Artificial Intelligence
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Thursday 3 September
Banks need conversational systems that can answer product questions, assist customers with account-related requests, and operate safely within strict operational and regulatory constraints. General-purpose language models do not reliably meet these requirements. They fall short when a task requires grounded information, correct tool use, or cautious handling of bank-specific sensitive situations. We introduce FiMI Banking, a controlled Indian retail-banking setting. We build it from vetted banking documents, structured ground truth, synthetic customer backgrounds, and banking tools. We evaluate two post-training approaches: preference optimization for response-level behavior, and reinforcement learning with verifiable rewards for multi-turn tool-use tasks. Preference optimization improves safe behavior substantially: out-of-scope refusal rises from 52% to 80%. Reinforcement learning improves edge-case performance from 0.509 to 0.718 and order-sensitive task performance from 0.590 to 0.679, while using 29% fewer generated tokens. These results show that preference optimization and verifiable-reward reinforcement learning address complementary requirements for reliable banking agents.
Combinatorics
The combinatorial Hopf algebra of signed graphs
The theory of combinatorial Hopf algebras (CHA) is a powerful framework for studying algebraic invariants of combinatorial objects. For instance, Stanley's chromatic symmetric function and the chromatic polynomial arise naturally from the CHA of graphs. In this paper, we define a CHA of signed graphs in order to study, in the same spirit, invariants of signed graphs such as the signed chromatic polynomial. The key ingredient of our construction is a well-chosen character, from which the signed chromatic invariants are obtained. Via the antipode, we obtain new proofs of combinatorial reciprocity results for these invariants. Moreover, by deforming the character with an additional parameter, we incorporate bivariate invariants into this framework, both for classical and signed graphs. In the classical case, this technique recovers the Tutte polynomial.
Read the full abstract
Analysis of PDEs
Interior Curvature Estimates of Semi-convex Solutions for the scalar curvature equation with Lipschitz Right-Hand Sides
In this paper, let $u\in C^4(B_{10})$ with $D^2u\geq -KI$ define a $2$-admissible graph $M=\{(x,u(x)):x\in B_{10}\}\subset\R^{n+1}$ satisfying \[ \sigma_2(\kappa[u])=f(x). \] We prove an interior curvature estimate depending on the Lipschitz norm of the right-hand sides. The proof combines a shifted Jacobi inequality for \(b=\log(H+J_0)\), a parallel hypersurface transformation that makes the Newton tensor uniformly elliptic and local boundedness estimate then reduces the pointwise bound to a weighted $L^1$ estimate, which is completed using the Jacobi energy inequality and integration by parts.
Read the full abstract
Fluid Dynamics
How do flapping avian wings exhibit superior aerodynamic performance?
This work investigates the unsteady aerodynamic performance and vortex dynamics of avian-inspired flapping wings using our in-house sharp-interface immersed-boundary solver, VorteXdyn. A falcon-inspired body-wing model based on NACA 4312 profile is employed to examine aerodynamic force generation and vortex evolution during steady forward flapping flight at Reynolds numbers of 2,500, 5000, and 10000 and Strouhal numbers of 0.18, 0.225, and 0.27. The influence of physiologies is examined using three wing configurations: a simplified wing without distinct feather structures, a wing incorporating feather-like structures (serrations) along the trailing edge, and a geometrically detailed wing incorporating multiple feather layers consisting of primary, secondary, and median feathers over its span. The aerodynamic performance of these configurations is quantified using the temporal profiles and time-averaged characteristics of the lift and drag coefficients and the lift-to-drag ratio. The associated vortex dynamics are characterized through the formation and evolution of multiple leading-edge vortices (LEVs), their spanwise coherence, circulation, characteristic size, and persistence over the wings' surfaces. Particular emphasis is placed on the spanwise development of the LEVs from the root to the wingtip, their interactions with the tip vortices, and the resulting wake evolution during the downstroke. Our results demonstrate that increasing geometric fidelity modifies aerodynamic force production, the formation and evolution of multiple LEV structures, vortex-vortex and votex-wing interactions, and wake topology. These findings provide insight into the aerodynamic role of feather morphology in three-dimensional flapping flight for the aerodynamic design of efficient bio-inspired flapping-wing micro air vehicles.
