logic
Dynamic Topological Completeness for $\mathbb R^2$
Dynamic topological logic (${\bf DTL}$) combines topological and temporal modalities to express asymptotic properties of dynamic systems on topological spaces. A dynamic topological model is a triple $\left <X,f,V\right >$, where $X$ is a topological space, $f:X\to X$ a continuous function and $V$ a truth valuation assigning subsets of $X$ to propositional variables. Valid formulas are those that are true in every model, independently of $X$ or $f$. A natural problem that arises is to identify the logics obtained on familiar spaces, such as $\mathbb R^n$. Slavnov showed that any satisfiable formula could be satisfied in some $\mathbb R^n$ for $n$ large enough, but the question of how the logic varies with $n$ remained open. In this paper we prove that any fragment of ${\bf DTL}$ that is complete for locally finite Kripke frames is complete for $\mathbb R^2$. This includes ${\bf DTL}^\bigcirc$; it also includes some larger fragments, such as ${\bf DTL_1}$, where “henceforth" may not appear in the scope of a topological operator. We show that satisfiability of any formula of our language in a locally finite Kripke frame implies satisfiability in $\mathbb R^2$ by constructing continuous, open maps from the plane into arbitrary locally finite Kripke frames, which give us a type of bisimulation. We also show that the results cannot be extended to arbitrary formulas of ${\bf DTL}$ by exhibiting a formula which is valid in $\mathbb R^2$ but not in arbitrary topological spaces.
Read the full abstract
Operator Algebras
Fixed-occurrence cases of the shuffle-word-embedding conjecture
For self-adjoint contractions in a finite tracial von Neumann algebra, binary shuffle traces with $m=1,2,4$ occurrences of the second variable admit simultaneous finite-dimensional approximation by commuting diagonal contractions. For every odd $m$, a commuting diagonal self-adjoint pair reproduces any prescribed finite set of shuffle traces exactly. When the first operator has finite spectrum and $m\geq3$ is odd, a weighted symmetric cycle reproduces all degrees simultaneously. We prove fourth-order trace-derivative positivity in finite tracial von Neumann algebras and obtain a positive measure representing the four-occurrence coefficients. For each even $m$, we characterize the contraction model property by positivity of an associated trace-derivative functional.
Read the full abstract
Optimization and Control
Quantitative stability of classical, entropic, and doubly entropic Wasserstein barycenters
In this paper, we show how the recent analysis of (sharp) strong convexity properties of the Kantorovich functional by the third author can be used to establish new global quantitative stability estimates for Wasserstein barycenters and their entropic regularizations with respect to perturbations of the population distribution. For classical Wasserstein barycenters, we improve the state-of-the-art Hölder continuity exponent with respect to the $\mathcal{W}_1$ metric from $1/6$ to $1/4$, while drastically weakening the assumptions on the population distribution. We also provide an example which shows that the exponent $1/4$ cannot be improved. Motivated by computational considerations, we also analyze regularizations of Wasserstein barycenters with outer and inner entropic penalizations. For outer entropic barycenters, we establish novel higher-order stability results and in particular give a quantitative bound on relative Fisher information. Another contribution of our paper is a strong convexity estimate for an entropic version of the Kantorovich functional via soft Legendre transforms, enabling us to extend our stability results to doubly entropic barycenters, as introduced by Chizat.
Read the full abstract
High Energy Astrophysical Phenomena
Long before plunge: Uncovering intermediate-mass black holes through their inspiralling compact object companions with LISA
The Laser Interferometer Space Antenna (LISA) will provide novel capability to discover and study intermediate-mass (~$10^2-10^5 M_\odot$) black holes at the centres of dwarf galaxies and in dense stellar environments. We systematically map out the LISA-detectable parameter space of binaries consisting of a low-mass compact object spiralling into an intermediate-mass black hole (mass ratio ~$10^{-5} - 10^{-2}$) on equatorial orbits with generic eccentricity. The depth of the LISA survey reaches a redshift ~$5$ ($0.1$) for binaries with a primary black hole mass $m_1$ ~$10^4 M_\odot$ ($10^3 M_\odot$). If the orbital companion is a white dwarf, which could be disrupted prior to plunge, LISA's survey depth is well matched to (and, in many cases, exceeds) the one of electro-magnetic surveys. Although the maximum distance at which a source can be detected depends weakly on the orbital eccentricity, the latter strongly affects the portion of the binary inspiral over which a source is detectable. The instrument's sensitivity time-volume varies by many orders of magnitude depending on the binary's component masses: it reaches ~$1 - 10^3 Gpc^3 yr$ (~$10^{-2} - 10^3 Gpc^3 yr$) for the more massive systems with $m_1 = 10^5 M_\odot$ ($10^4 M_\odot$) and decreases to ~$10^{-4} - 1 Gpc^3 yr$ for lighter primary black holes with $m_1 = 10^3 M_\odot$. Importantly, from tens of percent to the overwhelming majority of the total sensitivity time-volume (and therefore number of potential detections) comes from sources that do not plunge during the yr-long mission lifetime and are observed at an earlier stage of their inspiral. Despite the poorly known merger rates of this class of sources, our results show that LISA could identify from a handful to ~$10^3$ intermediate-mass black holes providing a new observational avenue for studying these elusive objects and the environments in which they reside.
