arXiv subjects

Zhen Li

Publications and source records attributed to Zhen Li.

At least 37 records · Page 2Linked to original sources

arXiv2026

MUGEN: Interactive Panoramic World Exploration via Camera Control

Interactive panoramic video generation aims to synthesize immersive 360\textdegree{} videos that remain visually coherent while following user-specified camera trajectories during exploration. However, progress is limited by a coupled data-and-model gap: existing panoramic video datasets are often short, weakly annotated, or lack camera trajectories, while existing camera-controlled video generation models are designed for perspective videos and do not directly support panoramic geometry. In this paper, we introduce MUGEN and Wan360 to address these limitations. MUGEN is a large-scale real-world panoramic video dataset tailored to interactive 360-degree world exploration, comprising over 1,300 hours of at least 4K panoramic videos with rich semantic and geometric annotations. Built on MUGEN, we further present Wan360, a camera-controllable interactive panoramic video generation model. Panoramic videos are commonly represented by EquiRectangular Projection (ERP), which unfolds a spherical 360-degree view into a rectangular frame with cyclic longitude seams and pole distortions. To this end, Wan360 introduces three parameter-free ERP-aware components: periodic longitude RoPE for seam-consistent positional encoding, ERP-aware padding for reducing boundary artifacts, and random roll yaw for consistent learning. For camera control, Wan360 uses a panoramic Plücker embedding that represents camera motion with ERP rays rather than perspective pinhole rays. Experiments show that MUGEN serves as a data foundation for panoramic world exploration, and that Wan360 enables high-quality, temporally coherent, camera-controllable 360-degree video generation.

cs.CV↗

arXiv2026

Positions Are Not Facts: The Mismatch Between KV Caches and Memory

When a fact changes, how should a language model update the history stored in its key-value (KV) cache? Hiding the old record is cheap, but it may still contain needed details or answer questions about the past. We compare hiding whole records, hiding only replaced values, and deleting old text and recomputing the cache. In a controlled quantity task, masking makes all eight models prefer the new value more strongly, yet six lose complete answers through unit errors or failure to stop; keeping the unit preserves all current answers. Later states also retain useful information from earlier records: on multi-hop updates, rebuilding these states at unchanged positions lowers historical accuracy by 20-41 percentage points, whereas moving the existing states has little effect. Keeping object dependencies and unchanged revision passages prevents many losses. Recognition is a separate challenge. Learned readouts recover distinctions missed by fixed cache similarities on synthetic record pairs. On natural text, text-detector-selected masks show no clear advantage over random masks at the same rate in 14 same-model detector-generator comparisons. Query-dependent access can avoid some losses, with additional storage or access costs. These findings identify what must be preserved beyond the replaced value when using a KV cache as updatable memory.

cs.CL↗

arXiv2026

OpenFlyScan: A Quality-Guided Aerial Reconstruction System for Consumer Drones

3D Gaussian Splatting (3DGS) provides high-fidelity scenes for large-scale embodied simulation, but constructing large-scale urban assets remains constrained by expensive equipment and delayed quality feedback. Preset surveys can leave complex surfaces insufficiently observed, with defects discovered only after reconstruction, requiring return visits and repeated processing. We present OpenFlyScan, a quality-guided aerial reconstruction system for consumer drones that integrates a GS quality model, a reacquisition planner, and a custom-designed mobile app. The model learns from GS rendering errors to predict regional reconstruction quality. Based on these predictions, the planner then generates complementary reacquisition strips to be executed through the app, which also supports automated oblique surveys and data transfer without additional hardware on board. Across real aerial scenes, the model effectively identifies regions that are likely to be poorly reconstructed. In the Expo West field experiment, targeted reacquisition improves PSNR at additional views by 10.95 dB. With consumer drones, OpenFlyScan integrates capture, targeted reacquisition, and reconstruction to support rapid, low-cost urban asset creation. Code and models will be made publicly available at https://openflyscan.github.io/.

cs.RO↗

arXiv2026

Accurate Motion Estimation with Bézier Control Point for Efficient Frame Interpolation

