Peer Revenue Returns subject to peer review

Papers

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

All of arXiv

Monday 31 August

Disordered Systems and Neural Networks

Resonance statistics, Fock-space branching, and long-range pair networks in slowly varying interacting chains

In a slowly varying aperiodic potential $h_i=h\cos(2\pi\beta i^n+\phi)$ with random power-law density interactions $V_{ij}/|i-j|^\alpha$, the resonant-object ensemble changes across $\alpha=2n$, wherever Hartree fragmentation is operative, from isolated two-level bonds to a mixture of fragmented multi-site clusters and isolated wing bonds that survive at finite density, while the pair-resonance scaling $x_{\rm pair}^{4-2n-\alpha}\ln x_{\rm pair}\propto h^2/(Vt_0)$ is unchanged~\cite{letter}. Here we develop the microscopic resonance theory underlying these results, together with its domain of validity. We derive the exact phase-averaged resonance statistics — the supply $N_0\propto L^{2-n}/h$, the correlated common-phase comb, and the closed-form Hartree variance — proving that the interaction leaves the leading supply law unchanged. Exact construction of the resonant Fock-space graph at $L\le18$ shows that the order-one forward-branching scale $h_{\rm FB}\propto L^{2-n}$ carries no giant component: one-step branching and connectivity are inequivalent. We separate the fixed-pair matching law from the shell-averaged $q\ln(1/q)$ law of the bond ensemble and develop the comb into a mesoscopic shell theory; we map the fragmentation domain, with its support threshold $V_*(\alpha)$ and the boundary-healing recursion; and we treat the marginal case $\alpha=2$, where shell and matching marginalities compound into a double logarithm. Three long-range thresholds emerge with distinct meanings — $\alpha=1/2$ (the exact variance threshold of the random Fock-space energy and the square-summability boundary of the leading LIOM-dressing estimate), $\alpha=2n$ (change of the local resonant objects), and $\alpha=2$ (marginality of the long-range shell sum) — none of which is, by itself, a localization transition.
Read the full abstract

Machine Learning

PRACTICE: From Experience to Expertise in Self-Evolving Embodied Agents

Recent studies have shown that multimodal large language models (MLLMs) can serve as embodied agents, translating language instructions and visual observations into executable plans. However, building agents that can continually improve through interaction and rapidly adapt to their environments remains challenging. Summing up experience from past interaction trajectories provides a promising solution, but existing experience-based methods often rely on manually designed prompting workflows to extract and update skills. Such fixed procedures may struggle to learn updated skills from new and diverse experiences. We introduce PRACTICE, which trains a skill learner to discover and maintain a persistent skill library from past interaction trajectories while keeping the task executor frozen. Given the historical accumulated skills and incoming trajectories, the skill learner produces structured batch-edits that add, refine, merge, or remove skills, and then hierarchical consolidate all collected edits into a consistent updated skill library. We train the learner with a two-stage curriculum. First, it learns basic skill generation and library maintenance from oracle trajectories. Then, by contrasting successful and failed trajectories from heterogeneous executors on the same tasks, it learn to identify invalid action patterns and recovery strategies. Finally, we apply online skill-edit distillation to align the skill learner with a stronger teacher on its current edit distribution to further improves the policy. Experiments demonstrate that a compact skill learner delivers consistent performance improvements across successive library-update rounds for multiple frozen executors. On EB-ALFRED and EB-Habitat, PRACTICE further outperforms the strongest experience-based baselines. Project resources are publicly available at: https://baai-agents.github.io/PRACTICE
Read the full abstract

All authors Ziyi Bai, Siqi Li, Tinglei Huang, Börje F. Karlsson

General Relativity and Quantum Cosmology

Geometric origin of intrinsic rigidity for extremal horizons

Our work clarifies the origin of enhanced symmetry of near horizon geometries. We relate this property to an existence of specific foliation by non-expanding horizons. The existence of this foliation can be shown in any spacetime satisfying null convergence condition. Our proof is geometric and based on analyzing null hypersurfaces in near horizon geometry spacetimes.
Read the full abstract

Computer Vision

CheXGround: Anatomical Region Tokens for Grounded Longitudinal Chest X-ray Interpretation

