arXiv subjects

Xinyu Wang

Publications and source records attributed to Xinyu Wang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Whisper-Flash: Acoustically Conditioned Parallel Drafting for Faster Whisper Decoding

Whisper is a widely used encoder-decoder model for speech recognition. Its encoder reads an utterance in one parallel pass, but its decoder writes the transcript one token at a time, which dominates inference time. Speculative decoding shortens such loops without changing their output: a small drafter guesses several upcoming tokens, and the original model verifies them all in one forward pass. We present Whisper-Flash, a two-layer drafter built on a property of speech recognition: the words still to be written have already been spoken. It reads Whisper's encoded audio and accepted decoder states and proposes eight tokens in a single forward pass. On the complete LibriSpeech test sets, Whisper-Flash processes $3.16\times/2.85\times$ as much audio per second as greedy decoding with identical outputs, and it remains faster at batch sizes up to 96 and under temperature sampling. Ablations show that direct access to the audio matters most.

cs.SD↗

arXiv2026

Budgeted Quotient-Residual Guidance for Frozen Pocket-Conditioned Molecular Diffusion

Pocket-conditioned molecular diffusion updates ambient atom coordinates, but many lead-optimization objectives are expressed on quotient features such as distances, contacts, and anchored substructures. We introduce budgeted quotient-residual guidance (QRG), an inference-time correction that makes these quotient objectives active without retraining the molecular generator. QRG lifts quotient covectors to metric-horizontal ambient directions and delivers them through a trust budget set by the frozen sampler's own step norm: quotient geometry chooses the direction, while sampler motion bounds the scale. We derive the horizontal lift, closed-form sampler-budget update, KL/kinetic interpretation around a frozen reverse step, equivariance conditions, and a product-budget split for budget-capped section and residual controls. Controlled quotient tasks confirm that sampler-relative delivery activates signals that raw local quotient gradients leave dormant. On frozen TargetDiff backbones, official seed-0 CBGBench ligand-generation/editing sweeps show practical quality-runtime gains: Local-QRG improves validity from 0.815 to 0.864 on fragment growing, 0.664 to 0.707 on scaffold hopping, and 0.681 to 0.712 on linker design, while PredNext-QRG improves fragment/scaffold and remains near-neutral on linker. Novelty remains 1.000 and diversity is preserved in the matched multi-seed molecular slice, giving task-dependent improvements without sampler retraining or backbone modification. Overall, QRG provides a lightweight route to quotient-aware inference for frozen molecular samplers with explicit runtime accounting.

cs.LG↗

arXiv2026

TopoCompress: Topology Aware Token Compression Algorithm for Distributed Edge MoE Inference

Mixture-of-experts (MoE) models improve capacity with moderate overhead by sparsely activating experts per token. However, deploying MoE across resource-constrained edge servers incurs substantial cross-server communication as experts are distributed across heterogeneous servers. Existing placement methods optimize for raw token traffic, while conventional compression considers semantics but ignores topology-dependent routing costs. Consequently, independent optimization leads to inefficient communication and resource utilization. This paper proposes TopoCompress, a deployment- and topology-aware token compression framework for communication-efficient distributed edge MoE inference. It jointly optimizes token compression, expert deployment/replication, GPU-CPU residency, and collaborative routing to balance cross-server transmission, quality, and resource use. To address the coupling between token-level compression and epoch-level deployment, TopoCompress employs a two-timescale alternating optimization. In the online fast loop, it identifies and compresses low-importance, high-routing-cost tokens and jointly routes surviving expert activations. In the offline slow loop, it updates expert placement, replication, and GPU-CPU residency according to post-compression traffic accumulated during online inference. We establish the feasibility, optimality, convergence, and computational complexity. Simulations demonstrate that TopoCompress effectively reduces cross-server traffic and deployment resource consumption while maintaining controllable inference quality, enabling efficient distributed MoE inference over bandwidth- and resource-constrained edge infrastructures.

