arXiv subjects

Zihao Wang

Publications and source records attributed to Zihao Wang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Dex-X: Learning Visual-Tactile Dexterous Manipulation From Human Videos with Simulated Interaction

Human videos are an abundant source of dexterous manipulation behaviors, but they lack tactile information that is crucial for contact-rich interaction. This raises a fundamental question: can robots learn deployable visual-tactile dexterous manipulation policies from human video demonstrations without robot-side data collection? We present DEX-X, a framework for learning visual-tactile dexterous manipulation from human videos through simulation. Our key insight is that simulation can serve as a tactile completion engine. Given monocular human demonstrations, DEX-X reconstructs hand-object interactions in simulation, where physically grounded contact dynamics provide tactile supervision unavailable in the original videos. Leveraging this recovered tactile information, we train visual-tactile dexterous manipulation policies and distill them into deployable policies operating on point-cloud observations and tactile sensing. We demonstrate zero-shot sim-to-real transfer on a dexterous hand-arm platform across diverse grasping and contact-rich tool-use tasks. The teacher policy achieves 65.9% average success across six task categories in simulation, while the distilled visual-tactile policy achieves 93% success on real-world cube picking and 53% on the challenging table-cleaning task. Zero-shot generalization to unseen object geometries is also observed on object-picking tasks. Our results suggest that simulated interaction is a key bridge between human videos and deployable dexterous manipulation policies, providing the missing physical supervision needed for scalable robot skill learning from Internet-scale human video data.

cs.RO↗

arXiv2026

Reliability-aware short-term roll prediction for unmanned surface vehicles via multi-task learning and adaptive centralization

Reliable roll prediction of unmanned surface vehicles (USVs) is essential for ensuring navi?gational safety and enhancing autonomous decision-making. While existing studies primarily focus on improving prediction accuracy, the quantification of prediction reliability remains insufficiently addressed. To bridge this gap, this paper proposes a reliability-aware prediction paradigm that integrates confidence assessment into the predictive pipeline. The architecture utilizes a multi-task learning structure where a shared feature extraction backbone feeds into dual heads: a regression head for precise roll prediction and a quantification head for confidence scoring. This configuration provides accurate prediction and corresponding confidence for risk?sensitive downstream tasks. In addition, an adaptive centralization strategy tailored for short?term real-time roll prediction is introduced to improve model generalization under varying operational conditions. Experiments conducted on a real-sea dataset demonstrate that the proposed method effectively quantifies the reliability of prediction results and maintains superior generalization under varying conditions, offering significant potential for practical engineering applications.

cs.LG↗

arXiv2026

Uni-VLaT: Whole-Body Tactile Adaptation of VLA Policies for Humanoid Loco-Manipulation

Physical contact often determines how a humanoid should respond during loco-manipulation, yet vision and proprioception alone are often insufficient to characterize physical interaction, especially when the contact region is occluded. Unlike sparse force or torque measurements at predefined regions, distributed tactile sensing preserves spatially resolved contact patterns across the robot body. We therefore study how to integrate such whole-body tactile information into vision-language-action (VLA) policies for contact-rich control. Our approach, Uni-VLaT, introduces a tactile pathway whose latent state is trained not only for action generation, but also to predict future tactile, proprioceptive, and visual representations. This predictive objective builds a tactile-anchored multimodal context, encouraging a more structured understanding of the physical world. We evaluate Uni-VLaT on five real-robot tasks covering tactile-triggered locomotion, sustained physical interaction, human-robot contact, and loco-manipulation. Uni-VLaT achieves a 75% average success rate, outperforming a baseline without tactile input by 43 points and a tactile-input baseline without predictive supervision by 7 points. Across two pretrained VLA backbones, our method improves Table Sweeping by 30 points on both backbones and Back-Tap Walking by 85-90 points. Ablations further show that contextualized tactile prediction and absolute future targets are critical to performance. These results indicate that predictive tactile learning provides an effective route for extending pretrained VLA policies to whole-body physical interaction.