In frame interpolation tasks, motion ambiguity in the training set causes models to generate blurry intermediate frames. Moreover, the assumption of uniform motion between frames during inference further leads to inaccuracies in the generated intermediate frames. To tackle these challenges, we propose an Accurate motion estimation algorithm with Bézier Control point, ABC-Inter, for efficient frame Interpolation. Specifically, ABC-Inter designs an Accurate Flow estimation Module (AFM) by decoupling two-frame features and mapping to corresponding coordinates to better estimate the optical flow between the two frames. Furthermore, ABC-Inter eliminates motion ambiguity in the training set by introducing Bézier control points that are computed using the input frames and the intermediate ground-truth (gt) frames. This allows the model to estimate accurate optical flow between two frames during the training process, thereby solving the blurriness problem in the generated intermediate frames during inference. Benefiting from the more accurate flow estimation between two frames, we can introduce additional frames and directly use multiple flows to calculate Bézier control points for modeling non-uniform motion without retraining the model. Simultaneously, to realize the estimation of non-linear motion using only two frames, we also introduce a new Bézier control point estimation module which achieves better motion estimation between the two frames by performing fine-tuning on the model in the second stage. Experimental results demonstrate that our ABC-Inter achieves state-of-the-art performance on multiple benchmark datasets and exhibits excellent visual perception.

cs.CV↗

arXiv2026

Modeling Clinical Workflow for SYNTAX Scoring from Coronary Angiography Videos

The SYNTAX score is a clinically established tool for assessing anatomical lesion complexity in coronary artery disease and guiding subsequent treatment. However, automated SYNTAX scoring is commonly formulated as a direct regression problem from coronary angiography videos to patient-level scores. In this work, we reformulate SYNTAX scoring as a vessel segment identity-preserving anatomical reasoning problem and propose a hierarchical modeling framework that explicitly aligns learning with the clinical workflow. Our approach maintains vessel segment identity across frames and views, estimates stenosis severity at the segment level, and aggregates evidence hierarchically according to coronary anatomy. Simultaneously, to address the scarcity of domain-specific data, we integrate and complete multiple public coronary angiography datasets, constructing a large-scale resource featuring completed vessel segmentation and derived structural annotations. Experiments demonstrate that vessel segment-level stenosis embedding enhances explanatory power and reduces prediction variability compared to baseline models, with the R^2 score improving by 0.201 and dev STD decreasing by 18.4%. These results highlight the necessity of structure-aligned modeling for reliable and stable automated SYNTAX scoring from multi-view coronary angiography videos. The GitHub link is https://github.com/VersaceSu7/SYNTAX_score_777.

cs.CV↗

arXiv2026

MaskVLA: Visual Masking Against Trajectory Overfitting of Vision-Language-Action Model

Vision-Language-Action (VLA) models integrate vision-language understanding with executable robot actions, enabling end-to-end learning for robot control. However, our empirical analysis reveals that existing models exhibit severe trajectory overfitting when finetuned on limited datasets. To guide the model in effectively utilizing wrist camera information, we propose MaskVLA, a masking-based fine-tuning strategy. By randomly masking a small portion of the main camera's visual information, the model is guided to autonomously learn more fine-grained, task-relevant, and effective visual features. This process leads to the emergence of robust policies, thereby enhancing the model's capability to tackle complex manipulation tasks and improving its generalization performance. Our method has been comprehensively evaluated on RoboTwin 2.0, achieving an average success rate improvement of 23.2% and 16.8% compared to $π_0$ and OpenVLA-OFT, respectively. Furthermore, experiments on real-world ALOHA robots also demonstrate the effectiveness of our approach.

cs.CV↗

arXiv2026

End-to-End Latency-Minimizing and Load-Balanced Request Scheduling for Edge LLM Inference in Agentic AI Services

Large language model (LLM)-powered agentic AI services increasingly demand low-latency inference, motivating the deployment of LLMs across distributed edge servers. However, heterogeneous communication and computing capabilities, together with dynamically evolving inference states, make the edge server selection for each incoming request time-varying and tightly coupled across slots. In this paper, we investigate an online request scheduling framework for edge LLM inference that jointly minimizes long-term average end-to-end latency and regulates workload distribution across heterogeneous edge servers. Two main challenges arise in this context. First, conventional latency models cannot accurately capture the fine-grained dynamics of multi-stage LLM execution. Second, the latency consequence of a scheduling decision is observed only after request completion, making immediate decision evaluation difficult. To address these challenges, we develop a cross-slot inference model that captures transmission, prefill, iteration-level decoding, and key-value (KV) cache evolution for each diverse request, and characterize server workload through a KV cache memory-time consumption metric. We propose the LYREO approach that transforms the long-term load-balancing constraint via Lyapunov optimization and employs reward redistribution with sequencebased return prediction to convert delayed outcomes into timely learning signals for earlier decisions. Simulations under various configurations demonstrate that LYREO consistently achieves lower latency and more balanced load distribution than representative learning-based and heuristic baseline schemes.

