arXiv subjects

Xu Zhang

Publications and source records attributed to Xu Zhang.

At least 19 recordsLinked to original sources

arXiv2026

Jailbreak Connectivity: Towards Diverse, Transferable, and Universal Multimodal Large Language Model Jailbreak

Multimodal large language models (MLLMs) are vulnerable to visual jailbreak attacks that bypass safety safeguards. Existing methods typically optimize isolated adversarial examples, limiting attack diversity, transferability, and generalization. We propose Jailbreak Connectivity (JC), a framework based on the observation that effective adversarial inputs can be connected through low-loss regions in image space. JC optimizes continuous adversarial paths: path sampling yields diverse jailbreak images, lightweight surrogate classifiers guide cross-model transfer, and a distribution-level objective enables universal attacks across harmful queries. On MM-SafetyBench, JC achieves average attack success rates of 84.01%, 85.77%, and 97.95% on Qwen3-VL-8B-Instruct, InternVL3.5-8B, and Gemma-3-12B-IT, respectively, exceeding the strongest baselines by up to 68.36 percentage points. In black-box evaluations, JC improves attack success rates over FigStep by 24.69 and 27.34 percentage points on GPT-5.6 and Gemini 3.1 Pro, respectively. These findings reveal structured vulnerabilities in MLLM safety alignment and inform more comprehensive security evaluation.

cs.LG↗

arXiv2026

Observability of wave equations with time- and space-dependent potentials: a counterexample under GCC and a positive criterion on the disk

We show that the geometric control condition (GCC) does not guarantee internal observability for a wave equation with a smooth time-dependent potential. On a bounded connected domain with analytic boundary, we construct a nonempty open observation set such that the pair satisfies GCC and the free Dirichlet wave is observable. For this same domain and observation set, we construct a smooth $2π$-periodic potential for which there exists a nonzero smooth Dirichlet wave that vanishes on the observed cylinder for all time. The wave is the unique mild solution in the space $X_Ω=L^2(Ω)\times H^{-1}(Ω)$ on every finite time interval, so no finite observation horizon yields an observability inequality for this fixed potential. The construction starts from a real-valued local counterexample to unique continuation across a noncharacteristic hyperplane. This local statement is of independent interest, since earlier constructions for the wave operator use complex-valued waves and potentials. The global counterexample is then obtained by joining flat oscillatory solutions along the edges of a polygonal observation set, repairing their sum at the vertices without changing its zero set, and using a spatial transition to a common time-independent pair near the boundary. The planar construction also yields counterexamples in every space dimension $n\ge3$. For comparison, and to identify a geometry that supplies both propagation and continuation, we show on the unit disk that an observation region containing a central disk and a neighborhood of an embedded cross-cut yields observability for every weak wave at sufficiently large observation time, with constants uniform over bounded measurable potentials. The proof uses a cut, logarithmic polar coordinates, and a Carleman weight whose favorable flux is confined to the observed artificial boundary.

math.OC↗

arXiv2026

Recall or Relearn? A Mixture of In-Context Experts for Recurring Concept Drift

Tabular foundation models (TFMs) are being applied to Web data streams such as network intrusion detection, where they adapt by updating the labelled rows of their context without retraining. When a past concept returns after leaving the context, a TFM faces a recall-or-relearn decision between restoring stored rows of the concept and relearning it from new labels. We propose MICE (Mixture of In-Context Experts), a training-free learner that cuts the stream into segments at detected changes. It merges a closed segment into a stored context when the TFM's estimated additional error on the segment is within a tolerance, under a capacity limit on stored contexts. A fast level weights the stored contexts and first-in-first-out (FIFO) windows by the log-loss of their predictions on the last labelled batch. We show that, with pure stored contexts and a margin condition, this fast level gains over relearning, in expected accuracy, the normalised positive area between the stored context's accuracy and the TFM's learning curve, up to a first-batch term. A slow level mixes the fast level's prediction with that of the FIFO window of the last labelled rows and bounds the discounted half-Brier regret of MICE against this window at every time. Learning curves measured beforehand predict the ordering of the observed recall gains. MICE outperforms reset, which empties the context at detected drifts, on grid streams with random boundary phases. Merging keeps fewer experts than pools that store every segment, and with fixed segmentation a time-matched snapshot pool matches MICE on the shifted grid. On real streams, MICE has the highest mean accuracy of all policies we compare.

