High Energy Physics (Phenomenology)
Papers
Fresh from arXiv, typeset properly. Highlighted words trade on the market.
All of arXiv
Friday 6 November
We study the production of spin-one $D_{s0}^*(2317)\bar D_s^*$, $D_{s1}(2460)\bar D_s^*$, and $D_{s1}(2536)\bar D_s^*$ molecular candidates in $B_s^0$ decays with an accompanying $\eta$, providing a complementary probe of their structure and production mechanisms. The meson-loop rescattering amplitudes are constructed from factorized weak sources and heavy-hadron chiral interactions. Constituent couplings are estimated using threshold-dependent compositeness relations, and the associated input uncertainties are propagated to the predicted rates. Opposite-$C$-parity partner hypotheses are included with inherited benchmark parameters. We calculate the branching fractions for two relative signs. Over the cutoff range $2\leq\alpha\leq5$, the larger central branching fractions reach several $10^{-6}$ to several $10^{-5}$, suggesting potential accessibility in future experimental measurements.
Computation and Language
AlienBench: A Benchmark of Structural Novelty in Language Model Outputs
Large language models (LLMs) are increasingly used for creative tasks such as writing, design, and idea generation. For these uses, it matters whether a model's ideas depart from familiar examples or stay close to them. Novelty is one standard measure of creativity, and current creativity evaluations ask human experts or LLM judges to rate how novel an output is. However, these ratings have no fixed reference for what counts as novel (or, in other terms, departure from the familiar). Measuring this departure requires a task with known familiar defaults. The structured imagination task from cognitive psychology is such a task. In this creativity test, people imagine a creature from another planet. The departure from features of Earth animals, such as bilateral symmetry and legs as mode of locomotion, is used as measure of creativity. We introduce AlienBench, a benchmark that adapts the structured imagination task to language models. Models describe alien creatures, and a panel of three LLM judges extracts ten biological features from the description and labels them as either Earth-typical or a departure. The sum of feature departures is the score of a description, from 0 to 10. Each model describes alien creatures under six prompt conditions that vary the framing of the task and the planetary environment. Each model generates 30 descriptions per condition (180 in total), and its overall score is the mean across these descriptions. We evaluate 10 language models, from frontier to open-weight models. Overall scores range from 3.28 (GPT-6 Luna) to 6.55 (Ministral 3 14B). Interestingly, open-weight models score higher than proprietary models in this task (5.40 vs. 4.58). Under the baseline prompt, the models depart from Earth defaults more often on average than 100 human participants on every feature. We release all prompts, generations, judge outputs, and scores.
Read the full abstract
Neural and Evolutionary Computing
Two-Layer Evolutionary Scheduling for Machine Learning Jobs in Multi-Cluster Environments
The rapid growth of large-scale machine learning (ML) workloads has led to unprecedented demands on GPU cluster scheduling systems. However, existing heuristic schedulers struggle to handle the complexity and dynamics of multi-cluster GPU environments, often leading to load imbalance, long queuing delays, and poor alignment to business objectives. This paper formulates the ML scheduling problem in multi-cluster GPU environments as online multi-objective optimization that minimizes the average and maximum weighted queuing times while maximizing GPU time utilization. To solve this optimization problem, we propose EVO-GP, a two-layer evolutionary online scheduling method that decouples global job-to-cluster allocation from intra-cluster GPU resource allocation. At the job-to-cluster allocation layer, an evolutionary algorithm assigns arriving jobs to feasible clusters to promote load balance across clusters. At the resource allocation layer, a priority function automatically learned offline via genetic programming is used to rank jobs and guide GPU allocation within each cluster. Extensive experiments on both real-world data and simulated workloads demonstrate that the proposed EVO-GP method consistently outperforms state-of-the-art baselines, while maintaining superior adaptability and scalability.
Read the full abstract
Computer Vision
AdaAlloc: Adaptive Visual Token Allocation for Long-Video Question Answering
Long-video question answering forces multimodal large language models to reason under strict visual-token budgets, creating a tradeoff between broad temporal coverage and fine-grained spatial detail. Uniform sampling preserves coverage but often obscures decisive local cues, while caption-based or frame-similarity retrieval operates over lossy proxies that can miss motion, state changes, and before-after relations. We recast long-video reasoning as a visual-token allocation problem: given a fixed budget, decide how much to spend on global context versus high-resolution local evidence, and where in the video to draw that local evidence from. We introduce AdaAlloc, a training-free inference framework that adaptively allocates visual tokens between a global video overview and localized visual detail. AdaAlloc first plans a global/local allocation policy, then uses temporal grounding to locate candidate evidence segments and refines them into a compact temporal memory. The selected policy controls both frame sampling density and resolution, combining sparse, lower-resolution global frames with denser, higher-resolution local frames drawn from the refined memory. The final answerer reasons over the allocated global context and refined local evidence. Across LVBench, MLVU, VideoMME, and LongVideoBench, AdaAlloc improves over budget-matched uniform sampling by up to 10.5 percentage points and outperforms the compared frame-selection and agentic methods on their reported benchmarks.