cs.AI↗

arXiv2026

Magnetic Reconnection Heating of Thin Accretion Disks around Kerr Black Holes

The magnetic reconnection process near a black hole can efficiently convert magnetic energy into the kinetic and thermal energy of plasma outflows, enabling the extraction of rotational energy from a spinning black hole. However, its impact on the surrounding accretion disk remains insufficiently explored. In this work, we develop a framework to investigate the 'heating' effect of magnetic reconnection outflow plasma on thin accretion disks around Kerr black holes. We incorporate the energy and angular momentum fluxes of reconnection outflow into the disk conservation laws as additional source terms, yielding a modified radiative flux. We investigate the dependence of the modified radiative flux on the black hole spin, reconnection radius, plasma magnetization, and outflow orientation. For a fixed magnetic field prescription, the contribution of the reconnection outflow plasma to the modified radiative flux decreases as the black hole spin increases. Its dependence on the reconnection radius differs inside and outside the ergosphere: inside, a smaller radius enhances disk heating through more efficient extraction of black hole rotational energy, whereas outside, a larger radius produces stronger heating due to reduced radial dilution of the outflow. The reconnection outflow plasma contribution also increases with plasma magnetization and the radial component of the outflow velocity. Our results demonstrate that magnetic reconnection can provide an additional source of energy and angular momentum for relativistic accretion disks, establishing a connection between plasma processes near rotating black holes and the observable accretion disk properties, providing a new perspective on the strong gravitational field regime.

gr-qc↗

arXiv2026

AlayaVista: Streaming World Modeling from Panoramic States to Perspective Video

Interactive video world models must maintain broad scene context under camera motion while producing high-fidelity observations with low latency. Existing approaches face a representation trade-off: perspective models operate on local views and must preserve off-screen content over long rollouts, whereas broader spatial coverage is typically obtained by synthesizing full-sphere videos or constructing explicit 3D representations. Motivated by the complementary roles of global context and selective local acuity in visual perception, we present AlayaVista, a camera-controllable streaming video world model that decouples panoramic world evolution from perspective observation synthesis. Given a single perspective image, AlayaVista constructs a 360-degree scene prior using a pretrained panorama expansion model and then evolves the scene as a camera-conditioned panoramic latent state. A latent viewport renderer maps this state to the requested perspective video latents, while a perspective refiner restores details, suppresses artifacts, and performs super-resolution. To support efficient streaming, we adapt the panoramic generator to chunk-autoregressive generation and distill both panoramic generation and perspective refinement into few-step processes. To provide the supervision required by this design, we construct MUGEN, a large-scale real-world panoramic video dataset containing 1,318 hours of videos at resolutions of at least 4K, together with rich semantic and geometric annotations.

cs.CV↗

arXiv2026

Predicting energy and structural response to force correction in molecular dynamics

We predict how force correction changes energy exchange and structural statistics by measuring leading response coefficients on shared reference trajectories. Residual power and the displacement virial distinguish the transfer of energy from the change in restoring forces, including intermittent reference updates. Independent simulations then test the predicted kinetic and configurational shifts. A matched-timetable experiment shows that reference timing affects heating through its coupling to the evolving state. In an anharmonic chain, the displacement virial predicts a structural shift missed by a power-only description. Local force constants in silicon predict a complementary directional tradeoff: scalar calibration repairs optical motion while degrading an already accurate low-frequency direction. Full nonlinear trajectories confirm this tradeoff and distinguish the benefits of static curvature correction and repeated reference impulses. Independent finite-temperature integrals in orthorhombic tin selenide (SnSe) support the configurational-response direction predicted from separate reference calculations. These results provide a physical basis for choosing how reference information enters molecular dynamics. The framework assesses force correction through its effect on atomic motion and statistical observables, beyond the accuracy of individual force evaluations.

cond-mat.mtrl-sci↗

arXiv2026

ForgeTrain: Forging Production-Grade Training Frameworks via Harness-Driven AI Development

Training large models still relies on general-purpose frameworks such as Megatron-LM, whose generality tax constrains scenario-specific optimization and adds runtime overhead through accumulated abstraction. AI code generation reduces the cost of building a framework, and makes it affordable to forge one per scenario. We propose Forge Engineering: building a dedicated implementation from scratch for each scenario and iteratively optimizing it toward peak performance under correctness and usability constraints. Dedicated implementations inherit no abstraction boundaries, so they can integrate optimizations across the stack and reach a higher performance ceiling. We instantiate this paradigm for training frameworks as ForgeTrain, which holds a trusted framework as a golden reference and relaxes equivalence monotonically from Bit-for-Bit to Surpass. Experiments across multiple model--hardware configurations show that ForgeTrain consistently produces correct training engines and improves MFU over established training frameworks by 4.7--33.2%. To our knowledge this is the first production-grade training framework forged end-to-end by AI to match or surpass its human reference.