Read the full abstract
Computer Vision
RARF: Region-Aware Rectified Flows for 3D Brain MRI Inpainting
Medical image inpainting has the potential to improve automated brain MRI analysis by reconstructing healthy tissue within pathological regions. We introduce RARF, a task-agnostic region-aware rectified flow framework for masked data generation. We instantiate the framework for 3D brain MRI inpainting as our submission to the BraTS Inpainting Challenge 2026. RARF restricts the stochastic interpolation process to the inpainting region, while the observed voxels remain fixed and provide patient-specific anatomical context. A three-dimensional neural network receives the partially voided image, with Gaussian noise filling the missing region, together with the inpainting mask and the corresponding timestep. The model is trained using masked flow-matching and reconstruction-consistency objectives, combined with mask-aware preprocessing and data augmentation. During inference, the learned velocity field transports the initial noise toward a plausible reconstruction of the missing tissue, which is then combined with the unchanged observed anatomy. Experiments under the BraTS evaluation protocol show that the proposed approach produces competitive reconstructions while maintaining anatomical consistency. Source code is available at: https://github.com/TomasGuija/rarf.
Read the full abstract
Computation and Language
Two-Stage Reinforcement Learning for Sound and Adversarial Test Generation in Code LLMs
Reinforcement learning (RL) has substantially advanced code generation with large language models (LLMs) through executable feedback. The feedback for coding problems mainly comes from specific test cases, where high-quality test cases are often scarce since they should be both sound and discriminative. We thus turn to study the auto-generation of test cases using the learned model. We find this is naturally an adversarial RL problem: the model is expected to generate effective test cases as counterexamples, depending on the solver's current failure modes. We propose Test Cases Scaling (TCS), a two-stage RL framework for effective test generation. Both stages train a test generator from a rolling policy-aligned buffer: Stage 1 generates tests consistent with the reference solution, and Stage 2 restricts the buffer to current failure modes and learns counterexample tests. Across TACO and LiveCodeBench, TCS improves both pass@1 and inference-time answer selection according to generated tests. We find the learned test generator also enables effective selection among other LLM outputs.
Read the full abstract
Number Theory
Galois representations ramified at one prime via relative deformation theory
Let $p$ be an odd prime and let $\mathbf{G}/\mathbb Z$ be a split connected reductive group with $\operatorname{dim} Z(\mathbf{G})\leq1$. Assume that a split maximal torus of $\mathbf{G}$ admits a cocharacter whose pairing with every simple root is odd, and impose an explicit root-theoretic condition on the reduction of $\mathrm{Lie}(\mathbf{G}^{\mathrm{der}})$ modulo $p$. We construct infinitely many continuous representations $\rho:G_{\{p\}}\longrightarrow\mathbf{G}(\mathbb{Z}_p)$ which are unramified at every finite prime different from $p$ and whose images contain a principal congruence subgroup of $\mathbf G^{\mathrm{der}}(\mathbb{Z}_p)$. When the center has dimension one, the representations may be chosen to have open image in $\mathbf{G}(\mathbb{Z}_p)$. The construction continues the author's earlier work on $\mathrm{GL}_n$-valued representations ramified at one prime, but replaces the residual unobstructedness used there by a relative lifting argument. For $\mathbf{G}=\mathrm{GL}_n$ all the required root-theoretic conditions are automatic for every odd prime. We therefore obtain, for every odd $p$ and every $n>1$, infinitely many representations $G_{\{p\}}\to\mathrm{GL}_n(\mathbb{Z}_p)$ with open image. In particular, this removes the weak Vandiver-type hypothesis occurring in the author's previous work, as well as the restriction $p\geq7$ in that construction.
