arXiv subjects

Zihao Wang

Publications and source records attributed to Zihao Wang.

At least 55 records · Page 3Linked to original sources

arXiv2026

LifeAgentBench: Benchmarking LLMs for Long-Horizon, Cross-Dimensional Lifestyle Health Reasoning

Personalized lifestyle health analysis requires long-horizon, multi-dimensional reasoning over heterogeneous lifestyle signals, and recent advances in mobile sensing and large language models (LLMs) make such support increasingly feasible. However, the capabilities of current LLMs in this setting remain insufficiently understood due to the lack of systematic benchmarks. In this paper, we introduce LifeAgentBench, a large-scale QA benchmark for long-horizon, cross-dimensional, and multi-user lifestyle health reasoning, containing 22,573 questions spanning from basic retrieval to complex reasoning. We release an extensible benchmark construction pipeline and a standardized evaluation protocol, deriving verifiable answers through executable queries and programs to support reliable assessment. We then systematically evaluate 13 representative LLMs on LifeAgentBench and identify key bottlenecks in long-horizon aggregation and cross-dimensional reasoning. Motivated by these findings, we propose LifeAgent, a tool-augmented reasoning baseline that decomposes complex queries, performs multi-step evidence retrieval, and invokes tools for deterministic aggregation. LifeAgent substantially enhances LLMs' capabilities on challenging reasoning tasks, achieving clear improvements over widely used baselines and showing potential for health reasoning in everyday scenarios. The benchmark is publicly available.

cs.AI↗

arXiv2026

Callability Is Not Operability: Controlled Interface Interventions for LLM Agents

A tool call can be perfectly valid yet still leave an autonomous agent unable to determine what to do next. For example, if an external effect commits but its response is lost, committed and uncommitted states may become indistinguishable to the agent even though they require different continuation actions. We study this gap between callability and operability: whether a tool interface exposes the action-relevant state and semantics needed for an agent to continue safely under operational uncertainty. We operationalize tool operability through Agent-First Tooling (AFT), a set of interface mechanisms spanning selective capability discovery, execution lifecycle and recovery, explicit external-effect semantics, machine-readable results, and postcondition verification. We introduce AFT-Bench, a controlled interface-intervention framework that holds the task, backend, initial state, injected failure, agent, and language model fixed while varying the interface exposed to the agent.

cs.SE↗

arXiv2026

PercepCap: Video Captioner with Structured Spatio-Temporal Perception

Video captioning requires fine-grained spatio-temporal understanding of videos, including spatial perception of where objects are located and temporal perception of when events occur. Existing MLLMs usually generate captions directly from video inputs without exposing the perceptual evidence behind descriptions. As a result, mistakes in spatiotemporal perception are only observed in the final caption, making it difficult to identify the underlying perceptual errors directly. To address these issues, we present PercepCap, a perception-aware video captioning framework that makes perceptual evidence explicit before producing the final caption. Specifically, PercepCap follows a perceive-describe generation chain, where the model first produces a spatiotemporal perception trace comprising object trajectories and temporal events, and then generates the final caption conditioned on the perceived evidence. To support this, we design a two-stage training strategy. Perceive-then-Describe Supervised Fine-tuning adapts the model from caption-only generation to the proposed perceive-describe chain, while Perception-Grounded Reinforcement Learning optimizes perception trace and caption quality with joint rewards over perception chain and the final caption. To support our two-stage training, we introduce Caption-Anchored Perception Data Construction. This pipeline builds the SFT and RL training data by first generating a caption-only description, extracting the objects and events it mentions, and grounding them back in the video with boxes and timestamps. This yields caption-aligned perception data that provides solid training ground truth, ensuring that the explicit perception trace and final caption refer to the same objects and events. Across direct caption and caption-to-QA evaluation, PercepCap consistently improves upon the Qwen3-VL baseline and demonstrates leading caption quality.

cs.CV↗

arXiv2026

Neural-Symbolic Reasoning over Knowledge Graphs: A Survey from a Query Perspective