cs.SE↗

arXiv2026

Code-to-Harness: Distilling Black-Box Optimizers from Self-Play

Can an agent learn a numerical search strategy through executable practice and then transfer that strategy as text? We study low-budget black-box optimization, where unaided language models remain well below strong classical optimizers. During development, an agent repeatedly writes and evaluates optimizer programs. It then distills the resulting program and practice record once into a 197-word primary Harness A, which is frozen before evaluation. Harness A reduces Gemini Flash regret by 48\% in an independent $N=30$ study ($p<.001$), enters the GP-BO performance range on the practice family, and lowers mean regret on all three held-out BBOB landscapes. The same text improves every tested Gemini executor and transfers to Claude Sonnet, reducing regret by 43\% and 49\% ($p\leq.005$). An independent end-to-end replication produces Harness B, a different program and text at the same performance tier. The same framework also attains the lowest regret on a sealed YouTube reward-tuning production benchmark. Executable practice is thus a viable way to discover a search policy, and language a portable medium for deploying it.

cs.LG↗

arXiv2026

From State Synchronization to Cognitive Self-Evolution: An Operational Architecture for Cognitive Digital Twins

As Digital Twin (DT) systems evolve beyond state synchronization toward task-oriented and knowledge-driven operation, Cognitive Digital Twins (CDTs) have emerged as an extension that incorporates cognitive capabilities into twin operation. Existing CDT studies often focus on specific enabling techniques, such as learning modules, knowledge graphs, and large language models, while providing limited insight into how cognition can be systematically integrated into DT architectures. To address this issue, this paper proposes a four-layer CDT architecture consisting of the physical layer, digital-twin layer, cognitive layer, and task layer. The proposed architecture establishes a self-evolving closed operational loop spanning these four layers, in which physical states are synchronized into digital representations, cognition constructs task-specific cognitive models through knowledge, memory, and attention, and task-level decisions are generated under practical constraints. Operational feedback further refines cognitive experience and updates relationships and annotations in the digital representation, enabling subsequent task interpretation, initiation, and reasoning to evolve with system operation. Based on this framework, two representative operation modes are characterized: user-request-driven cognition and self-driven cognition. We further discuss key enabling mechanisms and deployment challenges associated with semantic communication, knowledge querying, task orchestration, and closed-loop synchronization. A lightweight simulation study illustrates reliable closed-loop task feasibility under limited semantic information and improved operational efficiency through accumulated task experience. The proposed framework provides a structured foundation for the design and development of future CDT systems.

cs.AI↗

arXiv2026

IDMate: Finite-temperature error bounds for window-resolved self-consistent-field screening

We formulate a finite-temperature residual test in IDMate that bounds window-resolved electronic errors without a spectral-gap assumption. For a fixed Hamiltonian and exact electron number, strong convexity of the matrix Fermi entropy bounds the density-matrix distance and free-energy error within a selected window. The bound remains finite at spectral crossings and extends to weighted k points with a shared chemical potential. We also derive a distance correction for particle-number mismatch. Across $3{,}586$ stress trials in $70$ seeded perturbation ladders, $1{,}942$ proposals satisfy the screen with no observed violation of the $0.05$ window-distance criterion plus its numerical allowance. This criterion differs from the uncorrected exact-trace bound, which six of ten historical in-loop candidates exceed at the numerical-error scale. An accept-or-recover loop replaces ten reference-map evaluations while meeting terminal comparison criteria in three configurations that include oracle-subspace controls. Additional candidates built only from preceding-iteration orbitals yield two acceptances and one abstention. The silicon candidate has a window distance of $1.891\times10^{-13}$ but a normalized real-space density error of $3.754\%$. Analytic examples separate errors from complement occupations and interblock coupling. Window-level accuracy therefore does not imply full-state accuracy; the screen tests compressed proposal quality, independently of nonlinear SCF convergence or net acceleration.

cond-mat.mtrl-sci↗

arXiv2026

The energetics of force errors in machine-learned molecular dynamics

The energetic effect of a force error depends on atomic motion. We establish a directional residual-work coefficient combining directional curvature mismatch with the spatial distribution of the residual response. For conservative potentials force-matched at an anchor, it determines the leading signed work at the first crossing of a small force-error budget. At a 474-atom lithium-electrolyte interface, predictions fixed before future reference evaluations differ from measurements by less than 4.6% of predicted work across 24 prescribed endpoints. Changing only the initial velocity direction at fixed structure and initial total kinetic energy reverses the force-work ranking. At the same admitted time of 0.25 fs, one direction gives an 11.6% larger maximum force residual but 36.1% less work. The reversal recurs at a second structure. The framework connects force tolerances to reference-energy transfer, providing a physical basis for potential assessment and adaptive reference allocation.