Recent radiology multi-modal language models have made substantial progress in chest X-ray report generation, visual question answering, and temporal reasoning. While longitudinal chest X-ray interpretation compares sequential examinations to describe change, visual grounding aims to connect clinical language with localized image evidence. Although longitudinal modeling and visual grounding have each advanced radiology language models, how localized visual evidence can support longitudinal interpretation remains under-explored. We introduce CheXGround, a region-grounded longitudinal chest X-ray language model that represents paired studies through corresponding anatomical regions. CheXGround extracts anatomical regions from current and prior radiographs, encodes them as temporally enhanced Region-of-Interest (ROI) tokens, and combines them with global temporal image context during generation. To connect these region tokens with clinical text, we propose Temporal Region–Phrase Alignment, a pretraining objective that aligns temporal anatomical representations with localized report phrases. We evaluate CheXGround on single-study and longitudinal Visual Question Answering (VQA), longitudinal findings generation, temporal grounded VQA, and anatomical grounding. Across these tasks, CheXGround improves clinical language quality, temporal reasoning, and localization accuracy over recent baselines. Our results suggest that organizing longitudinal evidence at the anatomical level is a strong representation for grounded radiology language modeling. Project page: https://adonaydem.github.io/chexground-website
Read the full abstract

All authors Adonay Demewez Gebremedhin, Wessam Shehieb, Sara Alansari, Mohamad Alansari, Muzammal Naseer, Sajid Javed, Naoufel Werghi

Artificial Intelligence

Which Rules Matter Now? Policy-Centroid Routing Before an Intelligent System Acts

Before an intelligent system can decide whether an action is allowed, it must first know which rules the action has approached. A single proposed action can implicate several policy regimes at once. Their requirements may stack, overlap, or qualify one another, yet many remain written in natural language while the action itself arrives as an incomplete description of intent. The first problem is not judgment. It is attention. Policy-centroid routing creates a layer before adjudication. It compresses expressions within each policy regime into one or more representative centroids, places the proposed action in the same semantic space, applies a declared measure, and routes every regime crossing a declared threshold to authoritative review. Several regimes may trigger at once. The output is a review agenda, not permission, prohibition, legality, breach, compliance, certification, or enforcement. The paper develops six falsifiable propositions and seven follow-on studies comparing the hypothesis with structured workflows, lexical and semantic retrieval, hierarchical and direct classification, and selective prediction under matched review burden. The studies are designed to identify where policy geometry recovers applicable regimes, where compression loses rare or overlapping obligations, and where the mechanism should abstain. The paper includes a synthetic worked example and reports no empirical efficacy result.
Read the full abstract

Software Engineering

On the Prospects of Dynamic LLM Conversations in Software Development

Large language models (LLMs) have become an essential tool for assisting developers, yet we still lack knowledge on ways to effectively support their interactions during development activities. That is, the quality of interactions with a chat-based LLM still strongly depends on how developers phrase prompts and which information they include. Our goal is to evaluate whether interventions into these interactions with LLMs have an effect on software developers—be it harmful or beneficial. To this end, we conducted a four-month longitudinal study with third-semester computer science students working on a full-stack Web development project using chat-based LLMs under three conditions: (1) a context-aware group received intent-based conversation augmentation, (2) a proactive group received follow-up suggestions and tailored advice, and (3) a control group without intervention. Our augmentations are minimal: (i) to reduce confounding factors and (ii) to isolate treatment effects. Analyzing interaction logs and user surveys revealed no major differences in interaction patterns, indicating no detectable harmful effects in the measured outcomes when intervening in interactions. Moreover, we observed trends of increased satisfaction with the proactive treatment. The results indicate that even with minimal interventions, dynamic guidance mechanisms for developer-LLM interactions show observable effects, such that more severe augmentations may have the potential to substantially improve developer satisfaction.
Read the full abstract

All authors Annemarie Wittig, Alina Mailach, Janet Siegmund, Norbert Siegmund

General Relativity and Quantum Cosmology

External magnetic-field effects on dipolar-particle orbits and critical collisions in Kerr–Bertotti–Robinson spacetime