Read the full abstract
Computation and Language
Beyond Majority Vote: Multi-Perspective Adjudication for Medical Hallucination Detection
Understanding the frequency of factual errors in chatbot-generated text and evaluating systems that detect these errors is critical for determining chatbot safety. Yet factual-error detection is often treated as a single-pass, single-annotator labeling problem. In long-form chatbot responses, factual errors can be subtle and embedded within mostly correct text. We develop a multi-perspective annotation study of medically relevant chatbot responses, combining first-pass annotation, LLM-as-a-Judge (LaJ) candidate discovery, and two forms of adjudication: medical-expert and evidence-based fact-checking. First-pass annotators frequently miss factual errors later validated by adjudicators. LaJ improves candidate discovery, but is insufficient on its own: It misses factual errors that annotators catch. We also find disagreement among adjudicators, suggesting that adjudication over multiple candidate sources can improve benchmark completeness, but does not eliminate the need to apply judgment and expertise. Applied to an existing benchmark, this technique reveals a similar pattern of missing annotations. Together, these results suggest that in the settings examined here, single-pass hallucination benchmarks may achieve scale at the cost of undercounting factual errors. Multi-pass adjudication can improve coverage, but inferences drawn from the benchmarks are still sensitive to the judgment, expertise, and evidence used to determine error presence.
Read the full abstract
Computer Vision
WorldReward: Reward Modeling for Camera-Conditioned World Models
Camera-conditioned world models generate interactive videos in which commanded actions should induce the expected scene changes while appearance, geometry, and temporal dynamics remain coherent. Existing rewards assess these requirements separately: geometry-based rewards estimate trajectory execution but cannot judge the visual quality of the executed motion, whereas image-based rewards measure frame quality without capturing action execution or temporal dynamics. We posit that a vision-language model (VLM) offers a shared reasoning space for relating actions to their visual outcomes. However, judging a complete long video against its full action sequence creates a lengthy, noisy context in which short-lived local action evidence can be missed or diluted. We present WorldReward, a VLM-based pairwise preference reward model that unifies action-consistency and visual-quality evaluation for camera-conditioned world models. WorldReward decomposes paired videos into action-aligned chunks, organizes each chunk into structured visual evidence, and aggregates chunk-level decisions by voting into separate video-level action and visual-quality preferences. To train it, we construct a large-scale reasoning-augmented preference dataset using structured judgments generated by a frontier VLM and refined through tool-based agent auditing and targeted human review. We further introduce WorldReward-Bench, a human-annotated benchmark measuring reward-model agreement with human preferences across action consistency, appearance quality, and motion quality. WorldReward achieves the highest agreement on all three dimensions, exceeding GPT-5.5 by 3.42, 1.45, and 3.56 percentage points, respectively. When used for RL post-training of HY-WorldPlay 1.5, it consistently improves both action execution and visual quality across short- to long-term horizons.
Read the full abstract
Differential Geometry
Generalized Semi-Riemannian Submersions and Foliations
In this article, we introduce the concepts of generalized semi-Riemannian submersions and foliations, extending the classical framework to accommodate leaves with varying or degenerate causal characters, such as homogeneous foliations on semi-Riemannian manifolds and codimension-one lightlike foliations on Lorentz manifolds (including pp-waves). We establish that a regular foliation with a basic horizontal distribution is generalized semi-Riemannian if and only if it is transnormal. Furthermore, we introduce stationary foliations and prove that a stationary transnormal foliation induces a classical Riemannian foliation on the spatial rest space of a conformal observer. As a geometric consequence, we obtain a rigidity result ruling out stationary codimension-one lightlike foliations on positively curved Robertson-Walker spacetimes or static spacetimes. Finally, we derive an O'Neill-type sectional curvature formula for submersions with an involutive total distribution.
Read the full abstract
Optimization and Control
Asymptotic Analysis of Gradient Mapping-type Stationarity Measure for the Sum of Nonconvex Nonsmooth Functions and Applications to Proximal Gradient-type Algorithms
We propose a gradient mapping-type stationarity measure for the sum of two possibly nonsmooth nonconvex functions. Under suitable regularity assumptions, we show that Fréchet or proximal stationarity can be characterized through asymptotic vanishing of the proposed measure along some convergent sequence. We also establish analogous asymptotic results for a smoothing-based variant of the measure, which enables us to combine the proposed framework with smoothing techniques developed for nonsmooth optimization. As an application of our analysis of the stationarity measure, we provide an affirmative answer to an open question raised by [Olikier-Waldspurger, SIAM J. Optim., 2025] on whether every cluster point of a sequence generated by a proximal gradient method is a proximal stationary point under local Lipschitz smoothness of one component of the cost function. As a second application, we propose a proximal variable smoothing algorithm with a nonmonotone linesearch for minimizing the sum of two nonsmooth nonconvex functions under lower regularity of one component and prox-regularity of the other. For the proposed algorithm, we show that every cluster point of a subsequence such that the stationarity measure vanishes is a Fréchet stationary point.
