Searcharxiv⌕ Search

arXiv subjects

Yan Li

Publications and source records attributed to Yan Li.

At least 19 recordsLinked to original sources
arXiv2026

SenseNova-U1.5: Towards Native Unified Visual Intelligence

We launch SenseNova-U1.5, an 8B-MoT native unified multimodal model that understands, reasons about, and generates visual content within an encoder-free and VAE-free architecture. We strengthen its visual interface through spatially coherent patch reconstruction and scale its training with carefully curated generation and editing data, improved task formulation, structural prompt enhancement, and native resolutions of up to 4K. For post-training, we optimize specialized experts for visual aesthetics, bilingual text rendering, infographic generation, and image editing, and consolidate their capabilities through multi-expert on-policy distillation. Across extensive evaluations, SenseNova-U1.5 largely advances image fidelity, text rendering, complex composition, multi-reference editing, and interleaved generation, while improving instruction following and preserving subject identity, geometry, and unmodified regions. Despite limited exposure to structured formats in its generation data, SenseNova-U1.5 generalizes effectively to long, complex, and structured visual instructions, further proving that multimodal understanding can transfer to visual planning and creation. Together, these findings position native unified modelling as a promising path towards systems that perceive, reason and create within a fully end-to-end framework. We will open-source training code, including supervised fine-tuning, reinforcement learning, and on-policy distillation.

cs.CV↗
arXiv2026

UNISON: A Co-Designed Near-Memory Scheduler of Session KV Residency for LLM Agents

Large language models are increasingly composed into agent loops that plan, call tools, and resume the same task after each action. These loops press a shared memory hierarchy harder than conventional multi-turn chat, because they hold a growing key-value (KV) prefix across tool waits and place many sessions on one SRAM/HBM pool, so that eviction and hierarchical placement become a session-level efficiency problem orthogonal to compute-mode optimization. Existing proxies based on recency, timeout, or identity miss the mechanism information of the loop and therefore treat a live wait as a cold, discardable unit. We present Unified Native Inter-turn Session Orchestration Nexus (UNISON), an event-driven near-memory scheduler in which Survival-Penalty Eviction for Agent Return-gap (SPEAR) and Tiering in Idle-window DMA Events (TIDE) share one live ranking. SPEAR selects who leaves from a gap average and a turn-indexed hazard, while TIDE spends the observed wait as a DMA budget for who sits in the fast tier. On coding and general-mission benchmarks with three model families, totaling 1,415 sessions and 33,596 turns, the joint policy is the best non-oracle entry on every trace, raising hit rate by 0.3% to 23.1%, reducing AMAT by 22% to 51%, and lowering TTFT by 58% to 89% on long-horizon traces. A structural necessity analysis shows that the unified near-memory design cannot be decomposed into independent IPs or realized in software without re-introducing documented failure modes. The 28-nm CMOS scheduling core occupies 0.169 mm^2 at 13.6 mW and 150 MHz, a negligible overhead relative to the KV hierarchy it manages, reproducing the floating-point ranking at Kendall tau exceeding 0.998.

cs.AR↗
arXiv2026

ReDeck: Step-Level Render-Grounded Refinement for Document-to-Slide Generation

Document-to-slide generation is challenging because slides are dense editable artifacts that require both faithful content selection and precise spatial layout. Recent slide agents adopt iterative reflection, but typically follow a monolithic "one version, one feedback" loop: a slide or deck is rewritten, rendered afterward, and critiqued only at the turn boundary. This delayed feedback makes local failures such as overflow, overlap, clipping, and off-canvas placement difficult to attribute and repair. We propose ReDeck, a step-level render-grounded refinement framework that decomposes slide revision into atomic edit actions and returns renderer-derived observations after each step, turning refinement into "one edit, one observation." To balance local repair with global quality, ReDeck uses multi-granular feedback: step-level render feedback for spatial errors, a turn-level adaptive critic for semantic and design guidance, and a submission-level gate for hard layout validation. We further introduce DeckQuiz, a benchmark that decouples content fidelity, spatial correctness, and design quality. Across GPT-5.4, Claude-4.6, and Gemini-3.1, ReDeck consistently outperforms existing slide-generation agents, and ablations confirm that feedback timing and granularity are critical for reliable slide refinement.

cs.AI↗
arXiv2026

Probing Triton's Space Environment and Internal Structure: An Integrated Detection-and-Interpretation Framework