Knowledge graph reasoning is pivotal in various domains such as data mining, artificial intelligence, the Web, and social sciences. These knowledge graphs function as comprehensive repositories of human knowledge, facilitating the inference of new information. Traditional symbolic reasoning, despite its strengths, struggles with the challenges posed by incomplete and noisy data within these graphs. In contrast, the rise of Neural Symbolic AI marks a significant advancement, merging the robustness of deep learning with the precision of symbolic reasoning. This integration aims to develop AI systems that are not only highly interpretable and explainable but also versatile, effectively bridging the gap between symbolic and neural methodologies. Additionally, the advent of large language models (LLMs) has opened new frontiers in knowledge graph reasoning, enabling the extraction and synthesis of knowledge in unprecedented ways. This survey offers a thorough review of knowledge graph reasoning, focusing on various query types and the classification of neural symbolic reasoning. Furthermore, it explores the innovative integration of knowledge graph reasoning with large language models, highlighting the potential for groundbreaking advancements. This comprehensive overview is designed to support researchers and practitioners across multiple fields, including data mining, AI, the Web, and social sciences, by providing a detailed understanding of the current landscape and future directions in knowledge graph reasoning.

cs.AI↗

arXiv2026

The THESAN-ZOOM project: clumpiness of high-redshift galaxies and its connection to bursty star formation

Recent JWST observations have revealed diverse high-redshift galaxy morphologies, including a population with irregular and clumpy structures. The physical origin of these structures, and the extent to which observational biases shape their appearance, remain uncertain. We present a power-spectrum-based method for quantifying galaxy clumpiness across spatial scales, using the radiation-hydrodynamic simulation suite THESAN-ZOOM, which employs a state-of-the-art galaxy formation model that resolves the multiphase interstellar medium (ISM). Although the total stellar mass distributions in THESAN-ZOOM galaxies are usually smooth, clumpy structures appear in the H$α$, far-ultraviolet (FUV), and optical light distributions. Tracers sensitive to shorter-timescale star formation exhibit more pronounced small-scale structure ($\sim10^{2}$--$10^{3}{\rm pc}$). The corresponding projected light spectra follow $P(k)\propto k^{-1}$ to $k^{-2}$, with progressively shallower slopes for tracers sensitive to more recent star formation, reflecting enhanced small-scale power and greater spatial intermittency in young stellar populations. This behaviour is consistent with a highly compressible, shock-dominated ISM in which stellar feedback and outflows reorganise dense gas into filamentary and clumpy structures. We also find that galaxy clumpiness depends on the treatment of stellar feedback. Weaker early stellar feedback enhances small-scale power in both the mass and light distributions. Clumpiness also varies strongly over the bursty star formation cycle, implying that observed samples may be biased towards galaxies caught in phases of elevated star formation. Galaxy clumpiness, therefore, could provide a complementary probe of the bursty star formation in the early Universe.

astro-ph.GA↗

arXiv2026

MusicLayout: Explicit Structural Planning for Controllable Text-to-Music Generation

Text-to-music generation has advanced rapidly, but current systems still rely primarily on global text prompts, leaving the structural organization of generated music implicit and difficult to inspect, control, or revise before audio generation. To address this issue, we introduce MusicLayout, an explicit intermediate representation for controlling musical structure in text-to-music generation. MusicLayout describes a musical piece as a time-aligned layout of sections, textures, repetitions, variations, and instrument-level arrangements, serving as an interpretable planning layer between textual intent and the generated music. We integrate MusicLayout into a text-to-music framework built on a unified autoregressive formulation, where the model first generates a MusicLayout representation and subsequently predicts audio tokens conditioned on this representation within a single sequence. The resulting MusicLayout can be inspected and modified prior to audio generation, providing a mechanism for layout-level structural control. We evaluate MusicLayout through layout-conditioned generation, layout manipulation experiments, and matched-data ablations, providing evidence that explicit layout planning can improve long-range structural organization and support layout-level control. We have released the implementation as open source on GitHub at https://github.com/XaryLee/MusicLayout.

cs.SD↗

arXiv2026

WorldSimProbe: Diagnosing Simulator Faithfulness in Action-Conditioned World Models for Embodied Manipulation