Read the full abstract
Logic
On definable closure in closed $H$-fields
The definable closure of a real closed $H$-field $H$ in an ambient closed $H$-field extension $M$ is shown to be an immediate extension of $H$, except in two very special cases, where this is only true after adjunction to $H$ of a particular $H$-definable element of $M$ and then taking the real closure in $M$.
Read the full abstract
Quantum Physics
An Alternative to Quantum Theory
Despite the remarkable empirical success of quantum theory, the interpretation of measurement and the nature of entanglement correlations remain open questions. We outline an alternative framework that seeks to recover the empirical results of quantum theory from a holistic, scale-invariant ontology. Assuming only point particles in Euclidean space, characterised by ratios of masses (and charges) and by the ratios of their mutual separations, the theory describes shapes through a single scale-invariant function; no dynamical evolution is required. In principle, this framework provides a unified description of finite subsystems embedded in a configuration of infinitely many particles and offers a possible route to deriving the observed structures of atoms, molecules and macroscopic bodies without invoking standard quantum mechanics.
Read the full abstract
Sound
Textraction: Unified Text-Guided Speaker Extraction and Enhancement for Single- and Multi-Channel Audio
We present Textraction, a text-guided speaker extraction framework that enables the selective extraction of desired speakers or noise sources from an audio mixture using only natural-language descriptions. The proposed method operates within a single unified model that supports both single-channel and multi-channel inputs and it allows conditioning on either semantic textual descriptions of the target source or on its spatial description. Dedicated single- and multi-channel encoders feed a shared stack of time–frequency transformer blocks, which is conditioned on the prompt and on the recording configuration in every block through cross-attention and estimates the target directly from the mixture in one forward pass, without a separation stage or a source-selection stage. The system can extract an arbitrary subset of entities from a mixture, assuming up to two simultaneously active speakers and two noise sources, and interprets prompts specifying which entities to keep, which to remove, or both. In addition, we show that the framework naturally applies to a speech-enhancement setting, in which a single target speaker is present alongside background noise. To circumvent the application of computationally costly large language models used in prior work, we introduce a lightweight, task-adapted text encoder obtained by contrastively fine-tuning a compact text-retrieval model, which learns to be sensitive to instruction polarity (e.g., keep vs. remove) while remaining robust to paraphrasing. Experiments across attribute-based extraction, spatially guided extraction, and speech enhancement show that the proposed approach effectively isolates the desired entities under various acoustic and spatial conditions, and matches or surpasses dedicated spatial-extraction and enhancement baselines while offering substantially greater flexibility.
Read the full abstract
Representation Theory
Global-dimension spectra of gentle algebras
We classify the connected finite-dimensional gentle algebras of finite global dimension whose global-dimension spectra under derived equivalence are not intervals of integers. These are precisely the algebras with AAG invariant $[(2,4g)]$, where $g\geq2$, whose canonical line field is induced by a nowhere-vanishing vector field. Their spectrum is $\{2g,2g+2,\ldots,4g\}$. For each $g\geq2$, there are exactly two such derived-equivalence classes, distinguished by their Arf invariant. The geometric foundation is a realisation theorem for prescribed polygon weights in arbitrary genus. For an AAG invariant $[(m_i,n_i)]_{i=1}^b$ arising from such an algebra, the global dimensions attained in its AAG fibre form the integer interval $[\max_i\lceil n_i/m_i\rceil,\max_i n_i]$. By controlling the complete line-field invariant through local ribbon-graph constructions, we determine the spectrum in every derived-equivalence class.