Read the full abstract
Machine Learning
TUGB: A Two-stage Uncertainty-aware Generative Auto-Bidding Framework
Generative modeling, particularly Decision Transformers, shows promise in auto-bidding strategies. However, it struggles with limited exploration in static offline datasets, vulnerability to dynamic environmental shifts, and a reliance on heuristic return-to-go (RTG) specifications that may yield suboptimal performance on unseen traffic. To address these challenges, we propose the Two-stage Uncertainty-aware Generative Auto-Bidding (TUGB) framework. TUGB integrates generative sequence modeling with Reinforcement Learning (RL). It employs a two-stage pipeline where a pre-trained Generative Bidding Model is fine-tuned via an ensemble-based Uncertainty-aware Environment Model. This approach quantifies epistemic uncertainty to penalize high-risk trajectories and evolves the policy from RTG-conditioning to direct bidding optimization. Extensive experiments demonstrate that TUGB significantly outperforms SOTA baselines. Notably, full-scale deployment on JD Search Ad achieved a 2.32% increase in cost and 3.57% in target cost, validating its industrial efficacy, while its successful adaptation to subsequent mechanism upgrades and JD Food Delivery Ad further validates TUGB's superior generalization capability.
Read the full abstract
Computation and Language
MARS: Training Multilingual Search Agents for Retriever-Free Evidence Acquisition
Existing multilingual retrieval-augmented generation relies heavily on learned retrievers and similarity-based ranking from multilingual collections, which are vulnerable to cross-lingual retrieval errors and indexing overhead. In this paper, we explore retriever-free multilingual search, where the model directly interacts with raw corpora through executable shell commands. This paradigm shifts the bottleneck to multilingual search action generation, where exact search is sensitive to entity surface forms and outcome-only supervision provides limited guidance for multi-turn search. To address these challenges, we propose MARS, a multilingual search agent in direct corpus interaction (DCI) with reinforcement learning (RL). MARS introduces corpus-native entity grounding to recover reliable lexical anchors from corpus evidence, mitigating failure to retrieve evidence. Meanwhile, we present a redundancy-aware search reward to encourage concise and effective search trajectories, addressing training collapse in RL of multilingual scenarios. Experiments show that \ours~consistently outperforms strong retrieval-based and DCI baselines, while demonstrating robust cross-lingual generalization and substantially reducing redundant corpus interactions.
Read the full abstract
Distributed Computing
EventHOI: Event-Driven Resource-Efficient Hand-Object Interaction Logging on Smartglasses
Hand-object interactions (HOIs) can be captured by smartglasses to learn how people manipulate objects in daily life. Automatic logging of these interactions from an egocentric perspective provides time-stamped evidence of daily actions for applications like memory assistance. However, continuous RGB capture and wireless transmission exceed the energy and bandwidth budgets of smartglasses. We present EventHOI, an event-driven sensing and compression system that couples selective RGB capture with HOI-aware encoding to enable egocentric logging over Bluetooth Low Energy (BLE) while preserving downstream HOI logging performance. Our key insight is to use low-power event sensing both to determine when RGB capture is needed and to identify where visual detail must be preserved. A lightweight IMU-based egomotion cancellation pipeline suppresses egomotion-induced events so that residual HOI motion can guide RGB triggering and spatial compression. We build a smartglasses prototype and conduct an end-to-end evaluation of EventHOI on a 50-hour egocentric benchmark with 97 subjects and a 20-hour real-world dataset we collect from 10 participants, together covering a broad range of daily-life scenarios and environments. Compared to RGB-only baselines, EventHOI reduces always-on backend power consumption by 5.8-10.4x and uplink bandwidth by 2.9x, with minimal HOI recognition accuracy loss. The resulting RGB uplink of less than 130 kbps fits within commodity BLE bandwidth budgets, making stable BLE video streaming feasible for egocentric HOI logging. Demo video and project website: https://eventhoi.github.io
Read the full abstract
Software Engineering
When Routine Operations Break Architectural Pattern Relationships in LLM Agent Harnesses
LLM agent harnesses distribute tool mediation, state management and safety guardrails across native services, independently written plugins and configuration. Routine operations can change the runtime order or retained state that architectural pattern relationships require, without any code change. We study five relationships associated with the Guardrail Enforcement Point, Autonomy Controller and Blackboard patterns across five released harnesses, analysing a main sample of fourteen independently written plugins. Using PatternOps, our assessment procedure, we trace requirements from pattern descriptions, provider documentation and deployment profiles to the providers that realise them and the records that decide them. We then compare alternative allocations under supported operations. Such operations broke relationships silently: after a plugin was reapplied or an allow rule saved, policy stopped evaluating edits while every plugin stayed active, and a prohibited edit executed. Which alternatives qualified depended on the requirement. Allocations that kept the required order, or kept or rebuilt the needed state, held on their tested paths and operation sequences, while others prevented prohibited effects but missed edits rejected elsewhere. In these harnesses, a pattern relationship is a property of the running system and its history, so architects should evaluate it against the decision paths, state lifetimes and operation sequences their deployments use.