Read the full abstract
Machine Learning
VestigeKV: The NoPE-MLA KV Cache Carries Its Own Sparse-Attention Signal in a Vestigial Branch
A long-lived KV cache must be compressed before the queries that will read it exist. Selection by observed attention collapses there: on a NoPE-MLA model, H2O and SnapKV retrieve 0.00 and 0.33 of needles at 8x compression, because a token's importance has not yet been observed. VestigeKV instead derives a sparse attention pattern from a signal the cache already carries, occupying the sparse-attention literature's one unoccupied quadrant: training-free and query-independent. In NoPE-MLA the 64-dimensional decoupled branch is a vestige of RoPE that training repurposes into a salience channel; reading 11% of each row, it partitions the cache into an attended tier and a GPU-resident archive that no row ever leaves, reachable each step by a certified, query-adaptive trigger. Nothing is trained and cache rows are never quantized, so every quality effect attributes to selection and scheduling. On Kimi Linear 48B, retrieval holds at 1.00 under 8x and 0.96 under 32x from 8k to 65k context, with zero gap to full-row selection, and the recall tier holds 128x at 1.00 (8k). Both tiers stay on the GPU, so the win is speed, not memory: the per-step scan reads ~26% of the bytes dense attention would, and on a two-node sglang deployment the crossover sits at ~40k context, reaching 1.18x at 256k and 1.39x at 496k. The mechanism is exclusive to NoPE: the identical operator on a RoPE MLA collapses to 0.08, query-independent salience exists only without rotation, and query-universal exact merging is provably impossible under RoPE. All thresholds were frozen before their data; 20 archived verdicts and 8 closed routes accompany the paper.
Read the full abstract
Quantum Gases
Three-observable measurement of the critical velocity in a paraxial superfluid of light flowing past a mobile impurity
We report the measurement of the critical velocity of a paraxial superfluid of light flowing past a mobile impurity, realized using a dual-wavelength optical scheme in a warm vapor of $^{87}\text{Rb}$. The superflow is two-dimensional by nature, while the impurity features a highly repulsive potential, a large tunable radius, and a finite mass. The critical velocity is evaluated by simultaneously monitoring three complementary observables: the number of vortices nucleated in the wake of the impurity, the hydrodynamic force exerted on it, and its displacement velocity relative to the flow. These independent methods yield consistent values for the critical velocity, which remain systematically below the Landau criterion and decrease with the impurity radius, in qualitative agreement with theoretical expectations. Our results demonstrate the reliability of these combined diagnostics, establishing paraxial superfluids of light as a versatile platform for exploring the impurity problem in quantum hydrodynamics.
Read the full abstract
Image and Video Processing
Computationally Efficient Pathology Segmentation using Knowledge Distillation from Foundation Models
Automatic tissue segmentation is essential for large-scale analysis of histopathology whole-slide images (WSIs), but accurate pixel-level segmentation remains challenging. Pixel-level annotations are expensive to obtain, models pre-trained on natural images may transfer poorly to histopathology, and pathology foundation models, despite their strong representations, are computationally expensive to deploy at scale. We address these challenges with a foundation-model knowledge distillation framework for efficient tissue segmentation. We first train Virchow2-based segmentation teachers, including a LoRA-adapted variant, achieving state-of-the-art or highly competitive performance across four datasets: PUMA, IGNITE, BEETLE, and a private blood vessel segmentation dataset. We then transfer response-level and feature-level knowledge from these teachers into compact student networks. Distillation consistently improves student performance over supervised training alone, producing state-of-the-art or near state-of-the-art results with substantially fewer parameters and up to ten-fold higher inference throughput than foundation-model-based segmentation networks. These results show that foundation-model knowledge can be effectively transferred to efficient segmentation models for scalable deployment without compromising performance. Models trained on the public datasets will be released through TIAToolbox.