Read the full abstract
Machine Learning
Out-of-the-Box Tabular Foundation Models for Bayesian Optimization in Materials Discovery
Bayesian optimization is widely used in active learning loops for materials science. Gaussian processes serve as the de-facto surrogate model family, but require task-dependent kernel choices and can lack the modeling capacity for complex relationships. Tabular foundation models (TFMs) have recently emerged as the state-of-the-art for predictive tasks on structured data, without requiring bespoke kernels or hyperparameter tuning. We evaluate two TFMs, TabPFN and TabICL, as out-of-the-box surrogate models on two standard materials science benchmarks, comparing against a large field of baselines encompassing Gaussian processes, random forests, and CatBoost. Our findings indicate that TFMs are highly competitive surrogate models across the two benchmarks, without task-specific training for active learning or GP kernel selection. Our results motivate developing TFMs further as surrogates for active learning in materials discovery.
Read the full abstract
Algebraic Geometry
Semisimple Deformations of Contraction Algebras for Smooth Threefold Flops
Building on Bryan–Katz–Leung, Hua–Toda and Kawamata, we study deformations of contraction algebras and the local geometry of the scheme of associative algebra laws. Let $A$ be the basic contraction algebra of a completed projective flopping contraction of a smooth complex threefold. We construct a finite free polynomial family that is constant and semisimple away from the origin, with the original curve projectors fixed. For each restricted positive root $\beta=(\beta_i)$, with coordinates indexed by the retained vertices, the split generic fibre has $n_\beta$ matrix blocks of size $\sum_i\beta_i$, and the labelled projector $e_i$ has rank $\beta_i$ on each block. Here $n_\beta$ is the corresponding genus-zero Gopakumar–Vafa invariant. If the geometric generic fibre of a finite free associative formal deformation of $A$ is semisimple, it has the same matrix factors, and any lifts of the original projectors have the same ranks on those factors. The scheme of associative algebra laws is smooth at $A$. The closure of the corresponding semisimple orbit is its unique irreducible component through $A$. The proof uses the curve divisors of Katz–Morrison and Nabijou–Wemyss, presentations of the universal contraction quotient and its cocenter, and the orbit method of Jelisiejew–Smoktunowicz. These presentations prove the geometric Hua–Toda flatness assertions. In the geometric families considered here, the relative contraction algebra, its centre and its cocenter are finite free over arbitrary complete Noetherian local complex bases. On a smooth open neighbourhood of $A$ in the scheme of algebra laws, the centres and cocenters form dual vector bundles compatible with arbitrary base change. For the type $D$ algebras $D_{n,2}$ with $n\ge2$, the centre morphism gives counterexamples to the smoothness assertion in Conjecture 1.4(2) of Jelisiejew–Smoktunowicz.
Read the full abstract
General Relativity and Quantum Cosmology
Gravity Without Absolutes
By formulating his theory of universal gravitation in absolute space and time, Newton demonstrated the power of reductionism; it has been behind every major advance in physics since then. However, the lack of significant advances over the last half-century, together with unresolved problems in the dominant paradigms of physics and cosmology, suggest that holism may need to replace reductionism as the way forward. The ontological simplicity and historical significance of Newtonian gravity makes it the ideal theory for examining this possibility. Taking Newton's second law of motion to be the core of his theory, we show that its use within the reductionist framework of absolute space and time introduces an infinity of arbitrary solutions if used to describe $N$-body universes rather than subsystems within them. We show that Leibniz's principle of sufficient reason (PSR) eliminates the infinite arbitrariness and transforms the predicted nature of $N$-body universes; they are all shown to be unexpectedly interesting. An even greater change is enforced if one asks what principle could determine the number of particles in the universe. The PSR suggests it can only be infinite. We suggest that, in such a universe, correlation could replace causation while still giving the appearance that each state has been created by dynamical evolution from the past, whereas its actual explanation may lie entirely within the present state. While much work remains to develop these ideas into a viable theory, we are confident that it advances the case for holism.