Read the full abstract
Statistics Theory
Group Invariance of Statistical Models and Associated Statistical Manifolds
We study the group invariance of a parameterized statistical model that is closed under push-forward sample transformations given by a group action. We prove that the group action on the sample space induces an isometric action on the associated statistical manifold. Furthermore, it preserves the Fisher-Rao metric as well as the Amari-Chentsov tensor. Moreover, we establish that this isometric action is compatible with equivariant and sufficient statistics in the spirit of the Hall-Wijsman-Ghosh generalized Stein's theorem. This compatibility leads to a novel reduction of the original group-invariant model, where its Fisher information loss is explicitly quantified via conditional expectation. As an application, we apply this framework to a generalized cardioid distribution in circular data analysis, illustrating how the reduced model facilitates estimation and hypothesis testing for relative parameters.
Read the full abstract
Astrophysics of Galaxies
SEDust: A DUSTY-Based SED-Fitting Pipeline for Dust and Mass-Loss from AGB and LPV Stars
We present SEDust, an automated pipeline for spectral energy distribution (SED) fitting and the estimation of stellar mass-loss and dust-production rates. We apply this pipeline to quantify the mass-loss and dust injection rates in the dwarf galaxy NGC~6822, focusing on long-period variables (LPVs) and pulsating asymptotic giant branch (AGB) stars as major contributors to dust feedback into the interstellar medium. The total stellar mass-loss rate is found to be $(3.70 \pm 1.16) \times 10^{-3}\ M_{\odot} \mathrm{yr}^{-1}$. Two alternative methods (an LPV-based estimate and DESK) give totals that differ from this value by factors of $\sim$2.3 (lower) and $\sim$2.4 (higher), respectively, which we adopt as an estimate of the method-dependent systematic uncertainty. A positive correlation between stellar luminosity, intrinsic reddening caused by circumstellar dust self-extinction, and mass-loss rate is observed. The specific mass-return rate from AGB stars is $(7.40 \pm 2.30) \times 10^{-11}\ \mathrm{yr}^{-1}$, indicating that mass return from AGB stars alone is insufficient to sustain the current star formation rate of NGC~6822. This result implies the need for external gas accretion or additional mass-return mechanisms.
Read the full abstract
Computation and Language
The Invisible Gap: Romanized Indic Generation Fails Where Automatic Metrics Succeed
Hundreds of millions of Indians communicate with digital systems in romanized Latin characters – typing "aap kaise hain" rather than native script – yet romanized Indic generation is absent from NLP benchmarks and evaluation frameworks. We demonstrate that automatic metrics (BGE-M3, BERTScore) systematically misrepresent the quality of romanized Hindi, Telugu, and Kannada generation. We evaluate four LLMs (Claude Opus 4.8, GPT-5.5, Gemini 2.5-flash, Qwen3.5-35B) on 142 OpenAssistant query-response pairs. On a 72-response diagnostic subset, native speakers rate GPT-5.5 1.43 points higher than Claude Opus 4.8 on a 12-point rubric, whereas their full-dataset BGE-M3 means differ by only 0.004. Two distinct model-specific failure modes emerge: Gemini produces script spillage (16.9% of Telugu responses leak native script), while Qwen collapses to English-only output (16.2% of Kannada responses). We introduce a failure-aware evaluation policy that assigns zero to empty outputs, script-corrupted responses, and instruction collapses – revealing failures metrics alone cannot detect. Our results show that evaluating romanized Indic generation requires a coordinated bundle: semantic similarity, script compliance, failure coverage, and native-speaker acceptability.