cs.LG↗

arXiv2026

Efficient Remote Sensing Image-Text Retrieval via Interaction-Free Recall and Text-Guided Reranking

Remote sensing (RS) image-text retrieval plays a critical role in understanding massive RS imagery. However, the dense multi-object distribution and complex backgrounds in RS imagery make it difficult to simultaneously achieve fine-grained cross-modal alignment and efficient retrieval. Existing methods either rely on complex cross-modal interactions that lead to low retrieval efficiency, or depend on large-scale vision-language adaptation, which requires massive data and computational resources. To address these issues, we develop a fast-then-fine (FTF) retrieval framework centered on interaction-free multi-granular recall and parameter-free text-guided cross-modal reranking with the Balanced Text-guided Interaction Block (BTIB). Specifically, interaction-free multi-granular recall pre-computes gallery embeddings and matches them with the corresponding query embedding without full-gallery token-level text-image interaction, while text-guided cross-modal reranking with BTIB refines local semantic correspondence within the top-K candidate set. Furthermore, an inter- and intra-modal loss is designed to jointly optimize cross-modal alignment across multi-granular representations. Extensive experiments on public benchmarks demonstrate that the FTF achieves competitive retrieval accuracy while significantly improving retrieval efficiency compared with existing methods.

cs.CV↗

arXiv2026

EvoRouter: Cost-Efficient Multi-Agent Routing via Insight-Guided On-Policy Distillation

Serving web queries at scale with multi-agent systems requires jointly configuring large language models, collaboration modes, agent counts, interaction rounds, and skills. However, the resulting configuration space grows combinatorially, while sparse execution rewards provide limited fine-grained supervision for interdependent planning decisions. We propose EvoRouter, a self-evolving multi-agent routing framework driven by insight-guided on-policy distillation. A dedicated expert periodically synthesizes performance-cost feedback from configuration trajectories across tasks into reusable conditional routing insights about capability requirements, returns from resource scaling, and interactions among configuration choices. A privileged teacher combines these evolving insights with current execution feedback to provide dense token-level supervision for the router's own planning trajectories through on-policy distillation. Cost-aware reinforcement learning jointly optimizes task performance and computational cost. Together, these training signals internalize execution insights into the routing policy. The updated policy generates new trajectories that support subsequent insight extraction, forming an execution-driven self-improvement loop. We evaluate EvoRouter on 19 in-domain datasets spanning five domains and two out-of-domain benchmarks. EvoRouter achieves an average in-domain accuracy of 68.69%, outperforming the strongest routing baseline by 7.23 percentage points while reducing inference compute cost by 50.95%. Our code is open-sourced at https://github.com/CenYangtze/EvoRouter.

cs.AI↗

arXiv2026

The ALMA-QUARKS survey: A Keplerian Circumbinary Disk Candidate in the High-mass Star-forming Region IRAS 08303-4303