Action-conditioned world models (ACWMs) promise to provide embodied AI with scalable predictive simulators for planning, policy evaluation, and data generation. Realizing this promise requires precise action-conditioned transitions rather than merely plausible outputs. Yet their applicability remains difficult to establish because prevailing evaluations emphasize visual quality, task outcomes, or coarse rollout-level responsiveness without directly testing simulator fidelity. To address this gap, we evaluate ACWMs through the observable capabilities expected of physical simulators. Accordingly, we formalize Observable Simulator Contract, a minimal contract that any action-conditioned physical simulator should satisfy: supplied actions must induce corresponding agent motion, and environment responses must be grounded in that realized motion. To operationalize this contract, we introduce WorldSimProbe, comprising five controlled suites spanning local control sensitivity, global trajectory variation, source-diverse actions, interaction grounding, and dynamics. Suite-specific evaluators assess simulator-relative calibration, dense action-to-motion correspondence, false-interaction grounding, and primitive-level dynamics. We evaluate six open-source ACWMs on more than 18,000 instances across RoboTwin, ManiSkill, and LIBERO. World-SimProbe reveals systematic action-realization degradation across control variation, structured failures in interaction grounding and dynamics, and benchmark signals consistent with human judgments and downstream outcomes. Together, this capability-based framework provides a transparent, and standardized paradigm for diagnosing ACWM simulator fidelity beyond coarse, task-directed evaluation.

cs.RO↗

arXiv2026

MMAligner: Safeguarding Multimodal Large Language Models through Representation Calibration

Multimodal large language models (MLLMs) often refuse unsafe text prompts yet generate harmful responses to semantically equivalent multimodal inputs. Existing defenses either rely on external guardrails, which add inference overhead without repairing intrinsic flaws, or safety fine-tuning, which treats alignment as black-box optimization and may sacrifice utility or require large multimodal datasets. To identify the cause of this safety disparity, we analyze MLLM representations geometrically. We find that safety mechanisms learned from text persist across modalities: a shared safety subspace and refusal boundary remain effective, and representations inside this boundary consistently trigger refusals. However, unsafe multimodal inputs undergo a representation shift that places most of them outside the boundary, allowing them to bypass the model's intrinsic safety mechanism. This indicates that multimodal safety degradation stems from representation misalignment rather than the absence of safety capability. Based on this finding, we propose MMAligner, a safeguarding method that calibrates unsafe multimodal representations into the pre-existing refusal region. MMAligner applies a hard lower bound to ensure refusal, a soft upper bound to avoid excessive modification, and a preservation objective for benign inputs. Experiments across multiple open-source MLLMs show that MMAligner raises the average refusal rate on unsafe multimodal inputs to 99% with less than 2% utility degradation and minimal training data, substantially improving the safety-utility trade-off over existing baselines. (*Due to the notification from arXiv, "The Abstract field cannot be longer than 1,920 characters", the Abstract that appeared is shortened.)

cs.CR↗

arXiv2026

Photonic-chip-based generation of sub-100-femtosecond optical frequency combs

Sub-100-fs optical pulses and frequency comb sources have been revolutionizing a wide range of applications, from ultrafast optical science to optical frequency standard and measurement. To date, the leading techniques for generating such pulses in practical systems rely on tabletop mode-locked lasers, which inherently suffer from high system complexity, limited long-term reliability, and pronounced environmental sensitivity. Meanwhile, driven by advances in photonic integration, chip-scale approaches have sought to realize miniaturized pulse sources. However, simultaneously achieving sub-100-fs duration, ideal pulse shape, and a broadband flat-topped spectrum remains a significant challenge. Here, we address these challenges by combining two key photonic chip technologies: TFLN EO modulators for picosecond seed pulse generation, and highly nonlinear optical loop mirrors (NOLM) based on AlGaAsOI nanowaveguides for efficient temporal pulse cleaning and spectral broadening. In theoretical simulation and experiment, we show that for an input seed pulse centred at ~1550nm, a single-stage AlGaAs NOLM with a loop length of 1cm can produce flat-topped, nearly tenfold spectral broadening and over tenfold compression of pulse width, and more than 10dB suppression of pulse pedestals. Using initial EO comb pulses with ps-level durations at repetition rates of 10-20GHz, we demonstrate photonic-chip-enabled pulses with an unprecedented duration of 55fs and a flat-topped comb spectrum whose 10dB optical bandwidth exceeds 90nm. Our results highlight the remarkable potential of photonic chip technologies to realize high-repetition-rate, miniaturized sub-100-fs optical pulse generators with the prospect of superior stability and operability. The demonstrated photonic-chip-based sub-100-fs optical frequency comb sources may establish a new paradigm for both scientific research and practical applications.