Read the full abstract
Number Theory
The groups $\operatorname{SL}_2(\mathbb{F}_{13})$ and $\operatorname{SL}_2(\mathbb{F}_{19})$ are Galois over $\mathbb{Q}$
In this paper, we show that $\operatorname{SL}_2(\mathbb{F}_{13})$ and $\operatorname{SL}_2(\mathbb{F}_{19})$ are Galois groups of totally real extensions of $\mathbb{Q}$. For each of these primes $\ell$, we find a polynomial of degree $\ell+1$ whose splitting field has Galois group $\operatorname{PSL}_2(\mathbb{F}_\ell)$. These fields satisfy Böge's criterion and the associated central embedding problem has a proper solution with Galois group $\operatorname{SL}_2(\mathbb{F}_\ell)$. One can choose the resulting fields totally real via a quadratic twist. The degree $14$ polynomial is found through a genus one Hurwitz family of degree 14 covers. The degree 20 polynomial is found using Yang's modular equations for the Shimura curve $X_6^*(1)$.
Read the full abstract
Fluid Dynamics
Mach-number-dependent dissipative anomaly in isothermal compressible turbulence
Using a comprehensive set of three-dimensional, high-resolution direct numerical simulations, we investigate the existence of a dissipative anomaly in isothermal, homogeneous, isotropic compressible turbulence driven by solenoidal forcing. We find that the total kinetic-energy dissipation rate, as well as its solenoidal and dilatational components, approaches finite asymptotic values with increasing Reynolds number $Re$. The normalized mean dissipation rates collapse onto two distinct branches: one corresponding to the subsonic and transonic regimes, with root-mean-square Mach numbers ($M_{\rm rms}\lesssim 1$), and another to the highly supersonic regime, with ($M_{\rm rms}\ge 3$). This two-branch Mach-number dependence is most pronounced for the total kinetic-energy dissipation. For the solenoidal and dilatational dissipation rate components, the dependence on $Re$ depends in addition on the specific choice of the integral scale and root-mean-square velocity. Despite grid resolutions of up to $2048^3$ points the Reynolds numbers accessible are not sufficiently large to distinguish conclusively between a weak and a strong dissipative anomaly. We substantiate these findings using three complementary approaches: (i) a detailed analysis of the mechanisms responsible for dissipation generation, based on the corresponding dissipation-rate balance equations and their individual production terms; (ii) an investigation of precursors of anomalous dissipation using the Duchon–Robert framework extended to compressible flows; and (iii) a geometrical characterization of regions of intense dissipation. Taken together, these analyses provide consistent evidence for the existence of a dissipative anomaly in isothermal compressible turbulence, while leaving its precise weak or strong character unresolved.
Read the full abstract
Quantum Physics
On the geometry, typicality, and detectability of quantum magic
Quantum magic, or nonstabilizerness, is a central resource for quantum computation, motivating a systematic study of its mathematical structure and physical behavior. Here we characterize the geometry of the stabilizer polytope, obtaining near-tight bounds on its inradius, volume radius, and number of facets. These results reveal a pronounced separation between the typical presence of magic and its detectability. In particular, random mixed states obtained by tracing out an environment remain magical with high probability even when the environment contains nearly twice as many qubits as the system, demonstrating parametrically greater robustness to environmental mixing than bipartite entanglement. In contrast, commonly used magic criteria, including the Triangle and Pauli-norm-based criteria, lose detection power under much weaker mixing. More generally, when the number of environment qubits scales proportionally with the system size, certification with constant success probability requires exponentially many linear observables, even given their exact expectation values. We also prove that the number of facets of the stabilizer polytope grows doubly exponentially with the number of qubits, making exhaustive witness enumeration impractical for complete magic certification and motivating approaches that exploit the structure and symmetries of stabilizer states.