Read the full abstract
Analysis of PDEs
Microlocal properties of Cauchy evolution under metric variations
We develop the microlocal analysis of the Cauchy evolution for wave-type equations on the infinite-dimensional space of Lorentzian metrics. We regard the Cauchy evolution operator as a distributional function of the metric, and give this notion a precise meaning through restrictions to finite-dimensional families of metrics. We study the singularities of this metric dependence. For normally hyperbolic operators on vector bundles, we determine the resulting operator-valued wavefront sets. Their parameter covectors are given by integrals of metric variations along individual null geodesics. At ultrastatic ends, metric dependence also determines the time-translation generators and their positive- and negative-energy splitting. More generally, the parameter singularities distinguish the two characteristic sheets and can be used to reconstruct the ordinary Cauchy propagation relation, which describes propagation of singularities. Consequences include smoothing under parameter averaging and polar-cone bounds associated with averaged null-positive variations, linking the result to averaged null energy conditions. The analysis also covers the Cauchy evolution of the Dirac equation.
Read the full abstract
Computation and Language
Detecting AI-Authored Text in Authentic Student Writing: Evidence from PISA 2025
AI-text detectors reach near-perfect scores in evaluations that contrast fully human-written texts with texts generated from researcher-designed prompts. In authentic student writing, however, AI use is voluntary and often partial. We study this setting with the PISA 2025 FLA dataset, in which 15-year-olds in Germany wrote English argumentative essays with and without access to a gpt-4o-mini-based chatbot. Keystroke, clipboard and chat logs let us trace every character of the final essay to its origin. From this provenance, we derive text- and sentence-level labels for human-written, AI-generated and mixed text, and for text written after chatbot use without traceable AI material. We evaluate a feature-based detector and a classifier on frozen Llama-3.1-8B embeddings, both adapted from top-ranked shared-task systems. Both separate AI from human text reliably (macro-F1 up to 0.96 for texts and 0.95 for sentences), independently of spelling errors. The feature-based detector, however, relies on lexical complexity, which a simple student-style rewriting prompt removes (AI recall 0.92 $\to$ 0.60 with gpt-4o-mini, 0.41 with Qwen3). The embedding-based detector stays robust for whole texts (AI recall $\geq$ 0.94) but not for single sentences (AI recall as low as 0.59). Mixed text falls between the two classes, and text without traceable AI material is classified as human-written in about 95% of texts.
Read the full abstract
Data Structures and Algorithms
Optimizing the Steady-State Travel Time of Nash Flows over Time with Capacity Increases
We study budgeted capacity augmentation for single-source single-sink \NFOTs under constant inflow $u_0$. The goal is to increase capacities on arcs subject to a global budget constraint in order to minimize the steady-state travel time in the network. We show that this problem is weakly NP-hard on series-parallel networks and provide an exact pseudopolynomial algorithm as well as a fully polynomial algorithm that attains the optimum travel time for budget $B$ while spending at most $B + \varepsilon\cdot\max\{B,u_0\}$. On general networks the problem becomes strongly NP-hard. Moreover, unless $\text{P} = \text{NP}$, for every fixed $\varepsilon \in (0,1)$, no polynomial-time algorithm can achieve the optimal steady-state travel time while exceeding the budget by at most a factor of $n^{1-\varepsilon}$. Lastly, we give structural properties about steady-state travel time under varying arc capacities. We identify capacity thresholds determined by subsets of $s$-$t$-cuts and show that the travel time is constant in the interior of the cells resulting from the corresponding hyperplane arrangement.
Read the full abstract
Distributed Computing
Resetting-based designated information distribution
For a network of $n$ nodes under random gossip dynamics, we show that by introducing the additional resetting mechanism (with rate $\beta$) to a designated subset of $r$ nodes, the expected sum of all node values is asymptotically equal to $\tfrac{n}{r}$ times the sum of the initial values of the $r$ designated nodes, in the small-$\beta$ and long-time limit. We also analyze the convergence speed to this asymptotic state and provide an approximate high-probability confidence interval for the expected sum at finite time $t$ and finite $\beta$. Operationally, we propose a protocol that distributes the initial node values over the designated subset evenly (up to a prescribed precision), by appropriately turning on and off the resetting dynamics. Our work not only offers insights into classical problems such as distributed computing and network control, but also serves as a potential foundation for designing quantum algorithms that prepare desired states in quantum networks using only local operations.