physics.optics↗

arXiv2026

Shared Prefixes, Better Credit: Adaptive Routing for Multi-Agent Reasoning

Multi-agent reasoning (MAR) improves reasoning reliability through iterative solution exchange and refinement. Existing adaptive MAR methods typically learn routing decisions from query-level labels or trajectory-level returns, but such coarse supervision cannot accurately estimate the state-conditioned utility of individual operators in multi-step collaboration. We propose TreeCredit, a shared-prefix credit assignment framework for efficient adaptive MAR. Its core insight is to estimate operator utility through state-matched downstream comparisons, rather than directly attributing trajectory-level outcomes to preceding decisions. TreeCredit constructs shared-prefix collaboration trees by expanding candidate operators from the same intermediate state and assigns each state--operator pair a correctness-prioritized suffix credit based on the terminal correctness and cumulative additional cost of its complete continuation. These structured credits are converted into state-local operator preferences to train a lightweight pairwise state router, which dynamically selects the next admissible operator during inference. Experiments on six reasoning benchmarks show that TreeCredit modestly improves accuracy while substantially reducing inference cost, achieving a better accuracy--cost trade-off than representative MAR methods.

cs.AI↗

arXiv2026

PatchWorld: Gradient-Free Optimization of Executable World Models for Agent Environments

World models for interactive text agents must typically be learned from observation-action trajectories alone. Specifically, the environment returns text observations after each action, but does not expose a ground-truth latent state nor an inspectable transition model.A research gap remains in how to induce executable code as a world model in this black-box setting for prediction and agent decision making. We introduce PatchWorld, a gradient-free framework that turns offline trajectories into executable Python world models through counterexample-guided code repair.Instead of predicting the next observation with a black-box model, PatchWorld induces symbolic belief-state programs whose action updates can be inspected, replayed, and locally patched. Across seven AgentGym environments, PatchWorld-Simple achieves the highest code-based decision-making score among evaluated methods (76.4% macro success in live one-step lookahead), matching or exceeding LLM-based lookahead while invoking no LLM calls inside the world-model prediction module itself. We further find that a human-specified residual-memory bias improves surface observation fidelity but weakens agent decision-making utility. This reveals a tradeoff in executable world models, since improving observation fidelity can come at the expense of action-discriminative dynamics, and vice versa. Code is available at https://github.com/HKBU-KnowComp/PatchWorld.

cs.CL↗

arXiv2026

Algorithmic Separation between Constant-Depth and Logarithmic-Depth Neural Networks

Despite the empirical advantages of deep networks over shallow ones, theoretical depth separations largely concern approximation power, while algorithmic results are mostly limited to comparisons between two- and three-layer networks. In this work, we prove the first algorithmic separation between constant-depth and logarithmic-depth networks. Specifically, we identify a class of Boolean functions with hierarchically structured Fourier spectra that logarithmic-depth networks can learn efficiently using layerwise coordinate descent by reconstructing the spectra hierarchically and adaptively. We also exhibit a subclass for which every constant-depth, polynomial-width network with sufficiently regular activations and controlled spectral norms must incur constant $L^2$ approximation error under the uniform distribution over the hypercube.

cs.LG↗

arXiv2026

Who Pays the Price? Stakeholder-Centric Prompt Injection Benchmarking for Real-world Web Agents