We present the ALMA Band 6 continuum and molecular line observations of IRAS 08303-4303, a high-mass star-forming region, with an angular resolution of $\sim 0.3''$ ($\sim 700$ au at the distance of 2.35 kpc). Six dense condensations are identified in the 1.3 mm continuum, and at least four of them are associated with molecular outflows. The estimated outflow parameters, particularly mass outflow rates $\geq 10^{-4}$ $M_{\odot}$ yr$^{-1}$, suggest that the driving sources are high-mass young stellar objects. Two of the condensations are forming a wide protobinary system with a separation of $\sim 900$ au. They reside within a single rotating structure, identified as a circumbinary disk which has a mass of about 2 $M_{\odot}$. This structure is traced by multiple molecular lines and exhibits a Keplerian rotation pattern. Based on the Keplerian rotation assumption, the enclosed mass is derived to be between 32 and 61 $M_{\odot}$: 22-43 $M_{\odot}$ for the primary and 10-18 $M_{\odot}$ for the secondary. The circumbinary disk is on the whole gravitationally stable, yet some parts with Toomre $Q < 2$ indicate potential fragmentation. Various molecular lines reveal a streamer-like structure connecting to the circumbinary disk, suggesting transfer of material from the envelope to the protobinary system through it. Despite the observational rarity of such circumbinary disks around high-mass YSOs, the high binary fraction among massive stars underscores their significance. This source, therefore, serves as a key laboratory for understanding the formation of high-mass binary systems.

astro-ph.SR↗

arXiv2026

On the sum of least prime factors in short intervals

Let $p(n)$ denote the least prime factor of $n$ and $L_C(x)=Cx^{1/2}(\log x)^2$ with $C>0$. The sum of $p(n)/n$ over composite $n$ lying in the short interval $[x,\,x+L_C(x)]$, a question raised by Erdős and Graham, is studied. (i) The constant $c=8$ in the mean asymptotic is estimated \[ S(x)=\sum_{n 0$, the window sums $μ_C(x):=\sum_{x\le n\le x+L_C(x)}p(n)/n$ over composites have mean $4C$: $\frac1X\sum_{x\le X}μ_C(x)=4C+O_C(1/\log X)$, and second moment $\frac1X\sum_{x\le X}(μ_C(x)-4C)^2=O_C(1)$. (iii)Under the Maier-consistent bounded-fluctuation hypothesis, $μ_C(x)=4C+O_C(1)$ holds uniformly in $x$, and the Erdős--Graham question has an affirmative answer. Unconditionally, the uniform statement remains open; proving it would require resolving short-interval prime estimates at scale $(\log y)^2$.

math.NT↗

arXiv2026

REATS: LLM Reasoning-based Ensemble Learning for Adaptive Time Series Forecasting

Due to the diversity of real-world time series, no single forecasting model consistently dominates across all samples. Ensemble learning addresses this by combining complementary model strengths, yet existing methods rely on fixed rules or black-box models based solely on numerical inputs, failing to leverage LLM reasoning for interpretable weighting decisions. We propose REATS, which leverages LLM reasoning capabilities as an intelligent ensemble router that jointly processes textual temporal pattern descriptions and numerical features to produce interpretable, sample-adaptive ensemble weights through chain-of-thought reasoning. To enable effective LLM-based ensembling, we study its key design choices and propose: (i) a structured input pipeline that transforms raw time series into hybrid textual--numerical representations with fixed token cost, enabling rule-based chain-of-thought construction without API dependency, augmented with retrieved similar-sample priors; (ii) a diverse multi-row weight supervision scheme coupled with a token-efficient percentage-table format that reduces numerical complexity and mitigates LLM hallucinations; and (iii) a two-stage fine-tuning framework combining SFT with GRPO, where a reciprocal reward mapping transforms the continuous unbounded MSE gap into bounded signals with amplified near-oracle sensitivity, addressing the uniform sensitivity and outlier-dominated advantage compression inherent in naive reward designs for regression-based GRPO. Experiments on eight benchmarks demonstrate that REATS outperforms competitive ensemble baselines while providing natural language explanations and demonstrating strong transfer learning and out-of-domain generalization to unseen candidate models.