cs.NI↗

arXiv2026

FinInteract: Benchmarking Clarification and Intent Integration in Ambiguous Financial Question Answering

Large language model agents increasingly answer financial questions by searching regulatory filings. Such questions are often deceptively under-specified: Meta Platforms' "operating income" is $46.75B consolidated but $62.87B for the Family of Apps segment, and each reading is exactly verifiable against the filing. A capable agent should recognize the ambiguity and ask, rather than commit to a plausible but unintended reading. Existing financial benchmarks cannot measure this, because one gold answer per question cannot separate agents that resolve the ambiguity from those that guess the common reading, a blind spot we call the single-gold illusion. We release FinInteract, a bilingual (English/Chinese) benchmark of 173 instances that pairs each question with a default and an intended interpretation across a five-category ambiguity taxonomy, and grades whether an agent elicits the right clarification and then integrates it. Re-grading identical outputs against the default rather than the intended reading inflates GPT-4o's accuracy by 3.1 times, confirming the illusion. Beyond it, we find that models answer above 90% once the interpretation is supplied but at most 28.9% when they must elicit it themselves, that targeting is uneven across a taxonomy well powered for entity scope and metric definition and exploratory elsewhere, and that conditioning on the ambiguity category improves resolution at both inference and training time.

cs.AI↗

arXiv2026

InvestorNerd: An Investment and Financial Insights System Based on User Profiles

InvestorNerd is a web-based platform (investornerd.org) designed to educate and democratize financial understanding by providing accessible, AI-powered investment and personal finance insights tailored to potential user profiles. The system addresses a key challenge in helping everyday individuals, especially those without formal financial education, make sense of investment options and personal financial decisions. The platform offers three interactive tools: Stock Insights - Allows users to input any Stock, Mutual Fund, or Exchange Traded Fund (ETF) ticker to receive a summary of relevant news sentiment and quantitative metrics. Stock Insights Questionnaire - Builds on stock insights by letting users specify a preferred risk tolerance and sector interest. It returns categorized investment tables based on volatility and sector, sortable by dividend yield, return percentages, and others. General Insights Questionnaire - Provides users with personalized insights based on answers to an age-income-expenditure-savings questionnaire. Outputs include possible actions regarding savings strategies, account types (e.g., Roth IRA, UTMA), and potential loan options (e.g., FHA). The output is designed to be clear, unbiased, and educational, empowering novice investors with practical insights. This paper details the design, implementation, and evaluation of InvestorNerd, demonstrating how generative AI and open financial data can be integrated to create scalable, insight-rich tools for financial literacy.

cs.CE↗

arXiv2026

Edit-VAR: Taming Visual Autoregressive Model for Precise Video Editing

Text-guided video editing modifies target content while preserving the appearance and temporal coherence of unedited regions. Training-based approaches provide strong control but demand substantial data and computation. Training-free methods fall into inversion-free and inversion-based paradigms. Inversion-free approaches avoid trajectory recovery, but their source-preserving guidance can limit editing strength and leave semantic changes incomplete. Inversion-based approaches recover a latent trajectory before regeneration, where approximation errors can accumulate and cause source-content drift and temporal inconsistency. We introduce Edit-VAR, the first training-free and inversion-free framework for text-guided video editing with a pretrained visual autoregressive video model. Edit-VAR directly encodes the source video into multi-scale discrete tokens and performs probability-guided conditional token replacement for source preservation. Attention-guided token-wise and scale-aware modulation selectively relaxes source constraints over edit-relevant positions and generation stages. Scale-Decoupled Generation, implemented as late-scale constraint release, regenerates motion-consistent details and reduces texture fragmentation. Residual-guided token pruning further exploits redundancy at the final two high-resolution scales to reduce inference cost. Extensive experiments and a blind user study demonstrate that Edit-VAR outperforms existing training-free video editing methods overall in editing fidelity, source preservation, temporal coherence, and inference efficiency.