Strong magnetic fields affect particle dynamics through two distinct channels: gravitational backreaction deforms the spacetime, while direct coupling to an intrinsic magnetic moment depends on the relative orientation of the field and the dipole. We disentangle these effects by studying equatorial orbits and near-horizon collisions of electrically neutral magnetized particles in the exact Kerr–Bertotti–Robinson spacetime. The field-induced geometric deformation shifts turning points and circular-orbit domains and can eliminate a finite effective-potential well together with its bound orbits. Even without direct dipole coupling, it can also offset the Kerr periapsis advance and produce a finite-radius zero-precession orbit. Direct dipole coupling breaks the symmetry under magnetic-field reversal and shifts the radius, energy, and angular momentum of the innermost stable circular orbit in an orientation-dependent manner. The formal ultrarelativistic endpoints of these orbit branches, however, remain fixed by the background geometry and approach circular null orbits. In the Bañados–Silk–West mechanism, both magnetic effects modify finite-radius potential barriers and hence the ability of a critical particle to reach the near-horizon collision region. In the representative nonzero-field cases examined here, an exactly critical particle released from infinity is blocked before reaching an extremal horizon, although a locally admissible collision between critical and usual particles can still produce unbounded center-of-mass energy. Finite dipole coupling shifts the barriers but does not change the leading near-horizon divergence. Near a nonextremal horizon, an exactly critical particle is excluded and the collision energy remains finite.
Read the full abstract

Computation and Language

CLIN: an Objective Framework for Evaluating Creativity in Short Persian Literary Text

Evaluating creativity in large language model (LLM) outputs remains challenging because creativity is multidimensional and human-centered. We examine how reliably LLMs evaluate short literary text in Persian, a low-resource language, across multiple evaluation strategies and prompt formulations. We find that LLM-human agreement varies substantially across dimensions: alignment is stronger for structured TTCT-derived properties such as Originality, Fluency, and Elaboration, but considerably weaker for more subjective dimensions, particularly Emotion and Attractiveness. Judgments are also sensitive to prompt formulation, while few-shot prompting, ensembling, and multi-agent debate provide no consistent improvement. Motivated by this dimension-dependent behavior, we investigate whether structured creativity dimensions can instead be approximated using simple, interpretable proxy metrics. We introduce CLIN, which evaluates three TTCT-derived dimensions separately using topic-aware novelty for Originality, contextual lexical clustering for Fluency, and lexical diversity for Elaboration. These proxies achieve human alignment comparable to or better than the strongest zero-shot LLM judge in our setting while requiring substantially lower evaluation cost.
Read the full abstract

All authors Mohammad Reza Modarres, Armin Tourajmehr, Yadollah Yaghoobzadeh, Mohammad Taher Pilehvar

Information Retrieval

Learning from What You Retrieve: Online RL Fine-Tuning for Semantic Retrieval

In large-scale e-commerce retrieval, dual-encoder retrievers are op- timized for contrastive similarity, whereas downstream rerankers capture finer-grained relevance preferences; this objective mis- match limits end-to-end retrieval quality. Reinforcement Learning offers a way to use reward-model feedback for retriever adaptation, but we observe that standard policy-gradient updates can degrade embedding geometry, especially when the document index must remain frozen due to industrial constraints. To address this, we propose PAO (Positive-Advantage-Only), a selective RL optimization method. Our analysis reveals that in- discriminate penalization of negative samples (pushing away) in a frozen high-dimensional space disrupts pre-trained semantic man- ifolds. PAO selectively applies gradient updates only to retrieved items with positive advantages, effectively pulling query embed- dings toward high-reward regions while preserving global topo- logical stability. Experiments on both a massive industrial dataset and public benchmarks demonstrate that PAO significantly outper- forms standard RL and distillation baselines.
Read the full abstract

All authors Shaowei Wei, Chong Huang, Songtao Fang, Jin Zhang, Zhuojun Wang, Chengfu Huo

Differential Geometry

Quasi-convexity of energy functions along Teichmüller geodesics

Hyperbolic length functions are among the most fundamental ones on Teichmüller space, and they are quasi-convex along Teichmüller geodesics. In this paper, we investigate the same question for energy functions of harmonic maps in two natural settings, which may be viewed as nonlinear and higher-dimensional analogs of the length functions. For a fixed domain and a varying hyperbolic target, we prove the energy is quasi-convex along Teichmüller geodesics under a filling hypothesis. Furthermore, we generalize Masur's result on asymptotic growth of the length function along the Teichmüller geodesic determined by a Jenkins-Strebel differential to the energy functions. We also prove the quasi-convexity for covering maps between closed hyperbolic surfaces with fixed target and varying domains. We derive first and second variation formulas of energy functions along Teichmüller geodesics and explain why the natural global statement is quasi-convexity rather than genuine convexity.
Read the full abstract

Artificial Intelligence

Autoregressive Mosaics: Probing 2D Spatial Reasoning in Text-Only Language Models