cs.LG↗

arXiv2026

A Unified Interface Neural Network for Elliptic Interface Problems

We develop a Unified Interface Neural Network (UINN) framework for elliptic interface problems with discontinuous coefficients. A single shared network provides a unified representation of the solution in both subdomains, while a projection--correction construction embeds the solution and flux jump conditions directly into the neural trial space. This design aligns with the classical numerical principle of incorporating interfacial physics into the approximation basis to construct interface-unfitted numerical methods. The jump conditions are therefore satisfied exactly for any network parameters. Compared with methods based on separate phase networks and numerical formulations with interface penalty terms, UINN provides a unified representation with exact enforcement of interfacial jump conditions and interface-penalty-free optimization. Those features streamline the numerical formulations and can enhance training robustness. Numerical experiments on smooth interfaces at modestly large and high coefficient contrasts demonstrate UINN's solution accuracy, near-machine-precision satisfaction of the interface jump conditions, and enhanced robustness.

physics.comp-ph↗

arXiv2026

ANCHOR: Audio-Visually Grounded Chain-of-Thought Reasoning Benchmark

Multimodal large language models (MLLMs) perform well on audio-visual question answering, but correct answers do not necessarily imply that their stated reasoning is grounded in relevant perceptual evidence. Existing benchmarks mainly evaluate final-answer accuracy, offering limited insight into whether model-generated rationales correctly identify visual regions, temporal intervals, and audio events supporting predictions. We introduce ANCHOR, a benchmark and evaluation framework for measuring grounded-rationale alignment in multimodal reasoning, comprising 2,319 videos and 8,272 question-answer pairs with structured reasoning traces, temporal evidence, spatial bounding boxes, and salient audio cues, including a human-refined test set of 224 videos and 975 question-answer pairs. We separately measure answer correctness, hallucination, logical divergence, and conciseness, with direct spatial and temporal grounding measurements on a focused subset. Across the full benchmark, twelve contemporary MLLMs show a substantial alignment gap: models with comparable answer performance can differ markedly in whether their explanations are supported by annotated evidence. On a matched subset of 100 videos and 446 question-answer pairs, rankings remain highly stable across human-refined, Gemini-generated, and GPT-generated references (mean Pearson r=0.99), providing evidence against model-reference circularity within this subset. Grounding-cue ablations of three MLLMs on the full human-refined test set reveal a consistent modality hierarchy: removing video reduces accuracy by 38.4-41.8 points, while removing bounding boxes (3.2-7.7 points) and audio (4.2-5.6 points) causes smaller, complementary, model-dependent drops. These findings show answer-only evaluation obscures important grounding failures and position ANCHOR as a testbed for evaluating accurate and evidence-aligned audio-visual reasoning.

cs.CV↗

arXiv2026

OpenGCUA: Toward a Fully Open Stack for Generalist Computer Use

Computer use has reached a turning point: the focus is expanding from GUI-only interaction to systems that complete tasks on users' computers through visual and programmatic interfaces. We define a generalist computer-use agent (GCUA) as a system in which the model chooses how to observe, act, and organize its work across visual and programmatic interfaces. Yet frontier implementations, such as GPT-6 Astra's computer-use system, remain largely closed, limiting systematic study of their design. We introduce Open Generalist CUA (OpenGCUA), an ongoing project to develop and open the full stack for GCUAs with the community. This report presents the first release of OpenGCUA. Code execution is a key feature of our agent design: it lets intermediate results determine subsequent operations and enables models to apply their existing programming capabilities directly to computer interaction. We also propose harness-to-model decision transfer, moving selected observations and task-state decisions from fixed harness rules to the model. On OSWorld-v2, our full configuration improves task scores across all eight evaluated models, including GPT-6 Astra and Claude Opus 5, raising mean partial score by +19.6 points and reducing reported steps per task by an average of 53.0% relative to batched GUI-only execution. We release our implementation and prompts, evaluation trajectories from all eight models, and preliminary versions of a real-device harness and a cross-device SDK. The harness addresses real-device challenges not fully captured by benchmarks, while the SDK provides a common programming interface across computers, mobiles and browsers. Across four real macOS tasks under our real-device harness, selected Qwen3.8-Max runs use 19.4 times as many reported reasoning tokens as GPT-6 Astra, highlighting substantial differences between models in real-device execution efficiency.