cs.RO↗

arXiv2026

Investigating Human--AI Discrepancies via Multiple-Solution Problems

Frontier artificial intelligence (AI) models are benchmarked on whether they reach a correct answer. Yet many problems admit several correct answers and repeated attempts, by different people or by the same model resampled, trace out a distribution over them. In this work, we ask whether human and model reasoning lead to different distributions over valid solutions. Our testbed comprises 270 reasoning puzzles across five puzzle families. These multiple-solution puzzles each have 3 to 8 valid solutions and are simple enough that humans and models can solve them reliably. The resulting distributions differ markedly: models differ from one another, yet resemble each other far more than they resemble humans. Model distributions are, moreover, within every puzzle family, less diverse than human ones. We compare these discrepancies across puzzle categories, and trace how they respond to reasoning-effort settings, to prompting, and to perturbations of the puzzle that leave its solutions unchanged. Together, these results point at significant differences between human and AI problem-solving processes, and their choice among equally defensible solutions. As progressive deployment of AI systems in society comes into focus, evaluating such differences (beyond one-dimensional accuracy metrics) is increasingly important. Data and code are available at https://hai-discrepancies.github.io/

cs.CY↗

arXiv2026

SSIMuse: Assessing Data Replication in Symbolic Music via an Adapted Structural Similarity Index Measure

AI-generated music may inadvertently replicate training samples and raise plagiarism concerns. Similarity measures can quantify such replication, thereby offering supervision and guidance for music generation models. Existing similarity measures for symbolic music mainly target monophonic melodies and provide limited support for complex musical textures, which restricts their ability to assess replication in polyphonic music. To address this limitation, we explore a simple cross-domain idea and propose \textbf{SSIMuse}, the first adaptation of the Structural Similarity Index Measure (SSIM) from images to symbolic music. SSIMuse is based on an image-like representation of symbolic music using binary piano rolls. It reinterprets the SSIM components as note density consistency and local note event overlap, and searches for replication under temporal shifts, pitch transpositions, and musically meaningful duration ratios. Controlled experiments on polyphonic and melodic datasets show that SSIMuse detects replication of at least one bar under exact and time-scaled conditions. Real-case melody retrieval further shows that it remains competitive with one symbolic plagiarism detector and two audio embedding models. The code is available at https://github.com/Tayjsl97/SSIMuse.

eess.AS↗

arXiv2026

Ion-temperature-gradient turbulence from finite to weak magnetic shear regime

The tokamak, a toroidal magnetic device confining a hot plasma, is one of the most advanced approaches to fusion energy. A central obstacle is turbulent transport driven by the ion-temperature-gradient (ITG) mode. Operation scenarios that rely on a weak or zero magnetic shear core, such as the ITER hybrid scenario, have achieved markedly improved confinement, yet the underlying mechanism has remained unclear. Here we show that the radial width of ITG poloidal harmonics is constrained not only by the familiar parallel Landau damping, but also by the isotropy of micro-turbulence in the plane perpendicular to the magnetic field. In the weak-shear limit, micro-isotropy dominates and constrains the radial width at one poloidal wavelength, yielding a critical magnetic shear $s_{\text{crit}} \approx 1/(2π)$. Above this threshold, the extended Type II ballooning modes prevail; below it, the localized Type I modes emerge, comprising only two or three harmonics. Global gyrokinetic simulations spanning DIII-D, JET~and ITER parameters confirm this geometric criterion. Nonlinear simulations with sustained heating show that weak-shear plasmas spontaneously form internal transport barriers with the turbulence suppressed in the radial region determined by $|s| < s_{\text{crit}}$. This geometric criterion defines the weak-magnetic-shear regime and offers a fresh perspective on turbulent transport in fusion plasmas.

physics.plasm-ph↗

arXiv2026

A Comprehensive Review of Large Language Models for Nanophotonics: From Surrogate Modeling to Autonomous Design

Metasurfaces have revolutionized the development of photonic devices by enabling unprecedented precision in light manipulation. However, their design processes are often constrained by computationally expensive simulations and complex high-dimensional design spaces. Although deep learning has accelerated the design process by serving as a surrogate model, it remains constrained by task-specific architectures and lacks universal reasoning capabilities. This review surveys how Large Language Models (LLMs) are adding semantic interfaces, code generation, and tool orchestration to established numerical nanophotonic workflows. We first outline the development from classical neural networks to transformer-based models and their applications in nanophotonic design. We then review the emergence of LLM-related methods in nanophotonics and organize them into two operational modes: surrogate models that treat structure-spectrum mapping as a language task, and agentic systems that have been demonstrated to generate code, orchestrate selected simulation steps, and support closed-loop optimization. Furthermore, to identify future cross-disciplinary opportunities, we briefly explore applications of LLMs in research fields such as materials science and wireless communications. This review concludes by looking ahead to the next generation of multimodal foundation models with physical perception capabilities. In this vision, artificial intelligence is evolving from passive tools into active collaborators, participating in autonomous scientific discovery.