Triton, Neptune's largest moon, is a prime ocean-world target. Constraining ocean thickness, composition, and conductivity is essential for habitability assessment, but magnetic induction alone cannot resolve the thickness-conductivity degeneracy, and magnetic perturbations from Triton's space currents can obscure the internal induction signal. We present an integrated detection-and-interpretation concept linking four physically consistent calculations. Using `PlanetProfile', we construct a common radial interior structure (temperature, density, conductivity, seismic-wave speed). We then use `MoonMag' to compute the degree-one magnetic-induction response from that conductivity profile at the synodic, rotational, and orbital periods. We perform a multi-fluid `SWMF' simulation with the induced dipole as the inner-boundary condition and develop a Coulomb-gauge Poisson reconstruction to isolate space-current magnetic fields. Finally, we develop the `TritonSeis' workflow, three-dimensional seismic forward modeling plus hierarchical travel-time inversion, to constrain the ice-ocean and ocean-rock interface depths. We find that induction is substantially more sensitive to ocean conductivity than to layer thickness, and that space-current fields are comparable in amplitude to the internal induction signal. A five-station synthetic recovery test resolves both interfaces to first order, with errors of +8.4% for the ice shell and -12.5% for the ocean. Under a conservative noise assumption, the minimum detectable magnitudes are approximately 3.8-4.6 at epicentral distances of 100-1000 km. The Poisson reconstruction and end-to-end seismic recovery are, to our knowledge, the first such quantitative demonstrations for Triton. Coordinated magnetic, plasma, and seismic measurements are complementary and can break the conductivity-thickness degeneracy, providing a framework for future Triton exploration.

astro-ph.EP↗
arXiv2026

Initialization Is Critical: Advancing Federated Short-Term Load Forecasting under Load Heterogeneity via Model Initialization

Short-term load forecasting (STLF) provides essential information for numerous applications in modern power systems. However, accurate STLF often relies on fine-grained smart-meter data from distributed users, raising increasing concerns about data privacy. Federated learning (FL) has therefore emerged as a promising privacy-preserving paradigm for STLF. Nevertheless, this paper reveals structured heterogeneity in clients' load data. Specifically, clients exhibit different responses to exogenous factors and distinct temporal load profiles, which can degrade forecasting performance in FL. To mitigate these issues, this paper studies the role of model initialization in federated STLF, and proposes two initialization strategies from global and local perspectives. For global model initialization, when auxiliary public load data are available, a pretrained initialization strategy is developed to initialize the global model before federated training, thereby reducing client drift during the training process. For local model initialization, we propose SLIAvg, a sequential local initialization strategy that promotes a more consistent training process by allowing participating clients to start from progressively adapted models within each communication round. Since the proposed strategies only modify the initialization process, they are compatible with most existing FL frameworks and privacy-enhancing techniques. Experiments on real smart-meter data with two representative forecasting architectures demonstrate that the proposed strategies effectively improve forecasting performance, as evidenced by reduced client drift, improved convergence behavior, and lower forecasting errors.

cs.LG↗
arXiv2026

TrapVLA: Trapping Vision-Language-Action Models in Configured Failure Modes

This work introduces Configured Failure Trapping, a novel backdoor attack task against Vision-Language-Action (VLA) models, which aims to activate attacks through stealthy textual triggers and induce configured failure modes. Unlike prior backdoor attacks that treat any task failure as a successful attack, Configured Failure Trapping requires the attacker to control how the robot fails (e.g., causing the robot to grasp with a specified positional offset), making it substantially more challenging and hard to detect. To support the new task, we propose an effective data engine for synthesizing high-quality target trajectories and an automated suite for measuring configured-failure fidelity. Then, based on this foundation, we construct two new benchmarks, namely Trap-LIBERO and Trap-RoboTwin, that instantiate Configured Failure Trapping across four representative failure modes. To address this task, we identify sparse action deviation as a critical challenge and accordingly propose a novel method named TrapVLA, which explicitly learns trigger-induced action residuals to steer the policy toward the configured failure behavior. Extensive experiments across simulation benchmarks and real-world robotic settings show that TrapVLA effectively injects configured failure modes into VLA models while largely preserving performance on clean data. Project page: https://john-liua.github.io/TrapVLA/