Read the full abstract
Materials Science
Schottky barrier height modulation and memristive behavior in a mechanically exfoliated $MoS_2$ flake with multiple lateral electrodes
Van de Waals materials possess great potential as core active materials for novel electronics, optoelectronics, and sensing applications. In particular, transition metal dichalcogenides (TMDs) and other low-dimensional materials, such as hexagonal boron nitride and binary oxides exhibit memristive behavior, making them attractive candidates as building blocks for resistive memories and neuromorphic computing devices. In this work, we investigate two-terminal lateral memristors based on mechanically exfoliated few layer $MoS_2$ flakes. We study the influence of the Schottky barrier height (SBH) as a function of the contact area. We report different SBHs from 60 meV up to 208 meV in the same $MoS_2$ flake and analyze the effect of the SBH in the memristive-like $J-V_{DS}$ curves and the observed non-volatile/volatile resistance switching. Our results highlight the importance of contact engineering to obtain different memristive properties in $MoS_2$-based devices.
Read the full abstract
General Relativity and Quantum Cosmology
Direct product geometries in $n$-dimensional linear and quadratic gravity
Direct-product spacetimes offer a tractable route to exact solutions in gravity theories beyond general relativity. We study $n$-dimensional direct-product spacetimes of $(m_1+m_2)$, $(2+m)$, and $(2+\ldots+2)$ type in linear gravity and general pure quadratic-curvature gravity, allowing the two-dimensional subspaces to have nonconstant curvature. We classify the resulting solution spaces and determine how the allowed geometries depend on the gravitational theory, the dimension, and the curvature couplings. Metrics containing two-dimensional part(s) of nonconstant curvature are governed either by algebraic conditions on a curvature profile, or, in one special class of quadratic gravity, by a differential equation. Relations governing metrics with only constant-curvature $m$ dimensional subspaces are purely algebraic. We compare the two considered theories and study relations between their solution sets. We further examine conformal relations, generalized Petrov types, and compatibility with electromagnetic sources. Four- and six-dimensional cases are analyzed in detail, and so is the Weyl squared action (that reduces to Weyl conformal gravity in $n=4$), as a particular example of quadratic gravity couplings. Explicit analytic solutions illustrating the resulting structure are presented whenever possible.
Read the full abstract
High Energy Physics (Theory)
Cubic Curvature Corrections to Taub-NUT-AdS Black Holes and Their Thermodynamics
We investigate cubic curvature corrections to four-dimensional Taub-NUT-AdS black holes and their thermodynamics. We construct the complete analytic solution to the first order in the higher-derivative couplings, and derive the corrected Hawking temperature and Wald entropy. The NUT potential remains unchanged, while reconstructing the NUT charge and Gibbs free energy through the Maxwell relation leaves just one undetermined function of the NUT parameter. We then independently evaluate the Gibbs free energy using two improved Reall-Santos methods, based on either the field redefinition or the background subtraction. After matching the thermodynamic variables, both methods yield identical results and fix the undetermined function, providing a consistent determination of the mass and NUT charge. Our work therefore establishes that the thermodynamics of Taub-NUT-AdS black hole withstands the higher-derivative corrections.
Read the full abstract
Quantum Physics
Universal two-qubit control across dilution refrigerators via a 10-km coherent optical link
Distributed quantum networks are important for scalable quantum processors and for potential applications in communication, distributed computation, and sensing, in which direct classical links for adaptive control are essential to harness the shared entanglement. However, controlling superconducting quantum processors from a separate refrigerator over long distances remains elusive. Here, we introduce a coherent optical link (COL) that delivers microwave waveforms and their phase references between dilution refrigerators. Combining a laser, a microwave-to-optical transducer and an optical-to-microwave photodiode, the COL transmits microwave control signals over 10~km of fiber, while optical references derived from the same laser compensate for the frequency mismatch between devices. Through the COL, we deliver a universal gate set to a transmon qubit and a cavity qubit in another refrigerator and entangle them, and achieve a Bell-state fidelity of 80.6%. Quantum control delivered through the COL thus provides a classical backbone for modular quantum processors.