cond-mat.mtrl-sci↗

arXiv2026

MM-BrowseComp: A Comprehensive Benchmark for Multimodal Browsing Agents

AI agents with advanced reasoning and tool-use capabilities have demonstrated impressive performance in web browsing for deep search. However, existing benchmarks such as BrowseComp primarily focus on textual content, overlooking the prevalence of multimodal content. To bridge this gap, we introduce MM-BrowseComp, a novel benchmark comprising 400 challenging, hand-crafted questions designed to evaluate multimodal retrieval and reasoning capabilities. Unlike prior work, MM-BrowseComp incorporates visual prompts and necessitates the extraction of key evidence from web images and videos to complete questions, rendering text-only approaches insufficient. Additionally, we provide a verified checklist for each question, enabling fine-grained analysis of multimodal dependencies and reasoning paths. Our comprehensive evaluation of 27 state-of-the-art models reveals that even leading models like GPT-5-High with tools achieve only 24.25\% accuracy, highlighting the suboptimal multimodal browsing capabilities, establishing MM-BrowseComp as a rigorous new standard for the field.

cs.CL↗

arXiv2026

Exact branch-transfer criterion for common-mode Thomson heat cancellation in thermoelectric couples

Thermoelectric p- and n-type legs are commonly paired by matching their Seebeck magnitudes, although a cooler responds to heat transported through its complete electrical and thermal network. We decompose the leg coefficients into differential thermopower $α=S_p-S_n$ and common thermopower $M=(S_p+S_n)/2$. In a connected steady-state scalar thermoelectric network, a temperature-independent co-shift applied to every electrically active segment is an exact terminal null. A temperature-dependent perturbation of the legs relative to fixed leads is instead physical. At fixed current and shared isothermal endpoints, its first-order cold-port response is the action of $Γ_m=T\,dm/dT$ on the difference between the p- and n-branch oriented collection measures. We prove that every continuous $Γ_m$ cancels if and only if these measures are equal. In the constant-property, linear-common-mode limit, matching $R_i/K_i^{\rm leg}$ is sufficient and does not require identical legs. One- and two-dimensional calculations confirm the analytic reductions within their stated domains. For split thermal pads, the analysis gives the exact array law $ΔQ_{c,Σ}=\sum_j C_jI_jΔT_{c,j}$ and, for series elements with isothermal hot pairs, $IΔV_Σ=-ΔQ_{c,Σ}$. A representative seven-pair model gives corresponding increments of 7.87 mW and $-2.80$ mV. Branch transfer and endpoint topology therefore provide distinct material-pairing and device-test criteria for common-mode Thomson heat.

cond-mat.mtrl-sci↗

arXiv2026

Interactive Inference: A Neuromorphic Theory of Human-Computer Interaction

Neuromorphic Human-Computer Interaction (HCI) is a theoretical approach to designing better user experiences (UX) motivated by advances in the understanding of the neurophysiology of the brain. Inspired by the neuroscientific theory of Active Inference, Interactive Inference is a first example of such an approach. It offers a simplified interpretation of Active Inference that allows designers to more readily apply this theory to design and evaluation. The basic premise in Interactive Inference is that the user predicts a result prior to performing a task. User behaviour is modeled as Bayesian inference on progress and goal distributions that predicts the next action. The difference between the observed result and the prediction is what is processed by the brain. This discrepancy between goal and progress distributions, or Bayesian surprise, can be modeled as a simple mean square error of the signal-to-noise ratio (SNR) of a task. The problem is that the user's capacity to process Bayesian surprise follows the logarithm of this SNR. This means terminal execution errors rise quickly once average capacity is exceeded. Our model allows the quantitative analysis of performance and terminal error using one framework that can provide real-time estimates of the mental load of users that needs to be minimized by design. We show how three basic laws of HCI, Hick's Law, Fitts' Law and the Power Law can be expressed using our model. We then test the validity of the model by empirically measuring how well it predicts human performance and terminal error in a car following task. Results suggest that driver processing capacity is indeed a logarithmic function of the SNR of the distance to a lead car, and that terminal error follows the residual surprise predicted by a KL-divergence model. This result provides initial evidence that Interactive Inference can be useful as a new theoretical design tool.

cs.HC↗