arXiv subjects

Ran Xu

Publications and source records attributed to Ran Xu.

At least 19 recordsLinked to original sources

arXiv2026

Where Do Vision Language Models Fail at 3D Reasoning?

Vision-language models can get individual judgments right while struggling to recover and consistently use the exact properties of a 3D object. To characterize this gap, we introduce BedBench, a controlled benchmark of thirteen procedurally generated tasks spanning individual judgments and composite answers. Across twelve open-weight and four frontier models, per-color counting exposes a pronounced weakness: humans achieve 93.3% accuracy, compared with 7.2% for the best model. Targeted fine-tuning raises one model's accuracy from 3.0% to 68.6%, but reveals a further difficulty: its answers about the most common colors remain inconsistent with its own per-color counts. Supplying those counts as text improves its judgments, showing that recovering a property does not mean using it consistently. Reasoning, scaling, and added visual cues each improve individual judgments, but none makes exact answers reliable. To locate the difficulty, we probe Qwen3.5-27B: exact per-color counts remain recoverable at the language model's input, yet the model's answers rarely recover them (65.0% for the probe vs. 3.0% for the model). Image-removal controls further show that some apparent successes require no visual evidence. These findings motivate evaluating 3D reasoning by exact correctness, consistent use, and image dependence, rather than accuracy alone. Code and dataset are publicly available: https://github.com/leduckhai/BedBench

cs.CV↗

arXiv2026

Ancestor Lifting of Quasi-Cyclic Additive Codes and Certified Quantum Codes

Additive codes over $\Ffour$ that are self-orthogonal under the trace inner product correspond to binary quantum stabilizer codes. We study 1-generator $(1,2)$-quasi-cyclic (QC) additive codes and their exact quantum distances. For composite lengths $n=\ell m$, an ancestor-lifting theorem transports dual words of the folded ancestor to the length-$n$ code. A two-layer decomposition for $\ell=2$ and a CRT splitting for odd $\ell$ yield dimension laws and explain equality, escape, and mixed-branch behavior in the certified examples. A palindromic criterion ensures symplectic self-orthogonality, while reversal pairing and a symplectic Gram rank formula describe broader construction families. Exact distances are established by information-set decoding screening, Brouwer--Zimmermann computation on the additive dual, and independent witness verification. We certify twenty-one record-breaking binary quantum codes across twenty entries of Grassl's code table, with twenty-eight certified members in total. Propagation improves thirty-four further entries, and explicit constructions are supplied for three table-tying GV entries marked Code Missing Construction. Six additional candidates are reported only with rigorous upper bounds and heuristic screening evidence. The results connect divisor-lattice structure in additive quasi-cyclic codes with reproducible certification of exact quantum distances.

cs.IT↗

arXiv2026

SkillQR: Agentic Skill Discovery for Reason-Intensive Query Rewriting

Dense retrievers provide grounded evidence that mitigates hallucinations in outputs from Large Language Models (LLMs). Real-world queries are often verbose, ambiguously framed, or require multi-step reasoning, introducing irrelevant details that weaken semantic matching for dense retrievers. Query rewriting is a widely adopted pre-retrieval technique that leverages LLMs to reformulate queries and bridge the semantic gap between such queries and the retrieval target. Existing rewriting approaches either rely on expensive proprietary models, substantial training data, or iterative retrieval and rewriting, while largely applying a uniform rewriting strategy regardless of the retrieval corpus. We observe that effective query rewriting instructions are not universal: the optimal strategy depends on the retrieval target, rather than the model alone. We introduce SkillQR, a query rewriting framework that automatically discovers reasoning strategy sets from the retrieval target and dynamically routes a query to the appropriate rewriting strategy at inference time. Without relying on training data, SkillQR achieves competitive performance compared to fine-tuned rewriting models on the BRIGHT and R2MED benchmarks, while also providing lower inference latency than iterative rewriting approaches.