physics.optics↗

arXiv2026

Large-Amplitude Steady Solitary Water Waves with General Vorticity

We study two-dimensional steady solitary gravity water waves with general vorticity, allowing for overhanging free-surface profiles. The main challenges arise from the free boundary, the unbounded fluid domain, and the inherent complexity of the general vorticity setting. To address these, we introduce a conformal reformulation that reduces the problem to an equivalent system on a fixed strip, consisting of an overdetermined elliptic problem coupled with an elliptic boundary-value problem. This framework enables a local analysis without imposing restrictive assumptions on the vorticity. Using the center-manifold construction of Chen, Walsh and Wheeler \cite{chennonlinearity}, we first establish the existence of small-amplitude solitary waves for such non-trivial vorticity distributions. Subsequently, via a global analytic bifurcation argument, we prove the existence of continuous branches of large-amplitude solitary waves for analytic vorticity functions. Along these global solution curves, the free surfaces may develop overhanging profiles and need not remain graphs over the horizontal coordinate. Our results provide a new constructive framework for large-amplitude solitary waves with general vorticity, extending the existing theory beyond the classical settings of constant vorticity \cite{susannaarma} and non-overhanging profiles \cite{milesjma}.

math.AP↗

arXiv2026

Global bifurcation of water-wave fronts generated by surface pressure

We study two-dimensional steady gravity water waves with general vorticity in a homogeneous, inviscid fluid of finite depth, subject to a surface pressure that tends to distinct constants upstream and downstream. Starting from a fixed supercritical shear flow, we construct two global branches of monotone fronts connecting distinct laminar states, with increasing and decreasing profiles corresponding to the two signs of the pressure transition. Small-amplitude fronts are first obtained via the implicit function theorem and then extended by analytic global continuation.

math.AP↗

arXiv2026

Endpoint Riesz transforms on normalized graphs

For an infinite connected locally finite undirected graph with positive symmetric edge weights and normalized Laplacian $Δ$, we prove weak type $(1,1)$ estimates for two Riesz transforms. The edge transform has universal constant $2$, with no volume assumption. The vertex carré-du-champ transform has constant $H_1+1$ whenever $H_1=\sup_x m(B(x,1))/m(x)<\infty$. These endpoint constants apply to real-valued inputs and yield strong $\ell^p$ bounds for $1<p\le2$. The proof uses a discrete obstacle decomposition obtained by a contraction mapping, followed by elementary support estimates for the two gradients.

math.CA↗

arXiv2026

IndirectAD: Practical Data Poisoning Attacks against Recommender Systems for Item Promotion

Recommender systems play a central role in digital platforms by providing personalized content. They often use methods such as collaborative filtering and machine learning to accurately predict user preferences. Although these systems offer substantial benefits, they are vulnerable to security and privacy threats, especially data poisoning attacks. By inserting misleading data, attackers can manipulate recommendations for purposes ranging from boosting product visibility to shaping public opinion. Despite these risks, concerns are often downplayed because such attacks typically require controlling at least 1% of the platform's user base, a difficult task on large platforms. We tackle this issue by introducing the IndirectAD attack, inspired by Trojan attacks on machine learning. IndirectAD reduces the need for a high poisoning ratio through a trigger item that is easier to recommend to the target users. Rather than directly promoting a target item that does not match a user's interests, IndirectAD first promotes the trigger item, then transfers that advantage to the target item by creating co-occurrence data between them. This indirect strategy delivers a stronger promotion effect while using fewer controlled user accounts. Our extensive experiments on multiple datasets and recommender systems show that IndirectAD can cause noticeable impact with only 0.05% of the platform's user base. On public benchmark datasets, IndirectAD remains effective under very low poisoning ratios, suggesting that recommender-system poisoning may be more practical than previously assumed.