cs.CV↗

arXiv2026

FINSKILLOPS: A Self-Evolving Multi-Agent System for SEC Filing QA

Financial QA systems are typically improved before deployment through better retrieval, prompting, or agent coordination, leaving their reliability behavior fixed thereafter. In practice, new SEC-filing questions repeatedly expose heterogeneous errors in period, entity, evidence use, and calculation. Existing self-improvement methods can turn failures into new behaviors, but offer limited control over where a correction should apply or which previously correct answers it may break. We therefore frame post-deployment improvement as controlled behavioral maintenance: recurring failures should become scoped skill patches, and each patch should earn deployment with- out introducing regressions. We instantiate this view in FINSKILLOPS, a multi-agent system for SEC filing QA. FINSKILLOPS derives reusable skills from evidence-grounded, typed failure diagnoses and governs them through targeted validation, protected-case regression checks, negative controls, and versioned replacement or retirement. Across six financial QA benchmarks, a single frozen skill registry achieves the highest verdict-weighted correctness and reference consistency among the evaluated systems. Evolved skills raise correctness from 3.70 to 4.55 on our enhanced benchmark. In a separate 12-round operational study, only six of 33 proposed skills are promoted, while the monitoring non-correct rate falls from 20.0% to 12.5%. These results establish controlled skill scope, admission, and lifecycle management as the foundation for reliable self-improvement.

cs.AI↗

arXiv2026

Earth Surface Immune System for Rapid Monitoring of Unknown Anomalies

Earth surface anomalies, driven by escalating climate change, and expanding human activities, are increasing in both frequency and diversity, yet their limited historical data and unpredictability make them fundamentally different from conventional remote sensing targets. Existing methods address specific anomaly categories or stop at localization, leaving a gap between detection and actionable information. Here we present ESIA, an Earth Surface Immune System whose architecture is constrained by three principles from the biological immune system, refined over millions of years against equally diverse and uncertain threats. A non-specific innate immune stage treats anomalies as unobserved changes in time-series satellite imagery, generating binary localization maps at 14.51 km2/s without assuming any anomaly category, surpassing the strongest general baseline by 37% in F1. A specific adaptive immune stage applies negative selection to filter text prompts and matches surviving prompts with localized image patches through a multi-modal foundation model, enabling open-vocabulary recognition of unknown anomaly attributes including category, affected area, and damage severity, with recognition F1 exceeding 80%. A mutation mechanism tunes minimal embeddings at test time, adapting to each scene in 3.26s using a single reference image pair. We validate ESIA on a global-scale dataset covering 19,801.60 km2 across six anomaly categories, comparing against 22 models, and further apply it to quantify degraded farmland in the Dnipro Delta following the Kakhovka Dam collapse and assess burn severity from 2025 Palisades Fire in Los Angeles. This unprecedented flexibility in handling unknown anomalies opens new avenues for real-time disaster response and environmental surveillance.

cs.CV↗

arXiv2026

Dependency-Aware Trajectory Refinement for Efficient Multi-Turn Agent Fine-Tuning

Multi-turn agent trajectories often contain redundant rounds (failed tool calls, parallel sub-queries, verification-only steps) that inflate both training and inference cost. We propose viewing each trajectory as a \emph{round-level dependency DAG} that exposes which rounds are globally load-bearing for the final answer, and fine-tune agents on trajectories refined through this DAG. Given an LLM-annotated DAG, these edits are deterministic and interpretable, with optional rephrasing. Models trained on these refined trajectories consistently outperform those trained on the original trajectories at lower inference cost. Specifically, across four multi-modal QA benchmarks, our refinements improve downstream accuracy by up to $1.7$\,pp over vanilla SFT (and $5.7$\,pp over an LLM-deletion baseline) while reducing per-sample inference messages by up to approximately $40\%$ and inference tokens by up to approximately $48\%$, translating to substantial savings in compute and serving cost. Code is available.

cs.CL↗

arXiv2026

Emergent behaviors of the kinetic Motsch-Tadmor model in a phase-spatially extended setting