cs.RO↗
arXiv2026

Closed-loop AI achieves certifiable engineering design

Agentic AI has automated parts of scientific discovery, including paper generation, expert-level coding, therapeutic proposal, and autonomous experimentation. Complex physical engineering design remains a gap, because candidates must satisfy simultaneous constraints in fluid dynamics, solid mechanics, and structural stability. We introduce The AI Engineer, an agentic framework that couples large language models (LLMs) to deterministic engineering backends in a closed loop: natural-language requirements are converted into design-domain geometry and mesh; topology is optimized with bi-directional evolutionary structural optimization (BESO) coupled to the CalculiX solver; and member sizes are refined with particle swarm optimization (PSO) coupled to Zwind under offshore aero-hydro-servo-elastic load cases. To explore many designs without per-candidate certification cost, an Automated Reviewer scores each candidate on five dimensions (capacity, steel intensity, unit cost, constructability, and fatigue life) using piecewise-linear functions calibrated on 11 real floating-wind projects. Search terminates only when a candidate reaches a composite score $S \ge 85$ (grade A) with no subscore below 60. We validated this gate by submitting the top-scoring design to the China Classification Society (CCS) for Approval in Principle (AIP), which it passed; AIP is thus an external check that the reviewer tracks professional judgment, not the daily objective. The certified design outperforms the human-optimized TuQiang baseline, reducing steel mass and unit capital cost by 8.1% each while meeting all AIP criteria. This verification-closed regime, in which every proposal is judged by deterministic physics and codified limit states, distinguishes The AI Engineer from open-ended generative systems. Remaining limits include detailed design and fabrication-hard constraints.

cs.AI↗
arXiv2026

Complete Rigidity at infinity and Existence of the Levinson Cavity

We present a potential theoretic approach reducing the analysis of the asymptotic shape of free surfaces to the analysis of a precise ordinary differential equation resulting from the reduction process. Although the approach relies mainly on the principal part of the PDE operator to allow for a representation formula and is thus not restricted to problems of elliptic type, we present it at the clean-cut example of three-dimensional axially symmetric steady incompressible cavity flows, which are Neumann-type Bernoulli free boundary problems and for which frequency formulas are unknown and, if they do exist, insufficient to yield the very precise asymptotic behavior we prove here. In 1946 Norman Levinson derived by a power-law ansatz with a slowly varying correction a precise formula for the asymptotic shape of such cavities. However his result requires very strong assumptions such that it has remained an open problem for 80 years whether the cavity solutions we know to exist by a result by Garabedian-Lewy-Schiffer [12] actually share this asymptotic behavior, or whether at least one solution possessing the Levinson asymptotics exists. Here we answer both questions affirmatively, and we obtain complete rigidity at infinity of the Levinson solution in the class of axially symmetric solutions, that is, any solution satisfying mild and natural assumptions at the fixed boundary and infinity converges asymptotically to the Levinson profile $(\log r)^{-1/4}\sqrt{r}$.

math.AP↗
arXiv2026

CoinVE-200K: A Large-Scale High-Quality Dataset for Compositional Instruction-Guided Video Editing

The quality and diversity of instruction-based video editing datasets are steadily improving, yet existing datasets mainly focus on single editing operations and fall short in supporting compositional instruction-guided video editing. In particular, multiple editing intents must be jointly understood and faithfully executed within the same video. To address this issue, we introduce CoinVE-200K, a large-scale, high-quality dataset for Compositional Instruction-Guided Video Editing. CoinVE-200K contains 1080p video-editing pairs of up to 201 frames, covering diverse compositional scenarios where each sample involves 2 to 5 atomic editing operations. The instructions target humans, objects, and backgrounds, and cover edit types such as addition, removal, modification, and stylization. All samples are built through a carefully designed generation and filtering pipeline to ensure instruction faithfulness, visual quality, temporal consistency, and compositional diversity. We also introduce CoinVE-Bench, a benchmark for compositional-instruction video editing across diverse subjects, operation types, and instruction complexities. Furthermore, we present CoinVE-Edit, a 22B compositional video editing model built upon Wan2.1-T2V-14B and Qwen3-VL-8B-Instruct. CoinVE-Edit disentangles region-aware attention for different editing instructions, enabling precise multi-region editing while preserving irrelevant content and temporal coherence. Experiments on CoinVE-Bench show that CoinVE-Edit achieves strong performance in instruction following, compositional editing accuracy, visual quality, and temporal consistency.

cs.CV↗
arXiv2026

SPLIT-Q: A Scalable Sequential Quantum Computing Framework for Coherent Controlled Islanding