Read the full abstract
Information Theory
Reference-Dependent Utility Optimization for Goal-Oriented Communication Systems
Traditional communication resource allocation optimizes operational metrics such as rate, latency, energy efficiency, and reliability. Motivated by goal-oriented semantic communication, we develop a generalized allocation framework that transforms operational rewards through reference-dependent user utilities and evaluates them via expected utility, cumulative prospect theory (CPT), or related decision criteria. The formulation separates operational rewards, user-side valuations, and system-level decision criteria under uncertainty. We characterize utilities incorporating QoS/service reference levels, deficit sensitivity, post-reference saturation, and loss aversion. Weighted sum-reward maximization, proportional fairness, $\alpha$-fairness, and stochastic expected utility maximization emerge as exact special cases, while soft QoS-aware allocation follows from a deficit-penalty generalization. For deterministic orthogonal systems, we derive utility-weighted marginal-reward KKT conditions characterizing global optimality under concave composite objectives. For nonorthogonal and stochastic systems, we develop successive convex approximation (SCA) algorithms and establish stationary-point guarantees under explicit smooth and composite nonsmooth assumptions, including a CPT branch-stationarity result under eventual nondegenerate ordering identification. For globally solved penalized problems, we derive an $O(1/\lambda)$ bound on excess aggregate soft-QoS deficit, and on the deficit itself whenever a QoS-feasible allocation exists. Additionally, we establish an $O(1/\epsilon^2)$ outer iteration bound for the smooth fixed-step SCA best-response residual. Numerical examples demonstrate how reference points, marginal utility slopes, channel quality, and resource budgets influence active users and allocation policies.
Read the full abstract
Cryptography and Security
Brief introduction to post-quantum cryptography in industrial applications
The development of Post-Quantum Cryptography (PQC) is directly motivated by the necessity to secure information in the face of the quantum threat. This paper provides an overview of recent developments relevant to PQC adoption across several industrial applications, including IoT, healthcare, civil and infrastructure engineering, and defense. We present a brief, self-contained overview of PQC and its market positioning. We perform implementation experiments on selected quantum-safe methods and quantify several aspects, such as the quantum resources required to break common cryptographic schemes, as well as the performance of the most well-known PQC algorithms. We also discuss the resistance of different PQC classes to quantum attacks and the status of PQC implementations in different programming languages. Finally, we present benchmark results on KEMs and signature algorithms to help select the appropriate algorithm for a specific application based on criteria such as algorithm performance, data lifetime, and level of data sensitivity.
Read the full abstract
Artificial Intelligence
AuDiR: Diagnosing Audio Reasoning with Rule and Percept Interventions
Answer accuracy alone does not distinguish failures in acoustic perception, rule knowledge, and rule application. AuDiR (Audio Diagnostic Reasoning) examines these distinctions through controlled input interventions. Each item contains an audio clip, a multiple-choice question, a task-general rule, and a textual description of the acoustic observations needed to answer. Three conditions progressively supply the rule and observations; a fourth removes the audio while retaining the question and rule. The first correct diagnostic condition assigns an operational response label, without identifying a unique internal cause. The current benchmark contains 201 items in 18 families drawn from 9 sources, with answer references documented separately from percept extraction. In a preliminary evaluation of Gemini-3.8-flash, accuracy is 29.3% without supplied information, 31.3% with the rule, and 98.5% with the observations also supplied. No-audio accuracy is 23.9%, with a confidence interval containing the 25.0% chance rate. A family-level knowledge probe associates incomplete rule recall with larger gains from rule provision. These results show the benefit of supplying acoustic observations as text, while incomplete validation, possible recognition shortcuts, and non-monotonic responses limit stronger failure-attribution claims.
Read the full abstract
Computer Vision
Perturb, Aggregate then Reconstruct: Consensus Aggregation and Fourier-Attended Multi-Decoders for Reconstruction-based Multi-View Anomaly Detection
Multi-view anomaly detection is crucial for industrial inspection to identify defects hidden from single viewpoints. Yet, despite their strong success in single-view applications, reconstruction-based methods remain largely underexplored in multi-view settings. A key challenge in extending these methods lies in generating view-consistent pseudo-anomalies. To address this, we propose PAR, a reconstruction-based framework for enhanced multi-view anomaly detection. First, we introduce a feature perturbation strategy coupled with a consensus aggregation module that integrates these perturbed features across viewpoints to mitigate view-inconsistent anomalies and construct highly aligned pseudo-anomaly representations. Second, we employ a view-specialized multi-decoder architecture; by assigning a dedicated decoder to each view, we increase the model's representational complexity, enabling the generation of highly feasible, localized anomaly maps rather than less precise, view-agnostic reconstructions. Last, we incorporate Fourier attention into the decoders to emphasize global frequency-domain structures over local token evidence, effectively preventing identity mapping. Extensive experiments on the Real-IAD dataset demonstrate that PAR consistently outperforms state-of-the-art methods in sample-level, image-level, and pixel-level anomaly detection. Our code is available at https://github.com/SpatialAILab/PAR_mvad.