Read the full abstract
Artificial Intelligence
More Criticism Does Not Make a Better Review: EquiReview-R
AI reviewers can now produce many specific criticisms, but more criticism is not necessarily a better review. A review may miss a consequential weakness or retain an allegation that available evidence does not support. These failures require opposite corrections, yet generation-oriented systems and aggregate measures obscure the distinction. We therefore recast AI-assisted review as evidence-guided refinement of a structured concern set, with omission and overcritique treated as separate risks. Building on this formulation, we introduce EquiReview-R, which resolves existing concerns against localized evidence, searches for missing issues from independent and review-conditioned perspectives, and returns stop, continue, or defer. To expose the failure mode that motivates this design, we construct an evidence-linked trajectory corpus. Its retrospective analysis shows why revision must precede further search: nearly all concerns in a high-recall review lack a definitive evidential disposition, while an earlier refinement mechanism cannot revise them. On a frozen cohort of previously unseen papers, EquiReview-R satisfies the prespecified non-inferiority criterion for major omission, reduces major overcritique from 15.5% to 8.1%, and attains a one-sided omission upper bound of 9.9% while stopping on 52.4% of papers. Computation-matched controls, controlled pairs, and ablations show that the gain comes from revision rather than extra inference or shorter output. We release the corpus as ReviewTrace, an evidence-linked resource for studying review revision, disagreement, and provenance.
Read the full abstract
Chemical Physics
Diabatic Seam Space Sampling for Hydrogen Tunneling Systems with Nuclear-Electronic Orbital Theory
Hydrogen tunneling is central to many chemical and biological processes. Herein, we introduce the vibronic-SHAKE (V-SHAKE) approach to comprehensively sample energy-conserving molecular configurations enabling hydrogen tunneling. A constrained form of nuclear-electronic orbital multistate density functional theory (NEO-MSDFT) dynamics, where the tunneling hydrogen nucleus is quantized, is used to sample geometries in the diabatic seam space corresponding to the intersection of the reactant and product NEO-DFT diabatic vibronic surfaces. V-SHAKE is applied to hydrogen and deuterium tunneling in 4-cyanobutanolate and Z-4-hydroxybut-3-en-2-one. The vibronic coupling is found to vary significantly in the diabatic seam space, mainly due to changes in the donor-acceptor distance. The reaction coordinate and gradient of the vibronic coupling at the minimum energy crossing point are nearly orthogonal and are dominated by motions stabilizing the product relative to the reactant or decreasing the donor-acceptor distance, respectively. V-SHAKE provides fundamental insights and validation for assumptions underlying rate theories.
Read the full abstract
Machine Learning
Headroom-Drift Replay: A Primitive for Principled Replay Control in GRPO
RL-based post-training for reasoning models is increasingly bottlenecked by repeated fresh rollout generation, particularly in agentic settings where environment interaction dominates wall-clock cost. Replay can reduce this burden by reusing past trajectories, but existing methods typically embed it within larger training pipelines involving exploration, experience restructuring, or mixed-policy optimization. This makes replay's own contribution difficult to isolate. We ask a focused question: how far can principled replay selection alone go? We introduce Headroom-Drift Replay, a group-level replay control primitive for GRPO that separates reuse into two decisions. Headroom ranks stored groups by remaining learning value, while Drift gates them by compatibility with the current policy. The fresh on-policy stream remains unchanged, and the method adds no auxiliary generation or training machinery. Across mathematical reasoning, multimodal reasoning, and Agentic Search benchmarks, this single intervention outperforms naive replay and matches or exceeds broader replay methods on Avg Mean@32. In Agentic Search, where environment interaction dominates cost, it delivers comparable quality at materially lower wall-clock time.
Read the full abstract
Sound
Masked Autoregressive Speech Enhancement with Continuous Neural Audio Codec Representations
Most previous work on speech enhancement (SE) based on masked generative modeling relied on discrete token representations of audio signals, obtained using neural audio codecs (NACs). However, a recent study has shown that continuous latent representations of NACs can be advantageous for SE in terms of speech quality and intelligibility. In this work, we propose masked autoregressive SE (MARSE), a method for SE based on iterative decoding of masked clean speech frames using continuous NAC representations of speech. In particular, we investigate a set of different decoding policies, ceteris paribus, that is, using the same DNN (a Conformer model), the same NAC (the DAC codec) and the same training setup. The results show that MARSE enables a flexible trade-off between SE performance and computational cost. Audio examples and code are available online.
Read the full abstract
Number Theory
Arithmetic probability measures
We consider probability measures on compact subsets of the complex plane arising as limiting distributions of Galois conjugates of certain algebraic integers. These measures are characterized by infinitely many integral inequalities, one for each nonzero integer polynomial. We study a broad family of measures given by particular convex combinations of harmonic and equilibrium measures. We then show that, for measures in this family, it suffices to verify two conditions to guarantee all the required inequalities. Under additional arithmetic constraints on the underlying compact sets, we further show that the infinite collection of inequalities is equivalent to checking the inequalities associated with two explicitly specified integer polynomials.