cs.IR↗

arXiv2026

MTA-Zero: Self-Improving Multimodal Deep Search via Online QA-Solver Co-training

Multimodal large language models (MLLMs) have become strong visual reasoners, yet remain limited in open-world, multi-step search that requires grounding images, selecting tools, and integrating web evidence. Existing approaches typically construct and filter fixed offline corpora whose question distributions cannot adapt as the target deep-search agent improves. Building on MTA-Agent's image-grounded QA construction, we present MTA-Zero, an online multimodal QA-Solver co-training framework for self-improving multimodal deep-search agents that requires no precomputed multi-hop corpus. Starting from image-grounded VQA seed contexts, a trainable QA Generator retrieves evidence and constructs verified multi-hop questions online. A separate deep-search Solver then selects tools, gathers evidence, and answers the generated questions. To focus generation on the Solver's learning frontier, we derive a Goldilocks reward for the QA Generator from the Solver's group-wise rollout success rate: questions solved by all rollouts or by none receive zero reward, whereas questions near the current frontier are favored. We optimize the two policies independently with GRPO, without sharing parameters or gradients. Across six multimodal deep-search benchmarks, MTA-Zero improves a Qwen3-VL-8B Solver from 36.91% to 51.58% mean accuracy and achieves performance comparable to GPT-5 under the same agent workflow. The same framework also improves a newer Qwen3.6-27B Solver by 10.82 points. These results show that verified online question generation provides a policy-specific curriculum for continually improving multimodal deep-search agents.

cs.CL↗

arXiv2026

Optimal Dividends and Proactive Capital Injection over a Finite Horizon

We study a finite-horizon dividend and capital-injection problem for a Brownian surplus process with time-varying financing costs. Under a classical passive-injection policy, capital is supplied only at the solvency boundary. Anticipated increases in financing costs may instead make it optimal to inject capital while reserves remain strictly positive. We characterize this proactive financing mechanism through an auxiliary Dynkin game for the marginal value of liquidity. The game identifies an endogenous lower financing boundary and an upper dividend boundary, and yields restrictions on intervention timing and comparative statics with respect to the financing-cost path. Under a regular intervention-band condition, we recover the control value by spatial integration of the game value and verify that the optimal policy is two-sided reflection between the two boundaries. We establish this condition for a canonical class with decreasing marginal dividend rewards, increasing marginal financing costs, and a compatible piecewise affine terminal marginal value. Strictly decreasing financing costs eliminate positive-reserve financing contact and reduce the marginal game to a one-player stopping problem. Numerical examples illustrate the optimal policies and the effects of anticipated financing conditions on shareholder value and intervention thresholds.

math.OC↗

arXiv2026

Opera: A Verbal Critic Framework for Long-horizon Coding Agents

Long-horizon coding agents need timely corrections, yet feedback can be ineffective or even harmful when it misjudges ongoing work or fails to address the underlying problem. Existing critics focus on evaluating trajectories and generating feedback, but rarely track what happens after feedback is delivered. We present Opera, a verbal critic framework that treats each correction as a persistent note, followed until the diagnosed problem is resolved. Opera decides when to review through periodic and event-driven triggers, diagnoses issues with typed operators, audits feedback against visible evidence before delivery, and tracks the agent's subsequent actions to distinguish mere compliance from actual resolution. As a test-time critic, Opera improves the resolve rate of non-critic agents by up to 12.4, 15.0, and 8.9 percentage points on Terminal-Bench 2.1, a SWE-Bench Pro subset, and DeepSWE v1.1, respectively, across four policy models, and achieves the highest mean resolve rate among competitive critic baselines on all three benchmarks, and also improves policy models when the policy critiques itself. Beyond inference, Opera-guided rollouts provide approximately on-policy training data: fine-tuning Qwen3.5-9B on them improves its resolve rate on held-out SWE-Bench Pro repositories by 10.2 percentage points without a critic at inference time, matching fine-tuning on rollouts from a stronger model, while preserving its performance when switching harness, i.e., from Openhands to Terminus-2, which the latter substantially degrades. Our code is available at: https://github.com/dongyuanjushi/Opera.

cs.CL↗

arXiv2026

A Counterexample to Two Compactness Conjectures of Brudnyi

We construct a quasi-Banach function lattice $X$ on $\mathbb R$ whose quasi-norm is order continuous and invariant under every translation. Nevertheless, $X$ contains a bounded sequence with common compact support which is equicontinuous in $X$ and has no norm-convergent subsequence. This provides a counterexample to Conjectures 2.5 and 2.11 proposed by Brudnyi in \cite{Brudnyi2015}.