Read the full abstract
Computer Vision
Training-Free Active Test-Time Adaptation for Vision-Language Models
Test-time adaptation (TTA) for vision-language models (VLMs) is usually studied in a fully unsupervised regime, where adaptation relies on pseudo-labels, confidence scores, or entropy computed by the model itself. This is a severe restriction on hard shifted samples: the model is asked to repair its own uncertain predictions without any external evidence. At the other extreme, supervised online adaptation assumes labels can directly train or correct the model. We study the intermediate deployment regime between these two extremes. In Active Test-time Adaptation for VLMs (Active VLM-TTA), a pretrained VLM may spend an explicit labeling budget on selected test samples, but the prediction for the current sample must be emitted before its label can be used; queried labels are therefore costed, delayed evidence for future samples rather than free correction for the present one. Under this future-only constraint, the central question is how to reuse sparse labels conservatively. We instantiate this setting with ACTive VLM-TTA for Online Retrieval (ACTOR), a training-free plug-in module that routes uncertain samples among base prediction, label query, and memory-based correction using a consistency-and-reliability check over previously queried labels. Across ten cross-domain benchmarks and a domain-transfer benchmark based on ImageNet, ACTOR consistently improves strong VLM-TTA baselines on ResNet and ViT backbones with negligible inference overhead. Our implementation is publicly available at https://github.com/kaist-dmlab/ACTOR.
Read the full abstract
Mesoscale and Nanoscale Physics
Strong phonon-coupled trion emission in Janus homobilayer
Janus transition metal dichalcogenides with broken mirror symmetry provide a novel platform for exploring many-body effects mediated by spin-valley-phonon interactions. Strong electron-phonon coupling demands a precise match between a strong effective dipole moment and nanometer-scale charge confinement. Here, we report thermally robust and enhanced trion emission in a symmetric bilayer WSeS with an S/S stacking configuration. The inward-pointing antiparallel built-in electric fields induce a strong electron confinement effect at the interface, driving dominant trion emission at cryogenic temperatures. Furthermore, a massive electron-phonon coupling strength (3.9) resonant with the optical phonon A_1^2 energy (51 meV) contributes to the thermal stability of this trion emission at elevated temperatures. By tailoring the interfacial architecture in Janus WSeS homobilayers, this study overcomes the thermal decoherence limits inherent to traditional semiconductor systems, offering novel insights into room-temperature many-body interactions.
Read the full abstract
Machine Learning
BrainCoT: A Multi-Task Zero-Shot Brain Signal Foundation Model with Neurometric-Anchored Chain-of-Thought Reasoning
Brain signals exhibit substantial multi-center heterogeneity, strong inter-subject variability, and highly diverse task settings, which makes it difficult to build models that generalize beyond a single task. While recent brain signal foundation models leverage large-scale pretraining to learn generalizable representations, they often fall short in practice because they still require target-data-specific supervised adaptation, lack multi-task zero-shot capability, and rarely provide verifiable evidence to support their predictions. In this work, we present BrainCoT, a multi-task zero-shot Brain signal foundation model with neurometric-anchored Chain-of-Thought (CoT) reasoning. BrainCoT unifies heterogeneous tasks into an instruction-driven framework so that, after pretraining, a single model can be directly applied to multiple tasks in zero-shot settings without additional adaptation. It further introduces neurometric-anchored chain-of-thought grounded in neurometric evidence for verifiable reasoning, and adopts Decision-Consistent Rationale Training to keep process learning aligned with correct predictions. Across extensive evaluations, BrainCoT achieves an average ACC of 67.41% and AUC of 74.33% in zero-shot classification without downstream training data, outperforming the strongest linear-probing baseline trained with 1% labeled downstream data by 3.85% in ACC and 5.97% in AUC.
Read the full abstract
Quantum Physics
Quantum-Enhanced Classical Communications for DPMZM-based NG-PON2 Systems
This paper presents a framework for integrating quantum detection techniques into a Dual-Parallel Mach-Zehnder Modulator (DPMZM)-based Next Generation Passive Optical Network 2 (NG-PON2) communication system. We present a system model for the DPMZM-based NG-PON2 quantum communication system and characterize its signal-to-noise ratio (SNR), demonstrating its superiority over conventional MZM-based systems. We analyze the performance of various quantum modulation techniques, including binary phase-shift keying, quadrature phase-shift keying, and M-ary quadrature amplitude modulation. A comprehensive analytical framework is developed to evaluate the SNR, BER, and mutual information for both classical and quantum detection schemes within the DPMZM-based NG-PON2 quantum communication system. Our simulation results show that quantum detection significantly outperforms classical detection, and the DPMZM architecture further enhances the performance over the baseline MZM architecture by improving the effective modulation efficiency and receiver sensitivity.