Read the full abstract
Artificial Intelligence
Towards Numerical TOHTN Planning with SMT-based HTN-SAT Encoding
While HTN planning has received significant attention in recent years, support for numerical reasoning remains very limited. In this paper, we investigate numerical Totally-Ordered HTN (TOHTN) planning and show how standard SAT-based encodings can be naturally extended with SMT to handle numeric fluents. In addition, we introduce a benchmark suite for numerical TOHTN planning, providing a first common basis for evaluation in this setting. Experimental results show that this simple encoding already constitutes a competitive baseline. This work opens the way to more expressive approaches to HTN planning.
Read the full abstract
Machine Learning
RATL: Learning from Retrieved Residuals for Robust Multivariate Time-Series Forecasting
Retrieval-augmented generation (RAG) complements parametric models with retrieved external evidence. The same idea is attractive for continuous-output regression, but directly reusing retrieved target values is often not robust when samples differ in output level, numerical scale, or local dynamics. Moreover, conventional forecasting pipelines generally use residuals for model optimization and error diagnosis, but do not retain individual historical residual examples as memory that can be accessed at inference time.For multivariate time-series forecasting, we propose RATL, a plug-in residual-retrieval and feedback-correction method. RATL freezes a base forecaster to construct retrieval keys and turns its historical forecast residuals into a train-only memory specific to that base model. At inference time, RATL retrieves residual trajectories from similar historical contexts subject to causal availability constraints, then uses a set-aware router operating over forecast blocks and variables to select and combine these trajectories. Experiments show that historical residuals matched to the current context contain reusable forecasting information and that RATL improves frozen base forecasters in most experimental settings. Ablations further show that learned routing strengthens raw residual feedback, while validation-based correction-strength selection limits residual over-injection.On real-world benchmarks, we use iTransformer as the primary frozen base forecaster, compare against multiple strong forecasting baselines, and test transferability across backbones. The results show that RATL can further improve base-forecaster performance in most settings.Overall, RATL shifts the retrieved object from historical target values to base-model-specific historical forecast errors, providing a plug-in, residual-memory-based paradigm for learned feedback correction in continuous-output forecasting.
Read the full abstract
Computers and Society
Making Gender-Inclusive Practices Actionable: Evaluating a Research-Informed Computing Education Toolkit
The persistent gender imbalance in computing remains a global concern, and universities offer a key part of the pipeline to address it. Although research has identified practices that support under-represented student groups, translating this evidence into actionable guidance remains challenging. This paper first presents a novel web- based toolkit (TechMate) designed to address this gap by helping computing educators implement gender- inclusive initiatives through practical research-informed guidance. The toolkit defines over 25 actions, ranging from operational to strategic, and provides case studies and implementation resources. Second, this work reports on the evaluation of TechMate, capturing educators first impressions of its usefulness and usability through authentic tasks, and eliciting unanticipated insights about structural barriers to gender-inclusive practice in computing higher education. Eighteen computing educators of varying seniority from eight universities in Ireland assessed the toolkit using nine research-driven attributes: four usefulness measures (novelty, relevance, trustworthiness, actionability) and five usability measures (effectiveness, aesthetics, navigation, terminology, user satisfaction). Data was analysed using mixed methods: deductive coding, quantitative sentiment scoring, and inductive thematic analysis. Educators rated TechMate highly for user satisfaction and praised its novelty and trustworthiness, with over 60% of participants reporting positively on all attributes, suggesting that research -informed tools can be designed for real-world use. Thematic analysis highlighted other challenges, including women students isolation, institutional resistance to change, and the scarcity of women role models, while also surfacing proposed solutions such as leveraging student ambassadors and fostering supportive learning environments.