math.FA↗

arXiv2026

CLIFT: Conformal Self-Verification for Web Agent Training and Test-Time Scaling

Open-source web agents are now strong enough to execute realistic browser tasks, but training them with reinforcement learning still depends on weak supervision: binary task success is too sparse for credit assignment, while frontier-language-model judges are too expensive to call at every step and cannot be assumed available at deployment. We introduce CLIFT, a training and test-time scaling method built around conformal self-verification. During training, the agent answers natural-language verification questions about its own rollouts; a Compositional Conformal Certifier keeps only question signals whose URL-conditional evidence agrees with a training-time judge, assigns signed trust weights through polarity-aware lift, and blends the resulting verifier score into per-step rewards in a way that never subtracts from the judge baseline. At test time, the same certified bank is frozen and reused as structured evidence for Conformal Trajectory Selection (CTS): the agent samples a greedy rollout and one or more diverse retries, the self-verifier summarises each URL trace, and a conservative majority-vote rule chooses whether to swap away from the current incumbent without calling any external judge. This single mechanism supports three settings. On WebArena Infinity, CLIFT achieves state-of-the-art performance among open-source web agents. On VisualWebArena, a bank trained with the open model transfers to GPT-5.5 at test time and reaches state-of-the-art performance under the canonical harness. On Online Mind2Web, without training an agent on the benchmark, translating the certified question bank improves a live-web agent in zero-shot evaluation. Together these results position conformal self-verification as a way to turn costly judge feedback into a reusable training signal and a judge-free test-time scaling signal.

cs.CL↗

arXiv2026

VLAA-GUI: Knowing When to Stop, Recover, and Search, A Modular Framework for GUI Automation

Autonomous GUI agents face two fundamental challenges: early stopping, where agents prematurely declare success without verifiable evidence, and repetitive loops, where agents cycle through the same failing actions without recovery. We present VLAA-GUI, a modular GUI agentic framework built around three integrated components that guide the system on when to Stop, Recover, and Search. First, a mandatory Completeness Verifier enforces UI-observable success criteria and verification at every finish step -- with an agent-level verifier that cross-examines completion claims with decision rules, rejecting those lacking direct visual evidence. Second, a mandatory Loop Breaker provides multi-tier filtering: switching interaction mode after repeated failures, forcing strategy changes after persistent screen-state recurrence, and binding reflection signals to strategy shifts. Third, an on-demand Search Agent searches online for unfamiliar workflows by directly querying a capable LLM with search ability, returning results as plain text. We additionally integrate a Coding Agent for code-intensive actions and a Grounding Agent for precise action grounding, both invoked on demand when required. We evaluate VLAA-GUI across five top-tier backbones, including Opus 4.5, 4.6 and Gemini 3.1 Pro, on two benchmarks with Linux and Windows tasks, achieving top performance on both (77.5% on OSWorld and 61.0% on WindowsAgentArena). Notably, three of the five backbones surpass human performance (72.4%) on OSWorld in a single pass. Ablation studies show that all three proposed components consistently improve a strong backbone, while a weaker backbone benefits more from these tools when the step budget is sufficient. Further analysis also shows that the Loop Breaker nearly halves wasted steps for loop-prone models.

cs.CL↗

arXiv2026

Co-Evolving Harnesses and Models: On-Policy Correction Helps Weaker Models Catch Up Where Imitation Fails

Agent harnesses (the system prompt, tool set, execution hooks, and context-management scaffolding around a model) are a critical determinant of agentic task success. Automated harness evolution can enable smaller models to perform well on domain-specific tasks at a fraction of frontier-model cost. Since both the harness and model weights shape behavior, we ask how harness evolution and lightweight fine-tuning should be combined. Across seven enterprise agent tasks, we first evolve a harness with the weaker model, then find that a stronger expert often uses it more effectively, suggesting expert supervision could close the remaining gap. However, training the weaker model on the expert's complete trajectories under the evolved harness backfires: performance regresses on all seven tasks by 4 to 30 points across Qwen3-Coder and Gemma 4, even though the same procedure helps under the unevolved harness. Our analysis shows that imitation transfers knowledge and increases scaffold usage, but disrupts model-harness fit: the weaker model adopts the expert's planning strategy without the competence to execute it and no longer matches the harness evolved around its native planning style. We therefore develop an on-policy expert-correction pipeline, automated by a meta-level MLE agent, that localizes the failing turn in the weaker model's own rollout and asks the expert to rewrite only that turn. This preserves the model's planning style and combines the gains of harness evolution and model adaptation. Our results identify and resolve a source of contention between harness and weight updates, yielding a compatibility-preserving recipe for economical co-evolution on domain-specific enterprise tasks.