Read the full abstract
Superconductivity
Scalable operating currents in amorphous W-N/FePt superconducting rectifiers
We demonstrate high-efficiency, field-free superconducting diode behavior in W-N/FePt bilayers fabricated entirely by sputtering at room temperature. The devices combine amorphous W-N superconducting thin films with a 48-nm-thick FePt layer exhibiting remanent stripe-domain magnetization. We used lithographically patterned microbridges to compare W-N/FePt bilayers with single-layer W-N reference films at three nominal W-N thicknesses (18, 24, and 28 nm). The bilayers exhibit superconducting transition temperatures of approximately 4.1-4.3 K. At zero applied magnetic field, the current-voltage (I-V) characteristics exhibit pronounced nonreciprocity, controlled by the orientation of the in-plane remanent FePt magnetization relative to the transport current. The diode response is maximized for perpendicular alignment, reverses polarity upon magnetization reversal, and is strongly suppressed when the magnetization is aligned with the current. This behavior is consistent with asymmetric vortex entry and motion governed by the magnetic stray fields generated near the bridge edges. At 3 K with the remanent magnetization perpendicular to the current, increasing the W-N thickness from 18 to 28 nm widens the current interval between the dissipation thresholds for opposite polarities from about 0.3 to 1.55 mA. All three bilayers exhibit diode efficiencies of approximately 75-80% in this configuration.
Read the full abstract
Machine Learning
Learning to Select Sparse Linear Solvers with Convolutional Neural Networks
We investigate convolutional-neural-network (CNN) based automatic solver selection for sparse linear systems, a problem of practical importance where solver performance varies dramatically with matrix sparsity structure. A CNN is trained to regress log timing ratios between candidate solvers using spy plot images of matrix nonzero structure as input, achieving strong prediction accuracy on both held-out SuiteSparse matrices and real-world matrices from high-performance simulation codes.
Read the full abstract
Lattice Field Theory
Optimised Distillation for Nucleon Correlation Functions
The calculation and analysis of nucleon correlation functions is essential for understanding the observed spectra and structure of these ubiquitous hadrons. Two challenges arise in their analysis: at early Euclidean times, they are dominated by excited state contributions and at later Euclidean times, the signal-to-noise ratio degrades exponentially. This leads to a common challenge in identifying a reasonable window on which to extract the ground state energy. By modulating the contribution of each Laplacian eigenmode to the correlator via distillation profiles, we build optimised operators which improve the overlap with the nucleon ground state. In these proceedings, we present our progress on constructing optimised distillation profiles for nucleons and their excited state suppression in two- and three-point correlation functions.
Read the full abstract
Geometric Topology
The full length of short pants
Bers proved that any genus $g$ hyperbolic surface admits a pants decomposition of length bounded above by a function of its genus. The goal here is to provide an explicit upper bound for the sums of lengths which is on the order of $(g \log(g)) ^{3/2}$.
Populations and Evolution
ADZE 2.0: rarefaction estimates of allelic richness and private allelic richness for whole-genome data
Allelic richness and private allelic richness describe how alleles are distributed across populations, but both depend strongly on sample size. Rarefaction makes them comparable by computing the expected number of alleles in subsamples of a common size, and ADZE implemented this approach and an extension to alleles shared only by a set of populations. However, ADZE was developed for microsatellite data and cannot read sequencing data in variant call format (VCF). Here we present ADZE 2.0, which reads VCF, computes each statistic genome-wide or in genomic windows, and can treat blocks of phased SNPs as haplotype alleles. We also implement improved memory handling and multi-threading support. ADZE 2.0 reproduces the ADZE 1.0 estimates for human microsatellite data exactly. On chromosome 22 of the high-coverage 1000 Genomes Project data, private allelic richness for all 325 population pairs took 4.4 minutes on 32 threads. ADZE 2.0 is available at https://github.com/szpiech/ADZE.