cs.CR↗

arXiv2026

Monet: Measuring the Ecosystem of Open-Source Text-to-Image Models Tailored for Harmful Services

The open-source text-to-image (T2I) ecosystem enables rapid model development and sharing, but also hosts models intentionally tailored for harmful services, which we call Monets. Prior work has examined specific types of harmful T2I models on individual platforms, but a Monet does not exist in isolation. The broader Monet ecosystem, spanning model characteristics, cross-platform propagation, governance evasion, monetization, and downstream deployment, remains poorly understood. In this study, we present the first systematic, ecosystem-level measurement of Monets. Grounded in the policies of real-world model hubs, we construct a taxonomy of ten harmful service categories and identify 23,947 Monets across eight major T2I model hubs, with the most popular exceeding 19 million downloads. While some developers employ anti-theft mechanisms against unauthorized re-uploading, Monets propagate across platforms at scale, with 40.76% mirrored across hubs. Such propagation further enables governance evasion via cross-platform archiving, keeping 11.99% of Monets accessible after bans on their original platforms, alongside other evasion strategies including keyword obfuscation and model-level safeguard circumvention. Monets also anchor coordinated commercial campaigns---one spanning 668 models with 914 completed commissions and another advertising gray-market account-farming service---and reach users through GitHub projects and inference APIs, raising downstream child safety concerns. These findings expose the limitations of platform-siloed defenses and highlight the need for cross-platform threat intelligence, coordinated governance, and technical safeguards.

cs.CR↗

arXiv2026

Electronic Reconstruction Towards Topological Superconductivity in FeTe

The recent discovery of intrinsic superconductivity in stoichiometric FeTe films has renewed interest in the Te-rich end member of the iron chalcogenides for studies of unconventional and topological superconductivity, yet its intrinsic electronic structure remains unresolved. In this work, we combine molecular beam epitaxy, angle-resolved photoemission spectroscopy (ARPES), electrical transport measurements, density functional theory, and embedded dynamical mean-field theory to track the electronic reconstruction of 20-unit-cell FeTe films as Te annealing progressively removes excess interstitial Fe and drives the system from an antiferromagnetic metal to a superconductor. We find that this evolution is accompanied by recovered quasiparticle coherence, reduced electronic correlations, a Lifshitz transition, and a topological phase transition, yielding dxy-dominated hole and electron pockets that favor inter-pocket scattering. In addition, a shallow dxz/dyz-derived hole band located about 2 meV below the Fermi level may provide an incipient-band pairing channel, while scattering between the two electron pockets at M may offer additional pairing channels. High-resolution polarization-dependent laser ARPES measurements further reveal a topological surface state whose circular dichroism is consistent with the expected orbital-angular-momentum texture of stoichiometric FeTe. These results establish the intrinsic low-energy electronic structure of superconducting FeTe and identify the electronic states most relevant to superconductivity. The coexistence of intrinsic superconductivity and a topological surface state establishes stoichiometric FeTe as a promising platform for exploring topological superconductivity and possible Majorana bound states.

cond-mat.supr-con↗

arXiv2026

Minimality and Nonlinear Spectral Rigidity of Hypersurfaces in Real Space Forms

We study nonlinear mean-curvature spectral equations for hypersurfaces in Euclidean, spherical and hyperbolic space. For an admissible analytic response $F$ and ambient sectional curvature $c$, we prove minimality when the spectral parameter satisfies $σ\ge nc$, and strict mean-curvature bounds at gradient points when $σ 1$, and positive mixtures satisfying an explicit width condition. In particular, the result proves Chen's conjecture in Euclidean codimension one and biharmonic minimality in constant negative curvature, without completeness or holonomicity. A common real-analytic continuation and focal-trace argument treats all three geometries. Explicit umbilical models show that the unconditional spectral minimality range is sharp. Independently, we prove quantitative local rigidity, modulo ambient isometries, of compact nonminimal geodesic spheres in all three geometries and proper equal-radius Clifford hypersurfaces in spheres, for smooth laws with $F,F'>0$ near the model amplitude.

math.DG↗

arXiv2026

Halo assembly bias in the early Universe: a clustering probe of the origin of the Little Red Dots