Read the full abstract
General Relativity and Quantum Cosmology
On the interplay between waveform systematics and lensing signatures in gravitational-wave signals
The exceptional gravitational-wave event GW231123 exhibited unusually high total mass and near-extremal spins, while revealing strong waveform-dependent variations in its inferred source properties. Several follow-up studies favored a lensed interpretation, with reduced waveform systematics, raising the question of whether apparent lensing evidence could arise from waveform modeling systematics. Using GW231123-like numerical-relativity injections, we find that lensing degrees of freedom can reduce waveform discrepancies and yield non-negligible lensing support. However, this behavior is not systematic across injections, and cases where all waveform and lens models favor lensing while reducing waveform discrepancies are not found.
Read the full abstract
Combinatorics
A note on bounded ratios
We prove that the set of bounded ratios $\BR(X)$ on a semialgebraic set $X\subset\R^n_{>0}$ is the convex cone of linear forms that are nonnegative on the tropicalization $\trop(X)$. In particular, it is a rational polyhedral convex cone. For $X$ the set of Lorentzian polynomials with fixed M-convex support, it is the dual to the set of M-convex functions. We record an explicit counterexample to a conjecture of Huang–Huh–Soskin–Wang on the bounded ratios on Lorentzian polynomials. The bounded ratio in the counterexample corresponds to the non-hypermetric clique-web facet $\mathrm{CW}^1_7(1,1,1,1,1,-1,-1)$ of the cut cone on seven vertices.
Read the full abstract
Statistical Methodology
Causal Inference for Heterogeneous Extreme Quantiles with Heavy-Tailed Outcomes
We propose a framework for estimating conditional extreme quantile treatment effects (CEQTEs) in observational studies with heavy-tailed outcomes. Our procedure first estimates intermediate conditional quantiles using inverse-probability-weighted (IPW) quantile regression and then extrapolates them to extreme levels using extreme value theory. Under a linear conditional quantile model, we show that the conditional and marginal distributions of each potential outcome share a common extreme value index (EVI), motivating two complementary Hill-type EVI estimators based on conditional and marginal information, respectively. On the theoretical front, we introduce an IPW tail quantile score process that bridges regression quantile score processes and uniform tail empirical processes while accounting for treatment assignment. We establish its functional weak convergence under mild regularity conditions, without requiring a max-domain-of-attraction condition. This result provides the probabilistic foundation for the asymptotic analysis of the proposed CEQTE estimators. Simulation studies demonstrate favorable finite-sample performance, and an application to NLSY79 data reveals substantial heterogeneity in the effect of college education on extremely high hourly wages across confounder-defined subpopulations.
Read the full abstract
Geometric Topology
Khovanov monodromy groups via motions
For any link, we define a monodromy map from the motion group of the link to the group of automorphisms of the link's Khovanov homology. The image of this map is the unoriented monodromy group of the link. This map allows us to convert results concerning motion groups of links into ones about their Khovanov monodromy. In particular, we use a characterisation of the motion groups of split links to write their unoriented monodromy groups as an explicit semidirect product in terms of their unsplit pieces. We give some demonstrative examples, computing the unoriented monodromy groups of unlinks, Hopf links, and split links composed of pieces thereof.
Read the full abstract
Computer Vision
Sparse auto-regressive modeling for scene generation from multi-view images
Generating complete 3D scenes from sparse, unconstrained views is a fundamental challenge in 3D vision which requires reasoning beyond observed content while remaining computationally tractable. Existing feed-forward reconstruction methods are inherently limited to content visible in the input images, while 3D generative modeling is hindered by the high computational cost of dense volumetric representations and the scarcity of large-scale 3D supervision. We introduce SPAR3S, a sparse voxel-aligned 3D latent generative model for conditional scene completion without requiring ground-truth 3D data for supervision. Our key insight is to formulate 3D scene generation in a structured, compact, voxel-aligned 3D latent space where only occupied voxels are represented. We learn this sparse latent space directly from multi-view images using photometric supervision via differentiable 3D Gaussian Splatting. Given a partial set of observed voxels encoded from sparse input views, scene completion reduces to predicting the missing latent tokens and their spatial support within the voxel grid. To this end, we train a masked autoregressive transformer that jointly models voxel occupancy and latent token values, enabling efficient and spatially consistent generation of unseen regions. We demonstrate the effectiveness of our method on synthetic indoor scenes, achieving higher novel-view quality than prior work. We further validate its generalization on RealEstate10k, highlighting its applicability to real-world data.