Large language models (LLMs) trained only on text and code can sometimes generate programs that draw recognizable images. However, it is unclear whether this reflects an internal representation of 2D spatial layout or simply the ability to translate spatial descriptions into code. We introduce Autoregressive Mosaics (AM-Bench), a benchmark that separates these factors: First, a translation task gives a model a fully specified geometry of a picture in words as a prompt and asks for the code that produces it. Second, a layout task requires the model to compose an image from an underspecified prompt. Across eight open-weight text-and-code-only models, all models reliably translate specified geometry into code, but their open-ended layout performance differs substantially, indicating that these differences are not explained by code-generation ability alone. An output-medium ablation further shows that the interface or medium of expression that the model uses matters: replacing procedural code with raw SVG improves layout scores across all models. Finally, probing model activations shows that a coarse layout plan is present before generation, but reflects only the layout implied by the prompt. During generation, models track the evolving geometric state instead of executing an initially fixed plan. Overall, these results show that 2D spatial performance in text-only LLMs depends on both the model and the output medium, and is not explained by code-generation ability alone.
Read the full abstract

Machine Learning

Do VLMs Share Safety Neurons Across Modalities?

Vision-language models (VLMs) can comply with harmful requests delivered through images, even when their LLM backbones would refuse the same content in text. While prior work characterizes these jailbreaks empirically or at the representation level, how visual inputs perturb safety pathways at the neuron level remains uncharted. We close this gap with a causal, neuron-level analysis of safety mechanisms in 10 VLMs. We propose a two-stage detection pipeline with iterative ablation that accounts for self-repair, and introduce two modality-isolated benchmarks, ViSafe-Detect and ViSafe-Eval, which decouple visual and textual safety signals. Our analysis reveals: (i) Text safety in VLMs is localizable: $\sim$88 neurons ($<$0.01%) whose targeted ablation substantially reduces refusal. (ii) Text safety neurons constitute the dominant refusal pathway: ablating them is the only intervention that consistently and substantially reduces refusal across all models. (iii) Visual safety is high-dimensional and diffuse at the single-neuron level: text safety concentrates in $\sim$5 subspace directions while visual safety requires $\geq$50. This gap holds across architectures, explaining why current alignment has not closed the visual safety gap. Project page is at: https://jiaxuan-li.github.io/vlm-safety-neuron/ Warning: this paper may include examples of harmful content.
Read the full abstract

All authors Jiaxuan Li, Jiahao Zhang, Duc Minh Vo, Huy H. Nguyen, Pride Kavumba, Koki Wataoka

Computational Finance

Importance Sampling Enhanced with the COS Method for the Portfolio Risk Allocation

We introduce ISCOS, a cross-entropy importance-sampling calibration method for rare credit-portfolio losses. We derive Gaussian and Gaussian–inverse-Gamma proposals and analyse the propagation of finite-COS approximation errors to the fitted parameters. Numerical experiments for Gaussian and Student t-copula credit portfolios show the efficiency of this method.
Read the full abstract

Cryptography and Security

The Fragility of Jailbreak Robustness Across Operational States

Existing jailbreak evaluations typically characterize robustness using a single attack success rate (ASR) measured in a default configuration (the vanilla state). However, user-LLM interactions can induce diverse operational states beyond the vanilla state. In this work, we find that jailbreak robustness is highly fragile to operational-state variation: even when the attack remains fixed, changing only an ordinary system prompt not designed to affect safety can dramatically alter attack success rates. We systematically investigate this phenomenon across seven aligned models and three representative jailbreak attacks, observing substantial differences in ASR between vanilla and non-vanilla operational states. In one case, ASR increases by up to 56 percentage points (2% to 58%) solely due to a change in operational state. Remarkably, these increases occur even for attacks originally designed and optimized under vanilla-state evaluation. We further show that state-dependent robustness variation is systematically associated with differences in hidden representations along a refusal-related axis, and that projections onto this axis strongly predict jailbreak outcomes. Our results show that a single vanilla-state evaluation may not fully characterize jailbreak robustness, motivating evaluations that also examine how robustness changes across non-vanilla operational states.
Read the full abstract

All authors Yuna Park, Hwang Youn Kim, Yujin Kim, Won Woo Ro, Suhyun Kim, Jae-In Hwang