LLM-based web agents are increasingly deployed in real-world settings such as e-commerce, where they interact extensively with untrusted web content while executing actions that carry direct financial consequences. This makes them vulnerable to prompt-injection attacks, in which seemingly benign web content conceals adversarial instructions that manipulate the agent's behavior. Existing security benchmarks adopt an \textit{attack-centric} perspective, focusing on the technical feasibility of injections while overlooking the nuanced distribution of resulting harms. In practice, however, prompt-injection risk is victim-dependent: a single exploit can produce asymmetric consequences for different stakeholders, and the same attack pattern may exhibit substantially different effectiveness depending on whom it targets. To capture these properties, we introduce StakeBench, a stakeholder-centric benchmark that systematically categorizes and attributes harm in real-world web agent systems for online shopping. In general, StakeBench decomposes prompt-injection risk into 12 concrete attack objectives across three stakeholder classes, realized by 22 reusable templates and instantiated into 264 executable adversarial cases spanning 12 product categories, with each case evaluated along complementary outcome- and process-level metrics. Evaluating four deployable agent-backbone configurations across 3,168 attacked runs, we find substantial and heterogeneous vulnerabilities: no attack objective is reliably resisted by current LLM-based web agents, and outcomes span four qualitatively distinct modes. These patterns are missed by conventional attack-centric, single-metric evaluation, underscoring the need for stakeholder-aware assessment of LLM-based agents in real-world deployments.

cs.CR↗

arXiv2026

Interface-Confined Superconductivity with Thickness-Independent Superfluid Stiffness in (Pb,Sn)Te/FeTe Bilayers

Interface-induced superconductivity in FeTe-based heterostructures provides a promising route toward topological superconductivity, yet the roles of the neighboring layers topology, symmetry, and electronic structure remain unresolved. In this work, we employ molecular beam epitaxy to grow Pb1-xSnxTe/FeTe bilayers and use angle-resolved photoemission spectroscopy to track the evolution of the Pb1-xSnxTe layer from a trivial insulator to a topological crystalline insulator hosting multiple Dirac surface states. Electrical transport measurements reveal robust superconductivity throughout the entire composition range, with a nearly constant superconducting transition temperature of ~12 K despite substantial changes in the electronic structure and topology of Pb1-xSnxTe. Double-coil mutual-inductance measurements further reveal comparable superfluid stiffness across the topological phase transition and nearly thickness-independent superfluid stiffness despite large variations in the constituent-layer thicknesses, demonstrating that superconductivity is confined near the interface. These results establish that superconductivity in FeTe-based heterostructures is largely insensitive to the topology, crystal symmetry, and detailed electronic structure of the neighboring layer, supporting a primary origin in modifications to the FeTe layer. The coexistence of interface-confined superconductivity and tunable multiple Dirac surface states in Pb1-xSnxTe/FeTe bilayers provides a versatile platform for exploring topological superconductivity and interactions among multiple Majorana zero modes.

cond-mat.supr-con↗

arXiv2026

CRAFT: Clustering Rubrics to Diagnose Weak LLM Capabilities and Generate Targeted Fine-Tuning Data

Evaluations should do more than measure a models current performance. They should tell us what to fix for the next model iteration and provide a way to generate targeted post training data. Most evaluation pipelines identify weak examples, topics, or categories, but they leave the underlying capability failure implicit: they say where a model fails, not why. We introduce CRAFT, a method that converts any rubric based evaluation dataset into a model specific diagnosis of weak capabilities. CRAFT treats each grading criterion as a capability probe: it extracts a capability description from every prompt rubric pair, clusters these descriptions into a hierarchical capability tree, scores the target model at every node, and selects low performing nodes dynamically across tree levels, at the granularity where each failure is clearest. The selected weak capabilities then direct the generation of targeted supervised finetuning data. Holding the data generation, finetuning, and evaluation setup fixed, we compare CRAFT against prompt level EvalTree clustering and untargeted random generation on four open source models, two professional domains (finance and legal), and 13 held out benchmarks disjoint from the diagnostic data. CRAFT achieves the strongest finance domain average for all four models under repeated temperature decoding; on legal domain, it is strongest for three of four models and remains within the decoding variance bands of the best baseline on the fourth. Diagnosing weaknesses at the level of rubric criteria, rather than prompts or categories, thus yields both a sharper picture of what a model cannot do and measurably better models after finetuning on that diagnosis.