cs.CV↗

arXiv2026

From Documents to Spans: Span-Centric Learning for LLM-based Explainable Medical Coding

Medical coding, which assigns standardized ICD codes to clinical documents, is a fundamental yet challenging task in modern healthcare. It requires extensive knowledge of a large code space and the ability to identify and interpret sparse supporting evidence in long, noisy clinical documents. Most existing methods rely on document-level supervision, which provides limited guidance on supporting evidence and insufficient training signals for rare codes. To address these limitations, we propose Span-Centric Learning (SCL), which separates coding knowledge acquisition from its application in full documents. SCL first learns coding knowledge from short spans, then learns to use this knowledge in full documents from a small set of annotated data. Experiments show that, our approach enables an 8B LLM to surpass state-of-the-art methods using only 248 complete documents for training. It also provides explicit supporting evidence for each predicted code, with strong alignment to human annotations. Further analysis shows that the model can learn knowledge of codes absent from documents through spans and apply that knowledge to full documents.

cs.CL↗

arXiv2026

Look Before You Restore: Learning to Predict Before Acting for Agentic Image Restoration

Agents for mixed-degradation image restoration typically learn what a tool does to an image only after running it: failed calls waste computation and harmful outputs go unnoticed until too late. We propose PreActIR, an agentic restoration framework that predicts tool consequences before execution. For each candidate tool, a tool-conditioned world model forecasts changes in residual degradation, image quality, and fidelity, together with the risk of harm and the uncertainty of the prediction. The agent executes the top-ranked candidate, and an independent verifier commits or rolls back its output before the agent replans from the observed state. PreActIR also audits whether performance is limited by the selector or by the candidate space, and studies targeted tool adaptation as a controlled way to expand that space. In a controlled evaluation that fixes the candidate pool, verifier, and budget and employs reference-annotated degradation states, PreActIR uses 42.5% fewer real tool calls and 63.4% fewer rejected calls than reactive, execute-then-evaluate trial and error, while improving Y-PSNR by 0.92 dB.

cs.CV↗

arXiv2026

MAI-Gym: An Open All-in-One Platform for GUI Agent Evaluation and RL Training across Mobile and Computer Environments

Online reinforcement learning (RL) with verifiable rewards has driven rapid progress in mathematics, coding, and tool use, but scaling this success to GUI agents remains challenging due to the high cost of interactive environments and the scarcity of realistic training tasks. We introduce MAI-Gym, an open all-in-one platform that unifies GUI agent evaluation, data collection, and online RL training across mobile, computer, and web environments. Through a shared rollout pipeline, MAI-Gym ensures consistent agent behavior across evaluation and training, supports ten established benchmarks, and scales to 20,000 parallel environment instances. Beyond scalable infrastructure, MAI-Gym enables automated synthesis of diverse GUI applications and challenging tasks, including resettable mocked apps and coherent cross-app worlds with execution-validated task verifiers. Using MAI-Gym, we train MAI-Gym-9B, a unified 9B agent for mobile and computer use, achieving state-of-the-art performance among open GUI agents under 10B parameters: 78.4% on MobileWorld and 76.2% on OSWorld-Verified. Importantly, RL in a mocked train-ticketing app transfers to its real-world counterpart, improving task success on a phys- ical phone from 28.8% to 48.1%. We release the framework, synthesis skills, mocked apps, 2,000 verified tasks, and the model checkpoint to support repro- ducible research and further advances in scalable GUI agent training.

cs.AI↗

arXiv2026

AIPOCH Open-Science: An Open-Source AI Research Workbench with Reproducible .science Packages