Differential Geometry

Compact homogeneous complex manifolds

We study both the complex and Hermitian geometry of C-spaces, i.e., compact homogeneous spaces M=G/K admitting a G-invariant complex structure. Some results on Hodge, Bott-Chern and Aeppli cohomologies are given. We also prove classification results in the pluriclosed, parallel Bismut torsion and locally conformally Kähler (or Vaisman) cases, respectively, as well as existence results for balanced and Calabi-Yau with torsion metrics.
Read the full abstract

Disordered Systems and Neural Networks

Hartree Fragmentation and Long-Range Pair Networks in Aperiodic Chains

For fermions in a slowly varying aperiodic potential with random interactions $V_{ij}/|i-j|^{\alpha}$, the potential fixes the resonance supply $N_0 \propto L^{2-n}/h$. On crossing ${\alpha}-2n$, wherever Hartree fragmentation is operative, the resonant-object ensemble reconstructs: turning-point runs fragment into clusters whose intrinsic matching law resolves no logarithmic singularity, while isolated wing bonds survive at finite density and restore a reduced logarithm. The scaling $x^{4-2n-{\alpha}}_{pair} ln{x_{pair}} \propto h^2/(V t0)$ is unchanged. Random coefficient fluctuations sustain the $R^{-{\alpha}}$ coupling; uniform ones cancel to $R^{-{\alpha}-2}$.
Read the full abstract

Machine Learning

TDDM-Melatt: A Decoupled Memory and Diffusion Framework for Generalizable Encrypted Traffic Classification

The widespread adoption of encrypted traffic poses severe challenges to current security situational awareness systems based on network traffic monitoring. In existing dataset-driven training and testing studies, limitations such as shortcut learning induced by spurious feature correlations and sample imbalance caused by the long-tail distribution of real-world traffic result in weak generalization of traffic identification performance to real-world network traffic. To address these limitations, we propose TDDM-Melatt, a disentangled memory-based traffic classification framework with diffusion-based data augmentation. First, we design Melatt, a memory-decoupled traffic representation model, which employs Competitive Gating Long Short-Term Memory (CG-LSTM) to construct the encoder and decoder. We design a spurious-correlation-free pre-training and inference paradigm, employing strict topology anonymization and a frozen pre-trained encoder strategy to cut off the model's learning pathways for spurious features. During inference, classification is performed efficiently by a downstream classifier on the frozen representations. Second, we propose a Traffic Denoising Diffusion Model (TDDM) tailored to the characteristics of traffic data. Extensive experiments are conducted on 4 representative public benchmark datasets. Under strict flow-level splitting and anonymization, TDDM-Melatt outperforms 6 basic classification models and 6 SOTA representation learning models. The proposed method provides a new and effective technical pathway for encrypted traffic classification in real-world network environments.
Read the full abstract

All authors Ze Chen, Qiming Yu, Zijia Song, Guozheng Yang, Wei Yan

Dynamical Systems

Dimensions of surface repellers and attractors of non-linear planar IFSs

We establish that for a $C^r$-generic surface repeller ($1 < r \leq \infty$), its Hausdorff and box dimensions are exactly the unique zero of the sub-additive topological pressure function. As an application of our framework, we show that the attractor of a uniformly non-conformal and weakly irreducible planar non-linear iterated function system (IFS) satisfying the strong separation condition (SSC) attains its expected Hausdorff and box dimensions. Furthermore, as a direct consequence of our generic repeller theorem, we deduce that this dimension formula also holds for a generic $C^r$ planar IFS satisfying the SSC. Using this approach, we also extend the dichotomy result for graphs of Weierstrass-type functions of Ren and Shen (2021) by weakening their real-analytic requirement to arbitrary $C^r$ regularity for $r > 1$.
Read the full abstract

All authors Jialun Li, Wenyu Pan, Yao Tong, Disheng Xu

Differential Geometry

Perspectives on the Bott spectral sequence for Lie groupoids

The cohomology of a differentiable stack can be modelled by the Bott-Shulman-Stasheff bicomplex. However, computing its cohomology is still difficult in many examples. Improving on this requires additional methods such as using spectral sequences and finding models for the cohomology of each column of the bicomplex. This paper provides a first attempt of relating and unifying two approaches of this form. Firstly, Arias Abad has generalised the Bott spectral sequence to Lie groupoids by introducing representations up to homotopy. The second approach is the recent work on a model for the cohomology of differentiable stacks associated to proper and regular Lie groupoids by this author.
Read the full abstract

Soft Condensed Matter

Coarse-grained simulations of dsDNA polycatenanes and network formation in annular nanochannels with topoisomerase II