cs.AI↗

arXiv2026

WebDesignIter: Co-Evolving Design Knowledge for Repository-Level Front-End Code Generation

Front-end development accumulates change after change at the repository level, weaving complex cross-file dependencies that current LLM coding agents tuned for single-shot tasks cannot reliably track across multiple iterations, leading to functional regressions and code that resists maintenance. We argue the missing piece is design knowledge: architectural principles, module responsibilities, and structural constraints that developers lean on to keep code readable, maintainable, and evolvable as a system scales. To operationalize this, we propose WebDesignIter, a framework built around a persistent knowledge graph (WebAppArchKG) that fuses repository structure with design knowledge and keeps both in sync across development cycles. WebDesignIter works in two stages: design-informed planning pulls historical context and architectural overviews from WebAppArchKG to produce an implementation plan with corresponding test scripts, and design-aware generation executes that plan through targeted diff-based patches, validated by sandbox execution and automatic syntax repair. On Web-Bench, WebDesignIter delivers an average Pass@2 gain of 9.55 percentage points across nine foundation models over existing baselines. More importantly, WebDesignIter outperforms every general-purpose coding agent Claude Code, OpenHands, SWE-Agent, Codex CLI on every model configuration, posting the highest Pass@1 and Pass@2 while consuming 2530 fewer input tokens. Ablation singles out design knowledge as the most impactful component: stripping it drops Pass@1 by 11.40 percentage points, a degradation far larger than removing code-graph retrieval, patch-based generation, or sandbox verification, confirming that design knowledge provides a fundamentally more efficient and reliable path to repository-level code generation.

cs.SE↗

arXiv2026

Game Theory Driven Multi-Agent Framework Mitigates Language Model Hallucination

The application of lightweight Large Language Models in rule-based scientific domains remains severely limited by their tendency to mimic linguistic patterns rather than reproduce axiomatic reasoning, causing frequent hallucinations. Here, we show that G-Frame, an adaptive multi-agent framework integrating Bayesian and team game principles, establishes an automated closed-loop for high-quality data synthesis and model training. By forcing the internalization of domain constraints through structured reasoning, we synthesized a specialized corpus of 363,045 chains-of-thought and 199,589 question-answer pairs. The resulting 7B model OmniChem achieves performance parity with GPT 4o mini on custom benchmarks and ChemBench while exhibiting a 79.46% reduction in hallucinations relative to its base architecture. We further demonstrate the advanced capabilities of OmniChem in molecular design and synthesis planning. This work establishes a scalable paradigm utilizing adaptive multi-agents to overcome inherent reasoning deficiencies, offering a feasible pathway for accelerating knowledge discovery in specialized scientific fields.

cs.AI↗

arXiv2026

Measuring the Security of Mobile LLM Agents under Adversarial Prompts from Untrusted Third-Party Channels

Large Language Models (LLMs) have transformed software development, enabling AI-powered applications known as LLM-based agents that promise to automate tasks across diverse apps and workflows. Yet, the security implications of deploying such agents in adversarial mobile environments remain poorly understood. In this paper, we present the first systematic study of security risks in mobile LLM agents. We design and evaluate a suite of adversarial case studies, ranging from opportunistic manipulations such as pop-up advertisements to advanced, end-to-end workflows involving malware installation and cross-app data exfiltration. Our evaluation covers eight state-of-the-art mobile agents across three architectures, with over 2,000 adversarial and paired benign trials. The results reveal systemic vulnerabilities: low-barrier vectors such as fraudulent ads succeed with over 80% reliability, while even workflows requiring the circumvention of operating-system warnings, such as malware installation, are consistently completed by advanced multi-app agents. By mapping these attacks to the MITRE ATT&CK Mobile framework, we uncover novel privilege-escalation and persistence pathways unique to LLM-driven automation. Collectively, our findings provide the first end-to-end evidence that mobile LLM agents are exploitable in realistic adversarial settings, where untrusted third-party channels (e.g., ads, embedded webviews, cross-app notifications) are an inherent part of the mobile ecosystem.

cs.CR↗