AI-assisted research leaves instructions, computations, artifact revisions and checks whose relationships matter when another researcher receives the work. We present AIPOCH Open-Science, an open-source research workbench with portable .science session packages, and specify an exchange contract for these recorded relationships. Version and assessment targets remain attributable while receiving-workspace identifiers can be remapped, eligible contents can be omitted explicitly, and imported history remains separate from a writable fork. A controlled session tested an incorrect first version, a reviewed correction, an unreviewed revision and a planned but unexecuted step. We compared complete and partial exports with source records extracted independently of the exporter, then inspected imports into separate projects within the same installation. Both conditions retained the tested version and review associations, kept the unreviewed revision distinct from earlier assessments and represented excluded content as unavailable. A documentation-rich ordinary directory also supplied evidence for all seven static inspection questions. A separate registry case connected a table to its recorded task, execution and review; offline execution of its retained script and fixed input reproduced two saved CSV tables byte for byte. The study contributes an implemented exchange contract and a controlled examination of its preservation rules, content-selection constraints and interpretation boundaries.

cs.SE↗

arXiv2026

Extension of Sobolev functions on balls in infinite dimensions

We prove the existence of a bounded Sobolev extension operator $E:W^{p,1}\left( B,P \right) \rightarrow W^{p,1}\left( \ell^{2} ,P \right)$ using a completely new method, where $B\subset \ell^{2}$ is the unit ball and $P$ is any non-trivial centered Gaussian measure on $\ell^{2}$. This solves an open problem posed in the literatures.

math.FA↗

arXiv2026

Evolving Creatives with Clicks: CTR-Driven Advertising image Editing via Attribute Disentanglement and Preference Alignment

The optimization objective for e-commerce advertising creatives has shifted from aesthetic quality to online business metrics like click-through rate (CTR). Recent methods generate appealing ad images from product images by constructing CTR-based reward signals and refining prompt generation via Reinforcement Learning. Such paradigms equally penalize all attributes within an underperforming image, even though some are actually beneficial to clicks. In this work, we propose Evolving Creatives with Clicks (ECC), which reformulates creative optimization as fine-grained editing rather than generation from scratch. ECC selectively refines attributes that hurt CTR while preserving those that contribute positively. Specifically, faced with a rich attribute space across massive images, ECC first identifies 28 attributes significantly correlated with CTR via ANOVA and Random Forest regression. Based on attribute contrasts between images with disparate CTRs, we instruction-tune a prompt model to generate editing suggestions, then refine it with real click preferences since reference-derived suggestions are not uniformly beneficial. Furthermore, we introduce Consistent Preference Optimization (CPO) as ECC's preference optimization strategy, which enforces consistency across semantically equivalent shuffled suggestions to mitigate DPO overfitting and improve preference generalization. Extensive experiments demonstrate that ECC significantly improves the CTR of ad images, outperforming state-of-the-art models by a substantial margin. The complete data pipeline, including attribute extraction and analysis, the resulting dataset, and source code will be released upon acceptance.

cs.CV↗

arXiv2026

A Palais-Smale type condition for optimal control problems

We consider an optimal control problem with a general control constraint and a fixed initial state, for which convexity assumptions are absent. To establish the existence of optimal pairs, we introduce a control version of the Palais-Smale type condition adapted from critical point theory and prove that optimal pairs exist under this condition (see Theorem 2.1). Subsequently, we verify that a specific class of optimal control problems satisfies our Palais-Smale type condition (see Proposition 2.1). Furthermore, as an application, we present an example (Example 3.1) and show that, to the best of our knowledge, no existing result can be applied to establish the existence of an optimal pair in this case (see Section 3.1.1). Example 3.2 shows that finitely many Pontryagin maximum principle candidates do not guarantee the existence of an optimal control. Finally, Example 3.3 demonstrates that the proposed Palais-Smale type condition is sufficient, but not necessary, for the existence of optimal controls.

math.OC↗