We numerically investigate the behavior of a system of initially unlinked nicked dsDNA rings confined into an annular square nanochannel in the presence of TopoII. Channel confinement can enhance the catenation likelihood by bringing highly bent regions from different rings in close proximity, while at the same time reducing the emergence of knots. The annular channel topology simplifies the characterization of the system by removing periodic boundary conditions and can lead to the formation of circular catenanes.We characterize the equilibrium and dynamical properties of the steady-state system, including the amount of catenation and the topologies explored by the system, under different parameters of the model and for different levels of confinement. We argue that a similar setup could allow for a direct comparison between experiments and simulations under well characterized and controlled conditions, thus providing a way to select and tune computational models of TopoII, as well as provide a way to obtain dsDNA interlocked materials in a controlled fashion.
Read the full abstract

Machine Learning

Functional Degeneracy in Neural Networks: Measurement and Pruning

A central question in modern machine learning is how much a trained model can be compressed without changing its behavior, to reduce the memory, compute and energy required to deploy it. To study this, we quantify functional degeneracy through the behavioral recovery rank, the number of leading behavioral-Hessian eigendirections required to recover a model's performance to a given tolerance. Using it as a geometric benchmark for compression, we find that the rank flattens once the task is saturated, while the number of parameters retained by one-shot pruning keeps growing. This gap suggests that functional degeneracy is distributed across parameter directions and not exposed by individual weights or neurons.
Read the full abstract

All authors Maria Matveev, Pascal Esser, Ayush Bharadwaj, Lucius Bushnaq, Gitta Kutyniok

Representation Theory

Auslander correspondence for higher stable dg categories and cluster Morita theory

The notion of $d$-stable dg categories axiomatizes $d$-cluster tilting subcategories of stable dg categories. We establish an Auslander correspondence for $d$-stable dg categories: we characterize the $d$-stability of an additive connective dg category in terms of coherence, weak global dimension, and a duality on finitely presented modules. This gives a homological characterization of $d$-stability and reveals it as a twisted form of $(d+1)$-Calabi–Yau duality. For locally finite connective dg algebras, this interpretation becomes particularly transparent under Koszul duality, where $d$-stability corresponds to a shifted self-injectivity condition on the Koszul dual. Following the constructions of Amiot, Guo and Keller, for a $d$-stable dg category $M$, we introduce its $d$-cluster dg category $\mathcal C_{d,{\rm dg}}(M):=\operatorname{per}_{\rm dg}M/^\mathbb{L}\mathcal D^b_{\rm fp, dg}(M)$. Using our Auslander correspondence, we show that $\mathcal C_{d,{\rm dg}}(M)$ contains $M$ as a $d$-cluster tilting subcategory. In particular, every $d$-stable dg category can be realized as a $d$-cluster tilting subcategory of a stable dg category. We then develop cluster Morita theory: a pretriangulated dg category equipped with a $d$-cluster tilting subcategory $M$ is quasi-equivalent to $\mathcal C_{d,{\rm dg}}(M)$. Thus the connective dg structure of a cluster tilting subcategory determines its ambient dg category up to quasi-equivalence. As an application of cluster Morita theory, we prove a Morita-theoretic variant of Amiot's conjecture. More precisely, we establish a Calabi–Yau correspondence: for a locally finite $d$-stable dg category $M$ over a field, right $(d+1)$-Calabi–Yau structures on $\mathcal D^b_{\rm fp, dg}(M)$ are in bijection with right $d$-Calabi–Yau structures on $\mathcal C_{d,{\rm dg}}(M)$.
Read the full abstract

Audio and Speech Processing

Towards Balanced Spectral Reconstruction: Spectrally Adaptive Loss for Streaming Speech Enhancement

This paper proposes two spectrally weighted STFT loss functions for lightweight streaming speech enhancement, addressing the magnitude over-attenuation in mid-to-high frequency regions caused by the magnitude-phase compensation effect. The proposed sigmoid-weighted loss applies a smooth frequency-dependent modulation to the phase-aware contribution, while the signal-dependent spectrally adaptive loss further conditions the modulation on the ground-truth log-magnitude spectrogram. To evaluate the proposed objectives, we additionally design HyST-Net, a lightweight and competitive backbone with hybrid MHA-GRU spectral-temporal modelling for low-latency streaming scenarios. Experimental results exhibit consistent improvements in high-frequency spectral reconstruction for both losses. The spectrally adaptive loss further enhances the mid-frequency region, resulting in a more balanced spectral reconstruction across the full frequency range.
Read the full abstract