cs.AI↗

arXiv2026

RAPTOR: Role-Aware Private Training for Mixture-of-Experts

Differentially private (DP) fine-tuning methods treat sparse Mixture-of-Experts (MoE) models as a single dense block, ignoring that shared layers see all data while experts only see routed records. We identify and formally characterize three resulting failure modes: global clipping suppresses expert gradients, batch-level normalization dilutes sparse expert updates, and fixed privacy noise degrades signal-to-noise ratio on low-load experts. We introduce RAPTOR - a Role-Aware Private Training framework, which alternates shared and expert optimization and targets each failure directly, using expert-specific clipping and noise together with a public expected-owner denominator and a count-independent update schedule that avoids conditioning on private, realized expert counts. We prove the resulting mechanism satisfies $(\varepsilon,δ)$-DP: because each record is assigned to exactly one owner expert, per-expert mechanisms within a layer compose in parallel, so updating all $E$ experts costs no more, in privacy terms, than updating one, with shared and expert streams composing sequentially across training. We further derive a bias-variance decomposition of the public-denominator estimator showing its bias grows predictably with routing imbalance, yielding a privacy-free rule for selecting which layer to protect from routing entropy measured on a small public corpus. Experiments on Switch Transformer and OLMoE fine-tuning across GLUE tasks, and on DeepSeek-VL2-Tiny, show consistent gains over standard DP baselines across several privacy levels ($\varepsilon$), with the largest margins typically at the tightest budgets. Code and models are publicly available: https://github.com/leduckhai/RAPTOR

cs.LG↗

arXiv2026

The Daugavet property in symmetric operator spaces

Under mild assumptions, the main results of this paper characterize a (real or complex) symmetric operator space affiliated with a semi-finite atomless von Neumann algebra $(\mathcal M, τ)$ possessing the Daugavet property as $L_1(\mathcal M,τ)$ or $\mathcal M$ (up to equivalent norms, or even proportional norms). Our results are new even for complex symmetric function spaces, which extend results of \cite{AKM12, KMMW13, AKM15} concerning the Daugavet property for real symmetric function spaces.

math.FA↗

arXiv2026

Norms of multiplication operators: answering Fialkow--Loebl question

Let $\mathcal{M}$ be a factor equipped with a semi-finite faithful normal trace $τ$. Let $E(0,\infty)$ be a symmetrically normed function space and $E(\mathcal{M},τ)$ be the corresponding symmetrically normed operator space. Suppose that $a, b$ are $τ$-measurable operators affiliated with $\mathcal{M}$. It is shown that the range of the multiplication operator $S_{a,b}: x\mapsto axb$ on $\mathcal{M}$ is contained in $E(\mathcal{M}, τ)$ if and only if $μ(a)μ(b)$ belongs to $E(0, \infty)$, where $μ(x)$ stands for the generalized singular value function of a $τ$-measurable operators $x$ affiliated with $\mathcal{M}$. Moreover, we have $$ \|S_{a,b}\|_{\mathcal{M}\to E(\mathcal{M},τ)}= \| μ(a )μ( b) \|_{E (0,\infty) }, $$ which answers a question by Fialkow and Loebl (1984). We also consider the quasi-normed case, and show that the natural quasi-norm of weak $L_p$-space, $0<p<\infty$, is not monotone with respect to the logarithmic submajorisation.

math.FA↗

arXiv2026

Knowledge Graph Augmented Large Language Models for Disease Prediction

Electronic health records (EHRs) enable strong clinical prediction, but explanations are often coarse and hard to use for patient-level decisions. We propose a knowledge graph (KG)-guided chain-of-thought (CoT) framework for visit-level disease prediction on MIMIC-III. We map ICD-9 codes to PrimeKG, mine disease-relevant nodes and paths, and use these paths to scaffold temporally consistent CoT rationales, retaining only samples whose conclusions match observed outcomes. We fine-tune lightweight instruction-tuned LLMs (LLaMA-3.1-Instruct-8B and Gemma-7B) on two small cohorts (400 and 1,000 index visits) across ten PrimeKG-mapped diseases. Our models outperform strong classical baselines, reaching AUROC 0.66-0.70 and macro-AUPR 0.40-0.47. Without additional training, the models transfer zero-shot to the CRADLE cohort, improving accuracy from 0.40-0.51 to 0.72-0.77. In a blinded clinician study, KG-guided CoT rationales are consistently preferred for clarity, relevance, and correctness. Code is available at: https://github.com/JonathanWry/KG-guided-LLM-pipeline

cs.AI↗

arXiv2026

DarwinX: Evolving Agent Harnesses Through Natural Selection

An LLM agent's capability depends not only on model weights but on its harness: prompts, tools, skills, and control flow. Self-improvement loops already edit harnesses, yet single-lineage search is path-dependent and local wins often regress other tasks. We introduce DarwinX, which treats self-evolution as selection over a population of harnesses with the model frozen: a preserve-and-extend contract admits only variants that extend coverage without regressing, an archive keeps alternative lineages for recombination, and failure-, teacher-, and self-derived evidence share one edit interface. Fitness comes from each benchmark's own verifier: no gold solutions, no hand-picked winners. Across four benchmarks that progressively separate the evolution signal from the test, one loop adds about 17 points on average: Terminal-Bench 2.1 rises +7.7 to 83.2% on a matched base and to the verified frontier at 84.7% on a stronger one; TerminalWorld's held-out split reaches 68.3%, ahead of every off-the-shelf agent; WebArena-Infinity real-task pass@1 rises from 43.5% to 93.0% audit-clean; and a Terminal-Bench 2.1 harness transfers unchanged to SWE-bench Verified. What evolves is general agent competence, not benchmark-specific patches, so it survives changes of task, verifier, and base model. A frozen model need not be a fixed agent: harness selection turns evaluation compute into durable capability.