Read the full abstract
Algebraic Geometry
Cuspidal sections for varieties of arbitrary dimension
We define cuspidal sections for varieties of arbitrary dimension over fields of characteristic zero, extending the definition for curves by Grothendieck. Our notion of cuspidal sections is defined relative to a normal crossings compactification. We prove that the closure of the resulting set is independent of the choice of a compactification. For varieties iteratively constructed from fibrations, we compare our definition to an inductive notion of cuspidal sections, and discuss connections to the section conjecture of Grothendieck.
Read the full abstract
Signal Processing
DeepRT-Cal: Learning to Calibrate Ray-Traced Twin Channel Statistics with Low-Fidelity Environmental Geometry
Sixth-generation (6G) wireless systems increasingly rely on digital twin channels (DTCs) to represent and interact with physical radio environments. Ray tracing (RT) is naturally suited to DTC construction because it provides deterministic and physically interpretable channel predictions. However, high-fidelity three-dimensional environment models are expensive to obtain, whereas geometric inaccuracies in low-fidelity models can alter the detected multipath components (MPCs) and consequently distort received-power and delay-domain twin channel statistics. Building on the DeepRT ray-tracing framework, this paper proposes DeepRT-Cal, a geometry-aware cross-fidelity calibration framework that predicts high-fidelity twin channel statistics from low-fidelity geometry and RT features.The network encodes global link descriptors, masked low-fidelity MPC tokens, propagation-corridor geometry, and link-centered environmental objects. During training, detailed geometry is introduced as privileged information with structured masking, while deployment requires only low-fidelity geometry and low-fidelity RT information. A link-conditioned cross-attention module retrieves the environmental context relevant to each Tx–Rx link. The prediction heads estimate high-fidelity twin channel statistics, including total received power, RMS delay spread, mean excess delay, and a normalized power-weighted power-delay profile (PDP). Outdoor evaluation shows that, in an non-line of sight (NLoS) region, the method reduces power MAE from 16.80 to 3.88 dB and RMS delay-spread MAE from 304.89 to 98.19 ns. In a LoS-dominated region, power prediction improves, while RMS delay-spread accuracy remains comparable to direct RT.
Read the full abstract
Group Theory
Slim Polygons, Hyperbolicity, and a new Quasi-Isometric Invariant
We introduce the slimness function, $\mathrm{Slim}_X(n)$, of a geodesic metric space, which records the maximal slimness of its geodesic polygons as a function of the number of sides. We study the large-scale behaviour of this function and its relation to hyperbolicity, growth function, and quasi-isometries. We first establish that the slimness function is a natural quasi-isometric invariant in the hyperbolic setting. We then obtain quantitative estimates for geodesic polygons in hyperbolic spaces and bounded-valency graphs. In particular, we show that sufficiently slow growth of the slimness function forces tree-like geometry. We also prove a decomposition result for trees of hyperbolic spaces. We show that for a finitely generated hyperbolic group $G$ there are only two possibilities: either $G$ is virtually free and $\operatorname{Slim}_G$ is bounded, or $\operatorname{Slim}_G(n)\simeq\log n$. Finally, we investigate the interaction between polygonal slimness and the growth of connected bounded-valency graphs. Combining the resulting estimates with large-scale dimension theory, we obtain criteria implying that an unbounded graph has asymptotic dimension one.
Read the full abstract
Computer Vision
GlobalPuzzle: Globally Informed Genetic Search for Eroded Jigsaw Puzzle Reassembly
Jigsaw puzzle reassembly remains a highly challenging task, particularly when piece boundaries are irregularly eroded to simulate damage to cultural artifacts. On the GAP benchmark, existing methods achieve a perfect assembly accuracy of only 0.3% on $5 \times 5$ puzzles, whereas humans can solve jigsaw puzzles with remarkably high accuracy. Humans typically approach puzzle solving by identifying potentially adjacent pieces, evaluating the current assembly, and iteratively performing complex operations such as adding pieces, replacing incorrectly placed pieces, and moving entire connected groups of pieces. However, existing methods face a fundamental limitation: approaches that leverage global visual information typically predict piece arrangements directly, making it difficult to explicitly perform such complex iterative operations, while search based methods can rearrange pieces but primarily rely on pairwise boundary compatibility, lacking global visual information to evaluate and guide the search process. To bridge this gap, we propose GlobalPuzzle, a jigsaw puzzle reassembly framework that integrates global canvas level evaluation, action guidance, and genetic search. By incorporating global visual information into the evaluation and optimization of candidate assemblies, GlobalPuzzle aims to resolve ambiguous adjacency relationships that are difficult to determine from irregularly eroded piece boundaries alone, enabling flexible, globally informed reconstruction.