Computation and Language

Not All Fallbacks Are Failures: Understanding and Recovering from Fallbacks in Mobile Voice Assistants

Robust understanding of user input is a core requirement for voice assistants deployed in real-world environments. In practice, these systems encounter heterogeneous fallback situations caused by noisy audio input, transcription errors, ambiguous requests, incomplete utterances, or unintended activations. Existing systems typically respond with generic fallback messages, which do not resolve the underlying interaction failure and can degrade user experience. We study fallback handling in a deployed smartwatch-based voice assistant for general health support in everyday environments. Our analysis is based on six months of real-world usage data from more than 500 users, yielding a dataset of 3,030 anonymized, naturally occurring fallback-triggering utterances. We contribute (1) an operational taxonomy and the annotated VoxFallbacks dataset of these interactions, (2) a comparative evaluation of different models within a classification pipeline under practical deployment constraints, and (3) practical lessons for designing robust and cost-efficient fallback mechanisms. Results show that lightweight embedding-based classifiers outperform larger generative models on most classification tasks while requiring substantially fewer computational resources.
Read the full abstract

All authors Phillip Schneider, Alexandre Mercier, Joshua Oehms, Kristiina Jokinen, Florian Matthes

Lattice Field Theory

$D \to (K \pi)_{\mathbf{27}}$ at the SU(3)-flavour-symmetric point I: Methodology and strong phase determination

We present part one of an SU(3)-flavour-symmetric lattice QCD calculation of the amplitude for a $D$-meson decaying to a $K\pi$ final state in the 27-dimensional irreducible representation of the flavour symmetry group, denoted $(K\pi)_{\mathbf{27}}$. The Wilson–clover gauge ensembles used in this work, generated by the OpenLat collaboration, are tuned such that $M_\pi = M_K \approx 410\,\mathrm{MeV}$. Using the distillation framework, we construct a matrix of Euclidean correlation functions from pairs of single-hadron operators projected to definite spatial momentum. Solving a generalised eigenvalue problem yields the finite-volume energy spectrum that is used to determine the scattering phase shift from threshold up to $4 M_\pi \approx 1640 \,\mathrm{MeV}$, which sits below but plausibly within reach of $M_D^{\rm SU(3)} \simeq 1900\,\mathrm{MeV}$. The calculation is performed across three lattice spacings, and we apply two strategies in which the continuum limit is taken at different stages of the computation: (i) on the extracted scattering parameters and (ii) on the finite-volume energies at fixed physical volume before extracting the scattering parameters. We find consistent results across these methods for the scattering phase shift as a function of the centre-of-mass energy, $\delta_{\mathbf{27}}(E_{\sf cm})$. Taking a scattering-length-only parametrisation, we infer a value for the strong phase of the weak decay, $\delta_{\mathbf{27}}(M_D^{\rm SU(3)})=-38.4(2.4)^\circ$. We further describe the methodology for using the same operator basis to compute three-point correlation functions to extract $\langle (K\pi)_{\mathbf{27}}| H_W| D\rangle$, for the tree-level effective weak Hamiltonian $H_W$, and for relating such finite-volume matrix elements to the full decay amplitude. The complete analysis leading to the latter will be presented in a forthcoming manuscript.
Read the full abstract

All authors Matthew Black, Felix Erben, Maxwell T. Hansen, Fabian Joswig, Nelson Pitanga Lachini, Rajnandini Mukherjee, Srijit Paul, Antonin Portelli

Probability

Stability results for distribution-dependent stochastic Volterra equations

We investigate stability properties of distribution-dependent stochastic Volterra equations with respect to changes in the coefficients, the Volterra kernels, and the initial condition. Under Lipschitz continuity assumptions on the coefficients, we first derive quantitative stability estimates for strong solutions with explicit error bounds. We then prove a general convergence theorem for strong solutions under substantially weaker assumptions, replacing Lipschitz continuity by a continuity assumption together with uniform linear growth of the approximating sequence. Finally, we study the associated distribution-dependent Volterra local martingale problem and prove the stability of its solutions under convergence of the coefficients, kernels, and initial distributions.
Read the full abstract