The clustering of galaxies encodes key information about the structure and assembly history of their host dark matter (DM) haloes, providing a powerful probe of the origin of extreme high-redshift systems. While halo assembly bias has been extensively studied at low redshift, its behavior in the early Universe remains poorly explored. Using the large-volume, high-resolution Shin-Uchuu cosmological $N$-body simulation, we characterize halo assembly bias associated with formation time, concentration, and angular momentum across a wide range of halo masses and redshifts. We find that the sign and amplitude of assembly bias depend on halo mass for both concentration and spin. High-concentration and low-spin haloes are more strongly clustered below characteristic peak heights of $ν\sim 1.5$ and $\sim 0.75$, respectively, while the trends weaken or reverse at higher masses. Halo age bias persists at all redshifts but decreases toward higher masses and earlier cosmic times. We apply these results to assess whether clustering can distinguish competing formation scenarios for the Little Red Dots (LRDs). We find that the direct-collapse-black-hole (DCBH) scenario predicts the strongest large-scale bias and enhanced pair fractions, the self-interacting-dark-matter (SIDM) core-collapse scenario and low-spin compact-galaxy scenarios yield weaker clustering due to lower characteristic halo masses and spin-related secondary bias, and a primordial-black-hole (PBH) scenario predicts unbiased clustering. Our results demonstrate that halo assembly bias and characteristic host masses provide powerful diagnostics for constraining the physical origin of LRDs, offering testable predictions for upcoming clustering measurements with JWST and future deep surveys.

astro-ph.CO↗

arXiv2026

TTE-Flash: Accelerating Reasoning-based Multimodal Representations via Think-Then-Embed Tokens

Recent research has demonstrated that Universal Multimodal Embedding (UME) benefits significantly from Chain-of-Thought (CoT) reasoning. In this paradigm, a generative model produces explicit reasoning traces for a multimodal query, with the final representation extracted from an embedding token attending to both the query and the reasoning. Despite its effectiveness, the computational overhead of generating explicit CoT traces is often prohibitive. In this work, we propose replacing explicit CoT with latent think tokens, which are interpreted as latent variables that can produce explicit CoT traces as observed variables. By optimizing think tokens using CoT generation loss and subsequent embedding tokens using contrastive loss, we produce high-performance, reasoning-aware representations at a constant inference cost. Our study investigates two key architectural designs: 1) how think and embeddings tokens should be extracted from the same LLM backbone. 2) how the tokens should be trained as two dependent tasks. We introduce TTE-Flash-2B, a reasoning-aware multimodal representation model that outperforms its explicit-CoT counterpart on the MMEB-v2 benchmark, while producing latent think tokens that are interpretable both textually and visually. Furthermore, zero-shot evaluation across 15 video datasets reveals scaling behavior as the number of think tokens increases.

cs.AI↗

arXiv2026

Collaborative Memory for Multi-Agent VLM Systems

Vision-language model (VLM) agents combine specialized perception, tools, and reasoning to address complex visual tasks. In multi-agent settings, different agents inspect different image regions, video frames, or visual representations, so collaboration extends beyond distributed reasoning to distributed perception. This makes shared visual context a central problem in VLM agent collaboration. In this paper, we frame memory hierarchy, cross-agent sharing, and consistency mechanisms around the need to reconcile interpretations and update dependent reasoning. Effective collaboration requires agents to build on contributions from other agents, recover missing visual context, and reconcile differing interpretations as new evidence emerges. Shared visual memory preserves not only images or textual summaries but also the dependencies among observations, agent interpretations, and subsequent reasoning. Together, these design considerations shape how information flows and evolves across VLM agents. The proposed framework provides a foundation for building reliable and resource-efficient agent teams.

cs.AI↗

arXiv2026

Strongly stable CMC-one hypersurfaces in every hyperbolic space of dimension at least four

For every integer $d\ge4$, we prove strong stability for a subfamily of classical rotational hypersurfaces in $\mathbb H^d$ with normalized mean curvature one. The examples are complete, two-sided, properly embedded, and nowhere umbilic, with topology $\mathbb {R}\times\mathbb{S}^{d-2}$. An explicit positive supersolution yields a quantitative stability inequality for all compactly supported test functions. Consequently, endpoint horospherical rigidity fails in every ambient dimension at least four.

math.DG↗