Read the full abstract
Combinatorics
Quantitative group testing with a fixed shore: sharp asymptotics and exact formulas
We study adaptive quantitative group testing with one defective item in each of two disjoint sets. When one set has fixed size, we determine the leading term of the difference between the binary bound and the largest searchable opposite set size. The result extends to fixed unions of complete bipartite graphs and to positive weighted sums of the opposite set sizes. We determine exact eventual formulas when the fixed set has six, seven or eight items. The eight-item formula has a period-three correction. We explain this correction by optimizing the sizes of shared test pieces within fixed decision-tree prefixes. For each fixed finite family of such prefixes ending in star forests, the optimal deficit is eventually a rational polynomial plus a bounded periodic function. The asymptotic proofs are analytic; the exact formulas also use integer constructions and rational exclusion certificates.
Read the full abstract
Machine Learning
Large Language Model Deployment on Resource-Constrained Edge Devices: A Practitioners Survey
Large Language Models (LLMs) have transformed many areas of artificial intelligence (AI), but their substantial resource requirements limit their deployment on mobile and edge devices. Although LLMs have demonstrated impressive capabilities in various applications, the next challenge is deploying these models on resource-limited edge devices. Recent innovations in software optimizations and specialized hardware have made it feasible to deploy LLMs on edge devices. Edge deployment offers benefits including local data processing, cost savings, reduced latency, and personalized responses. This paper reviews the latest techniques that make the deployment of LLMs on edge devices feasible.
Read the full abstract
Computer Vision
Emo-LiveSync: Real-Time Universal Lip Synchronization with Explicit Emotion Control
Lip synchronization, the task of generating realistic lip movements synchronized with audio signals, is fundamental to applications such as film dubbing, virtual avatars, and digital content creation. Recent GAN-based methods offer computational efficiency but often lack visual fidelity, while diffusion-based approaches achieve high-quality results at the cost of substantial computational overhead. Moreover, most existing methods are limited to photo-realistic human faces, restricting their applicability to stylized characters and non-human entities. In this paper, we propose an efficient universal lip-sync video generation framework that achieves high-quality results without compromising computational efficiency. We build on LivePortrait, a warp-based portrait animation model, and introduce a training strategy that aligns audio signals to an implicit motion space by distilling knowledge from a pre-trained video-to-motion model. This enables real-time inference while maintaining high visual fidelity, and naturally generalizes to diverse stylistic characters beyond photo-realistic humans. Furthermore, we introduce an expressive control gate that enables explicit manipulation of emotional expressions for content enrichment. Extensive experiments demonstrate that our method achieves state-of-the-art performance in both visual quality and lip-sync accuracy while ensuring real-time inference.
Read the full abstract
Functional Analysis
Characterization of Attenuated V-Line Transform and Direct Reconstruction from Samples
The V-line transform (V-LT), a generalized Radon transform (RT), is widely used in Compton scattering imaging (CSI). The central mathematical problem of CSI is to reconstruct a function from its V-LT. Unlike V-LT, the attenuated V-line transform (AV-LT) accounts for the medium-induced loss of intensity, and has been demonstrated to improve image quality. The present paper makes two main contributions. (1) We establish the fundamental characterizations for AV-LT including operator properties ($L^2$-boundedness, operator norm etc.), a necessary and sufficient condition for the preservation of compact support, and sharp estimation of Sobolev smoothness. These characterizations remain unexplored, and they reveal essential differences among AV-LT, RT and V-LT. (2) We develop a direct reconstruction method based on AV-LT samples that does not require solving any equation. Our sampling formula gives exact reconstruction for band-limited functions. We also provide approximate reconstruction for functions in $H^s(\mathbb{R}^2)$ with $s>2$.