The kinetic Motsch-Tadmor (in short, KMT) model is a kinetic flocking model with a normalized communication weight. In this paper, we study the emergent dynamics of the phase-spatially extended KMT model. We first establish a global well-posedness theory in the fully noncompact spatial-velocity setting. To this end, we introduce a direct Lagrangian formulation in which the unbounded part of the initial velocity distribution is separated from an interaction-generated bounded remainder. This decomposition allows us to construct global Lagrangian weak solutions without truncating the velocity distribution and to propagate finite phase-space moments. We then investigate the long-time collective behavior of the resulting solutions. When the initial velocity support is compact while the spatial support is allowed to be noncompact, a time-varying effective-region argument yields a uniform contraction mechanism for the normalized interaction and leads to exponential weak support flocking. When both the spatial and velocity supports are noncompact, support-level flocking is in general impossible. Nevertheless, the same effective-region mechanism, combined with the Lagrangian decomposition and a bootstrap argument for the interaction-generated remainder, yields exponential weak moment flocking. In particular, pairwise spatial moments remain uniformly controlled while velocity fluctuations converge exponentially to zero. These results provide a unified framework for the well-posedness and flocking dynamics of the KMT model beyond the compact-support regime.

math.AP↗

arXiv2026

FrontierChallenge: Evaluating Scientific Workflow Completion

Scientific agents increasingly analyze data, execute code, and produce research artifacts, yet most benchmarks emphasize final answers, isolated programs, or a single domain. We introduce FrontierChallenge, a cross-domain benchmark comprising 300 end-to-end scientific workflows. In this paper, we release and evaluate 97 of these tasks, spanning quantum chemistry, molecular dynamics, materials characterization, analytical chemistry, life science, and electrochemistry/environment. Each task provides fixed inputs and specifies a bundle of required scientific deliverables. We evaluate twelve frontier models with three agent scaffolds. Pass Rate measures the fraction of tasks satisfying the full-completion criterion, while Avg. Score captures partial progress. Each of the best-performing configurations completed only 20 of the 97 released tasks, yielding a Pass Rate of 20.6%. Partial progress translated especially poorly into complete delivery in analytical chemistry and electrochemistry/environment: Avg. Scores reached 87.6 and 94.9, but the highest Pass Rates were only 4% and 0%. Among non-passing Claude Code trajectories, 75.5% still ended with language claiming completion. Complementary HDS6 process scores correlate strongly with task outcomes, supporting FrontierChallenge as a benchmark of Heavy Duty Solver capabilities. These findings show that neither high partial scores nor confident claims of completion reliably indicate that a scientific task has been fully delivered, highlighting the need to evaluate end-to-end workflow execution and the completeness of scientific deliverables together.

cs.AI↗

arXiv2026

"Tab, Tab, Bug": Security Pitfalls of Next Edit Suggestions in AI-Integrated IDEs

Modern AI-integrated IDEs are shifting from passive code completion to proactive Next Edit Suggestions (NES). Unlike traditional autocompletion, NES is designed to construct a richer context from both recent user interactions and the broader codebase to suggest multi-line, cross-line, or even cross-file modifications. This evolution significantly streamlines the programming workflow into a tab-by-tab interaction and enhances developer productivity. Consequently, NES introduces a more complex context retrieval mechanism and sophisticated interaction patterns. However, existing studies focus almost exclusively on the security implications of standalone LLM-based code generation, ignoring the potential attack vectors posed by NES in modern AI-integrated IDEs. The underlying mechanisms of NES remain under-explored, and their security implications are not yet fully understood. In this paper, we conduct the first systematic security study of NES systems. First, we perform an in-depth dissection of the NES mechanisms to understand the newly introduced threat vectors. It is found that NES retrieves a significantly expanded context, including inputs from imperceptible user actions and global codebase retrieval, which increases the attack surfaces. Second, we conduct a comprehensive in-lab study to evaluate the security implications of NES. The evaluation results reveal that NES is susceptible to context poisoning and is sensitive to transactional edits and human-IDE interactions. Third, we perform a large-scale online survey involving over 200 professional developers to assess the perceptions of NES security risks in real-world development workflows. The survey results indicate a general lack of awareness regarding the potential security pitfalls associated with NES, highlighting the need for increased education and improved security countermeasures in AI-integrated IDEs.