Number Theory

Density of the multidimensional Lagrange spectrum

The Lagrange spectrum is a classical object in number theory, defined as the set of values of $\liminf_{q\to\infty} q\, \mathrm{dist}(q\alpha,\mathbb{Z})$ where $\alpha$ runs through irrational numbers. It has a complicated structure, with the discrete part, Hall's ray, and a transitional part in between. One can similarly define Lagrange spectrum in the multidimensional set-up, and until now not much has been understood about it. In this paper we prove that, unlike in the one-dimensional case, the closure of the multidimensional Lagrange spectrum is equal to the interval between $0$ and its supremum. The proof relies on a correspondence between Diophantine approximation and dynamics on the space of unimodular lattices and proceeds by studying a dynamical counterpart of the Lagrange spectrum that we call dynamical Lagrange spectrum. The latter is shown to be equal to the interval between $0$ and its maximum by means of an argument utilizing the higher rank nature of the set-up. A passage from full dynamical spectrum to the density of the Diophantine spectrum is achieved by applying equidistribution of expanding translates of horospheres in the space of lattices.
Read the full abstract

Rings and Algebras

$\sqrt{\Delta}$-Fine Rings

We introduce and study the so-termed {\it $\sqrt{\Delta}$-fine rings}, a new class of rings that generalizes the classical fine rings introduced by Călugăreanu-Lam in J. Algebra & Appl. (2016) by requiring that every nonzero element $r \in R$ can be written as $r = u + a$, where $u$ is a unit and $a \in \sqrt{\Delta(R)}$. We establish that every such ring is simple, every abelian $\sqrt{\Delta}$-fine ring is indecomposable, and most notably, the matrix ring $M_n(R)$ over a $\sqrt{\Delta}$-fine ring $R$ is again $\sqrt{\Delta}$-fine for every $n \ge 1$. As a consequence, we characterize all semi-local $\sqrt{\Delta}$-fine rings as those rings which are precisely the simple Artinian rings. We also examine group rings, providing conditions under which they are either $\sqrt{\Delta}$-fine or generalized fine, where the latter class was introduced by Zhou in J. Algebra & Appl. (2022), and conclude our work with the difficult open question asking of whether each $\sqrt{\Delta}$-fine ring is necessarily fine.
Read the full abstract

All authors Peter Danchev, Omid Hasanzadeh, Ahmad Moussavi, Arash Javan

Probability

An operator-theoretic characterization of Gaussian measure singularity under mean shifts

For two Gaussian random elements on a separable Hilbert space with common covariance operator, the fourth-order moment tensor structure of their mixture admits an operator-valued representation obtained from the purely quadratic component of the class-conditional second moment. We identify the tensor mechanism generating the completely diagonal coefficients of this fourth-order representation after covariance standardization, and show that these coefficients are governed entirely by the coordinatewise Cameron-Martin energy of the mean shift. This yields a necessary and sufficient condition, together with an explicit eigenvalue, for the Fisher discriminant to be an eigenfunction of the induced coordinatewise operator. We further introduce aggregated fourth-order spectral functionals for two complementary constructions, one built from fourth-order moments and the other from a product-of-expectations counterpart, and prove that they are asymptotically equivalent precisely when the underlying Gaussian measures are mutually singular. These results provide a spectral realization of the classical Cameron-Martin criterion through higher-order moment operators, explaining the probabilistic origin of fourth-order spectral quantities previously proposed for Gaussian discrimination and functional data classification and relating them to the "near-perfect classification" regime.
Read the full abstract

Optics

Robust bistable vortex light bullets in graded-index multimode fibers

We explore theoretically the formation and evolution of spatiotemporal vortex bullets, pulses confined in both space and time while carrying orbital angular momentum, inside a graded-index multimode fiber. Through a two-fold approach, combining variational analysis with full numerical simulations, we establish the existence of a bistable vortex soliton family, characterized by distinct radial and azimuthal quantum numbers. These wave packets exhibit a Laguerre-Gaussian multi-ring topology in their spatiotemporal structure. We identify the experimental conditions necessary for the generation of such structured states of light. A stability analysis based on Vakhitov-Kolokolov criteria reveals an upper limit for the propagation constant, below which vortex bullets are found to be stable. Stationary and non-stationary dynamics of vortex bullets with different topological charges are analyzed fully, exploiting analytical techniques that are supported by full numerical simulations. These results enhance our understanding of self-trapped, spatiotemporal vortex solitons in a graded-index medium and may lead to potential applications in the area of classical and quantum information processing.
Read the full abstract

All authors Ashis Paul, Anuj P. Lara, Govind P. Agrawal, Samudra Roy