Read the full abstract
Combinatorics
Claw-free minimal factor-critical and bicritical graphs
A graph $G$ is called $k$-factor-critical if $G-S$ has a perfect matching for every subset $S\subseteq V(G)$ with $|S|=k$, and is minimal if $G-e$ is not $k$-factor-critical for every edge $e$ of $G$. The cases $k=1$ and $k=2$ correspond to factor-critical and bicritical graphs, respectively. In this paper, we obtain complete structural characterizations of claw-free minimal factor-critical graphs and claw-free minimal bicritical graphs. The former is characterized by ear decompositions. For the latter, we first characterize the $3$-connected case using the theory of line graphs and then extend the characterization to the general case by a suitable $2$-sum operation.
Read the full abstract
Data Structures and Algorithms
Vertex Deletion on $H$-Induced-Minor-Free Graphs: Subexponential FPT Algorithms and Approximation Schemes
We study vertex-deletion problems on string graphs and, more generally, on graphs excluding a fixed graph as an induced minor. For every fixed graph $H$, fixed nonnegative integer $t$, and fixed monadic second-order formula $\phi$, we consider the problem of finding a minimum-cardinality set $S$ such that $G-S$ has treewidth at most $t$ and satisfies $\phi$. This captures classic problems such as Vertex Cover, Feedback Vertex Set, and deletion to graphs of any fixed bounded treewidth. We give an algorithm with running time $2^{k^{2/3} \log^{O(1)} k}n^{O(1)}$ on $H$-induced-minor-free graphs, where $k$ is the optimum deletion-set size. This resolves in the affirmative the open problem highlighted by Bonnet and Rzążewski [Algorithmica 2019] and Berthe, Bougeret, Gonçalves, and Raymond [ICALP 2025] regarding the existence of a subexponential parameterized algorithm for Feedback Vertex Set on string graphs. Our algorithm significantly improves over existing parameterized algorithms for more restricted classes of problems on more restricted classes of graphs. When the property defined by $\phi$ is hereditary, we also obtain a quasi-polynomial time approximation scheme, namely an algorithm that produces a $(1+\varepsilon)$-approximate solution in quasi-polynomial time. Both the sub-exponential algorithm and the quasi-polynomial time approximation scheme build on the same new structural and algorithmic insights.
Read the full abstract
Analysis of PDEs
Complete uniqueness for two-dimensional asymmetric incompressible jet flows
Alt, Caffarelli and Friedman established the existence of two-dimensional asymmetric incompressible jet flows in [Asymmetric jet flows. Comm. Pure Appl. Math. 35, 29–68, 1982]. In their formulation, the constant fluid speed $\lambda$ on the free boundaries and the downstream asymptotic angle $\theta$ are selected as part of the solution. In this paper, we prove that the asymmetric jet is unique for a prescribed nozzle and mass flux. We first establish the strict vertical monotonicity of the stream function $u$ and introduce the partial hodograph coordinates $\xi=x$ and $\eta=u(x,y)$, which map every fluid domain onto the same fixed strip. Under this framework, each jet is characterized by a height function $y=h(\xi,\eta)$, which determines the vertical position along the streamline $u=\eta$ at horizontal coordinate $\xi$. By analyzing the difference between the height functions of two candidate jets on this strip, we derive an elliptic equation with Dirichlet conditions on the fixed walls and conormal conditions on the free boundaries. Combining these mixed boundary conditions with the asymptotic behaviors, we first establish the uniqueness of the free-boundary fluid speed $\lambda$. For the resulting common speed, a local perturbation of the height function near a nozzle endpoint, together with the maximum principle, proves that the two height functions must coincide. Inverting the partial hodograph transformation then establishes the uniqueness of the physical jet.
Read the full abstract
Computer Vision
UniRGBA: Unifying RGB and Alpha in Pretrained Video Diffusion Models
Recent advances in video diffusion models have achieved remarkable photorealistic RGB synthesis, yet generating videos with a native alpha channel remains an open challenge. Existing approaches often rely on matting-style data and non-native RGBA designs, causing them to specialize in real-object scenes and to degrade severely on visual-effect content whose semi-transparency is spatially non-local and non-elliptical. We present UniRGBA, a unified framework whose key insight is that RGBA generation does not require a dedicated alpha encoder or complex latent-space restructuring—a minimal, targeted extension of the existing RGB pathway is sufficient. UniRGBA-VAE extends a causal video VAE to 4-channel RGBA via zero-initialized channel expansion, symmetric selective layer unfreezing (5% trainable), and a six-term progressive reconstruction loss anchored by a KL reference loss that prevents latent drift, followed by a two-stage schedule that introduces a lightweight GAN objective. We then apply standard DoRA to the pretrained DiT, reusing the unified 4-channel latent without any alpha-specific architectural changes. On a visual-effect RGBA benchmark, UniRGBA substantially outperforms Wan-Alpha on RGBA reconstruction while preserving RGB fidelity. For generation, UniRGBA outperforms Wan-Alpha on RGBA text-to-video tasks and matches or surpasses the original pretrained Wan2.1 backbone on RGB-only metrics, demonstrating that native alpha support can be integrated without compromising the underlying RGB prior or the model's architectural versatility.