cs.CR↗

arXiv2026

A Hyperbolicity Atlas of Large Language Model Hidden States

LLM hidden states are ordinary vectors, but the distances among those vectors may still show hierarchical structure. To our knowledge, this paper is the first systematic study of whether prompt-token hidden states in contemporary LLMs exhibit Gromov Hyperbolicity (GH), a distance-based measure of tree-likeness. Using 818,904 sample-layer measurements from ten open-weight models across MATH500, HumanEval, WinoGrande, and TruthfulQA, we build a GH map over four axes: parameter scale, layer depth, model family, and input domain. The clearest pattern is depth, not scale: middle layers usually form a high-relative-hyperbolicity plateau, while final layers often become substantially more tree-like. Scale effects are weak and non-monotonic, matched 7/8B model families differ strongly, and domains interact with model specialization. These findings make GH useful as a practical diagnostic: it shows where hierarchical distance structure appears, how specialization changes it, and which model-layer-domain comparisons deserve closer analysis.

cs.CL↗

arXiv2026

GenAI Models Capture Urban Science but Oversimplify Complexity

Generative artificial intelligence (GenAI) models are increasingly used for scientific data generation, yet their alignment with empirical knowledge in urban science remains unclear. We therefore ask whether generated urban data reproduce empirical regularities and support repeatable experiments. We introduce AI4US, a framework for evaluating data synthesis and conditional intervention across text and image modalities. Four cases spanning urban systems, within-city structure, neighbourhood vitality and streetscape perception are evaluated against published parameters, observed urban data and human judgements. Generated outputs recovered recognizable scaling and distance-decay patterns and yielded measurable associations between neighbourhood morphology indicators and pedestrian activity, while model judgements showed positive agreement with sampled human choices. Controlled changes to urban conditions produced repeatable output responses across all four cases. However, generated data often compressed empirical numerical coverage, local variation and visual diversity. In an inspectable GenAI model, intermediate activations linearly distinguished some urban relationships, while activation edits changed only selected output scores. AI4US provides an empirically grounded approach for assessing GenAI as the virtual urban laboratory in urban data synthesis and controlled model experiments, while revealing its tendency to simplify empirical complexity.

physics.soc-ph↗

arXiv2026

GridFlow: Structured Latent Flow for Seamless City-Scale 3D Point Cloud Generation

Generating realistic 3D city environments from remote sensing data is important for simulation, urban planning, and mixed reality, yet existing point cloud generation methods are limited to single objects or bounded indoor scenes and cannot handle the scale, seamless tiling, and partial observability challenges of city-scale generation. We present \ours{}, a multi-stage framework that generates dense, colored point clouds ($10^5$ points per $150\text{m}{\times}150\text{m}$ tile) at city scale, conditioned on satellite imagery, semantic segmentation maps, and digital surface models (DSM). A \emph{Grid-Aligned VAE} encodes each tile into a topology-preserving latent grid where tokens correspond to fixed spatial regions, enabling spatially coherent multi-modal conditioning and compact latent-space edge consistency that implicitly aligns thousands of boundary points for seamless cross-tile generation. A conditional rectified flow model synthesizes geometry latents from the fused multi-modal conditions, and an orientation-aware diffusion colorizer separately handles satellite-visible horizontal surfaces and occluded vertical façades. To support standardized evaluation, we build on public 3D data sources to introduce \emph{City3D-MultiGen}, a benchmark of $163$K densely annotated tiles from Melbourne and London with aligned point clouds, satellite images, semantic maps, and elevation data. Experiments show that \ours{} outperforms adapted point cloud generation baselines across all geometry metrics and produces visually coherent colored point clouds with seamless boundaries over arbitrarily large urban extents. Our benchmark details are available at https://huggingface.co/datasets/e32/City3D-MultiGen