cs.NE↗

arXiv2026

Disjointness-preserving mappings on Calkin operator spaces and positive isometries

Let $E(\mathcal{M},τ)$ and $F(\mathcal{M},τ)$ be two Calkin operator spaces affiliated with a semifinite von Neumann algebra $\mathcal{M}$ equipped with a semifinite faithful normal trace $τ$. We show that if $\mathcal{M}$ is atomless, $τ$ is finite, and $E(v,τ)\not\subseteq F(\mathcal{M},τ)$, then every order-measure continuous and disjointness-preserving mapping $T:E(\mathcal{M},τ)\xrightarrow{\rm into} F(\mathcal{M},τ)$ is identical to the zero mapping, which establishes a noncommutative version of Abramovich's theorem. We also show that every positive isometry $T$ from a normed $\mathcal{M}$-bimodule $E(\mathcal{M},τ)$ of $τ$-measurable operators into another $F(\mathcal{M},τ)$ preserves disjointness provided that the norm of $F(\mathcal{M},τ)$ is strictly monotone. As an application, we obtain the general form of $T$, which extends and unifies several results due to Abramovich, de Jager, Conradie, Veksler and Sukochev et al. \cite{SV,HSZ20,Abra1991,vek,dC20}.

math.FA↗

arXiv2026

3D-Aware VLMs with Implicit and Explicit Geometries

Despite rapid progress, most existing vision-language models (VLMs) built from 2D visual inputs often struggle when handling various 3D tasks that require fine-grained spatial understanding and reasoning. To bridge this gap, we present VLM-IE3D, a unified framework that enhances the 3D spatial awareness of VLMs by equipping them with both implicit and explicit 3D geometries learned from RGB videos. Our VLM-IE3D introduces Implicit Geometry Tokens (IGTs) that capture high-level geometric priors from input videos, as well as complementary Explicit Geometry Tokens (EGTs) that encode detailed geometric structures from reconstructed 3D attributes. On top of that, VLM-IE3D comes with a 3D-aware adapter that effectively fuses the two types of geometric representations with 2D visual cues. This RGB-only design injects strong 3D inductive biases for fine-grained spatial understanding and reasoning without requiring any additional 3D inputs. Extensive experiments show that VLM-IE3D achieves superior performance consistently across various 3D tasks including 3D video detection, 3D visual grounding, 3D dense captioning, and spatial reasoning. Code and models are available at https://github.com/Vegetebird/VLM-IE3D.

cs.CV↗

arXiv2026

Look Less, Think Faster: Joint Token-Compute Adaptation for Multimodal LLMs

Multimodal Large Language Models (MLLMs) have recently demonstrated strong performance across vision-language tasks. However, their high inference cost, arising from both the large number of input visual tokens and the heavy computation of the large language model (LLM), remains a key barrier to practical deployment. Recent work attempts to reduce the cost by adaptively optimizing individual dimensions, e.g., pruning redundant visual tokens or skipping LLM layers and heads. Nonetheless, prior approaches typically treat these dimensions independently and overlook a fundamental coupling: the available compute resources must be dynamically allocated across all dimensions based on the input content. To bridge the gap, we propose SmartVL, a unified adaptive inference framework that jointly controls vision token number and model compute capability in response to varying input contents and compute budgets. SmartVL introduces a vision-side token controller that dynamically selects informative visual tokens and an LLM-side compute controller that adaptively adjusts LLM computation. Importantly, these controllers are trained to coordinate with each other so that the overall inference cost satisfies a target budget. To allow this joint scheduling, we connect the controllers using a shared budget encoding and leverage a differentiable latency estimator for end-to-end training. This design enables SmartVL to learn cross-stage allocation strategies that adapt to both input complexity and runtime compute constraints. Experiments across multiple MLLM benchmarks demonstrate that, with joint scheduling, SmartVL consistently outperforms prior adaptive methods and achieves superior accuracy-efficiency Pareto frontiers. Project page: https://www.schaterji.io/publications/2026/jointtokencompute.

cs.CV↗