Read the full abstract
Mesoscale and Nanoscale Physics
Coulomb screening of the orbital inverse Faraday effect in finite-size diffusive conductors
We study the effects of Coulomb screening on the second-order dc response in finite-size diffusive conductors. Starting from the Boltzmann equation supplemented by the self-consistent electrostatic field, we develop a kinetic theory of the inverse Faraday effect for a two-dimensional disk driven by a circularly polarized ac electric field. In this formulation, the first-order charge modulation and the induced electric field are coupled through the Coulomb interaction, and the self-consistent first-order solution determines the nonlinear source for the second-order dc circulating current responsible for the induced magnetic moment. The radial problem is solved numerically using an angular-harmonic expansion and the Chebyshev collocation method. It is shown that in the non-self-consistent approximation the frequency dependence of the second-order dc response has a maximum when the diffusion length associated with the radiation frequency becomes comparable to the sample radius. This behavior is found to be strongly modified by self-consistent screening. We demonstrate that Coulomb screening shifts the maximum of the dc response to a higher drive frequency and suppresses the peak amplitude. This behavior is observed consistently for the area-integrated circulating current and the induced magnetic moment. The obtained results clarify the limitations of the non-self-consistent description of the inverse Faraday effect and show that Coulomb self-consistency is crucial for quantitative predictions of optically induced dc magnetization in finite-size diffusive conductors.
Read the full abstract
Machine Learning
WearableBridge: Task-Guided Latent Transport for Free-Living Activity Recognition
Wrist-worn accelerometers make activity monitoring feasible at scale, but incidental arm motion, device orientation, and shifting context blur the signal that separates activities. Calibration studies supply higher-quality references such as concurrent thigh recordings or supervised laboratory wrist sessions, but they are expensive, cover few participants, and are rarely available at deployment. How can limited reference data improve recognition for participants who wear only a wrist device? We propose WearableBridge, which holds the reference encoder and classifier fixed and learns a task-guided transport map, the bridge, that carries each wrist feature into the region the classifier associates with its activity. Reference features or the classifier's predictions indicate the destination, and a classification loss penalizes wrist features that land in the region of another activity. We instantiate WearableBridge with three bridges: Mapper regresses onto reference features, Distillation matches the classifier's class probabilities, and MeanFlow learns a one-step flow from wrist toward reference features. On 40 held-out free-living PAAWS participants, all three bridges improve macro-F1 in every training run, and the strongest, MeanFlow, raises mean macro-F1 from 35.1% to 56.0% with concurrent thigh references (Thigh2Wrist), and from 54.0% to 63.5% with group-matched laboratory wrist references (Lab2Freeliving). Fitted only on PAAWS, MeanFlow also raises macro-F1 on the CAPTURE-24 cohort from 42.8% to 54.7%. The gains replicate on selfBACK, hold across four reference encoders, and grow with temporal context. A study can therefore collect reference recordings for a calibration subset and improve wrist-only recognition for every other participant, training only a bridge of 0.13 to 2.44 million parameters while the reference encoder and classifier stay unchanged.
Read the full abstract
Dynamical Systems
Hausdorff dimension of planar box-like self-affine sets with rotations
We study the Hausdorff dimension of proximal planar self-affine sets whose defining maps have diagonal or anti-diagonal linear parts, with both types present. Such systems are irreducible but not strongly irreducible. Under a projective exponential separation condition, we show that the Hausdorff dimension equals the minimum of $2$ and the affinity dimension, thereby extending the dimension formula of Hochman–Rapaport to all proximal irreducible planar systems. When the affinity dimension is at most an explicit threshold determined by the Lyapunov exponents, the formula holds assuming only that there are no exact overlaps. Above the threshold, it holds under the strong open set condition. These dimension formulae are proved by approximation with finite diagonal subsystems. Assuming projective exponential separation, we also determine the pointwise dimensions of the coordinate projections of the equilibrium state. Below the threshold, the projections are not exact-dimensional: the pointwise dimension of each exists almost everywhere and takes exactly two values, each on a set of positive measure. At or above the threshold, both projections are exact-dimensional with dimension one.