cs.CV↗

arXiv2026

Magpie: Real-Time World Renderer for Interactive Games

Modern game development relies heavily on conventional graphics pipelines. High-quality visual content requires modeling, material authoring, animation, lighting, effects, and runtime optimization, making asset production expensive and extending the development cycle of game prototypes. Recently, video foundation models are beginning to change film and video production, but games differ from linear media, they require not only continuous and realistic imagery, but also stable and reproducible gameplay rules, object states, and interaction outcomes. We present Magpie, a real-time generative world-rendering system for interactive games. Magpie separates gameplay execution from visual generation. Designers define scenes and rules in a game engine. At runtime, the Game Engine resolves player actions and maintains world state, while an independent Render Server generates visual output from white-box frames produced by the engine. Magpie provides a system-level implementation path for applying generative models to real-time game rendering. It preserves gameplay designability and reproducibility, and reduces the dependence of early game prototypes on complete visual assets.

cs.CV↗

arXiv2026

MIRAGE: How Conversation State Shapes Historical Evidence Use in Multimodal Personal Agents

Multimodal large language model (MLLM) agents are increasingly used as personal assistants for long-running tasks. Their utility depends on continuity: agents must retrieve and use earlier evidence across dialogue, files, and workspace state. However, agents can generate plausible answers even when access to that history has degraded, causing outcome-only evaluation to overestimate true evidence use. We present MIRAGE (Multimodal Interaction Retrieval, Attribution, and Grounding Evaluation), a controlled study of historical evidence use under conversation-state variation in multimodal personal agents. MIRAGE holds evidence objects, questions, and scoring fixed while varying only conversation state, and evaluates whether an agent can determine answerability, recover the correct source, and answer from it. Across seven frontier and open-weight multimodal backbones, we find that: 1) pre-compaction depth and post-compaction continuation form distinct, non-monotonic failure regimes rather than a single degradation curve; 2) open-weight models rely heavily on context continuity and are reluctant to spontaneously switch to tool-mediated retrieval when provenance fails; and 3) retrieval pressure improves source attribution in deep pre-compaction states for tool-compliant models, but consistently regresses after compaction, where stored evidence has already degraded. These findings show that historical evidence use should be evaluated under state variation, rather than inferred from outcome-only correctness.

cs.CL↗

arXiv2026

Weak flocking for phase-spatially extended kinetic Cucker-Smale equation in a confining force field

We study the quantitative fast and slow weak flocking of the phase-spatially extended kinetic Cucker-Smale model in confining potential fields. The confining potential is allowed to be nonconvex. In contrast to the phase-spatially confined case, the communication weight may have no positive lower bounds, while spatial and velocity diameters may remain infinite. Moreover, an indefinite Hessian of confining potential prevents the direct use of convexity-based coercive estimates. To overcome these difficulties, we identify three structural assumptions that are sufficient for the noncompact hypocoercive method: quadratic confinement, a globally Lipschitz force, and a strict virial inequality. The admissible class of confining potentials includes genuinely nonconvex radially symmetric ones and localized oscillatory perturbations of the harmonic potential. Weak flocking analysis combines three key ingredients: a microscopic mechanical-energy estimate, a time-varying effective region, and a macroscopic Lyapunov functional. For this, we first establish global existence for initial data with finite second moments. In the exponential mechanical-energy class, we further show the uniqueness and finite-time Osgood-type stability in 1-Wasserstein distance. For polynomially decaying initial mechanical-energy tails, we derive the optimal algebraic decay exponent of the fluctuation energy throughout the admissible communication-decay regime. The optimality is verified by a symmetric countably infinite particle solution for a fixed nonconvex potential. For exponentially decaying tails, a two-stage localization argument yields exponential weak flocking on an optimal exponential time scale. These results show that well-posedness and weak flocking persist even for fully noncompact data under genuinely nonconvex confining forces.

math.AP↗