Growing integration of distributed energy resources increases power-system variability and uncertainty. During disturbances, these effects can intensify generation-load imbalances and cascading failures. Controlled islanding limits their propagation by partitioning a compromised grid into connected, electrically sustainable islands. However, classical methods face rapidly growing computational costs as network size and island count increase. Quantum optimization offers an alternative for exploring this combinatorial partition space. Yet monolithic quantum formulations encode all assignment decisions in one circuit, causing qubit demand and circuit complexity to scale with network size. In this study, a qubit-bounded sequential distributed quantum approximate optimization algorithm (QAOA) framework is proposed to tackle coherent controlled islanding under limited quantum resources. It formulates the optimization as boundary-conditioned regional quadratic unconstrained binary optimization (QUBO) subproblems that are solved sequentially within a fixed qubit budget. Thus, circuit width remains independent of network size, with aggregate quantum workload scaling linearly on bounded-degree networks. Evaluation covers eleven IEEE systems from 9 to 300 buses using IBM quantum computing resources, with Gurobi and monolithic QAOA as references. Across all systems, the framework recovers feasible Gurobi-optimal partitions under noise, confirming the resilience of its solution quality. The results further show that the proposed method substantially reduces quantum-resource demand and circuit complexity relative to monolithic QAOA, allowing large islanding problems to be addressed within current hardware limits. The proposed framework provides a feasible and scalable pathway for quantum optimization in large-scale power systems.

quant-ph↗
arXiv2026

Adaptable Fingerprinting with Nonlinear Shrinkage for Climate Change Detection and Attribution under Variance Heterogeneity

Detection and attribution of climate change relies on fingerprinting--a linear errors-in-variables regression framework in which both predictors and responses exhibit internal variability governed by a proportional covariance structure, subject to a variability inflation factor. Accurate estimation of the scaling factors (regression coefficients) depends on inferring the precision matrix of the regression errors from limited climate model control runs. In high-dimensional settings, existing approaches often overlook the variance inflation of the predictors and suffer from imprecise precision matrix estimates, yielding biased estimators, underestimated uncertainties, and confidence intervals with poor coverage. We propose a nonlinear, rotation-invariant shrinkage framework for estimating the precision matrix that restores the asymptotic optimality of the total least squares estimator in high-dimensional regimes. Our procedure jointly estimates the scaling factors and the variability inflation factor, thereby correcting estimation bias, and incorporates consistent variance estimators to enable valid uncertainty quantification. We also develop a residual consistency test to assess model adequacy. Numerical studies demonstrate precise estimation, improved confidence interval coverage, and higher efficiency. Applied to annual mean near-surface air temperature data from 1951--2020, our method produces narrower and more reliable confidence intervals, yielding refined attribution results.

stat.ME↗
arXiv2026

Theoretical analysis towards accurate optomechanical detection of quantum gravity effects

Optomechanical systems offer a promising platform for observing dynamical signatures of quantum gravity through precision measurements of quantum harmonic oscillator dynamics. However, most existing analyses consider only the linear radiation-pressure interaction while neglecting higher-order optomechanical couplings and laser phase noise. These neglected contributions can be comparable in magnitude to the predicted quantum-gravity corrections and may therefore introduce spurious signals or mask the genuine physical effect. Here we reanalyze two experimentally realized platforms, a Fabry-Perot optomechanical system and a membrane-in-the-middle optomechanical system, by incorporating the complete nonlinear dynamics and realistic laser phase noise. Using measured device parameters, we derive revised protocols for generalized uncertainty principle tests and establish practical sensitivity bounds. Our results demonstrate that previous idealized estimates significantly overestimate the achievable resolution, underscoring the necessity of including higher-order interactions and implementing effective laser phase noise suppression in realistic assessments of optomechanical quantum gravity tests.

quant-ph↗
arXiv2026

OccAnyScene: Towards Unified Indoor-Outdoor 3D Occupancy Prediction

3D occupancy prediction is fundamental to scene understanding, yet existing 3D semantic occupancy methods are typically specialized to fixed scene types and occupancy protocols. We introduce Cross-Scene 3D Semantic Occupancy Prediction, a new task setting which requires a single model to handle heterogeneous indoor and outdoor scenes with varying cameras, spatial ranges, voxel specifications, and semantic taxonomies. This setting poses a fundamental challenge: achieving metric-consistent yet scene-adaptive image-to-3D lifting across varying camera configurations and scene scales. To address this challenge, we propose OccAnyScene, a pixel-frustum-centered Gaussian framework built upon a pretrained depth foundation model. Specifically, the framework employs Pixel-Aligned Frustum Feature Aggregation to construct a camera-aware frustum query for each feature pixel, and Frustum-Parameterized Gaussian Construction to decode each query into multiple Gaussians whose positions and sizes are constrained by the predicted pixel depth and corresponding frustum geometry. OccAnyScene sets new state-of-the-art results, achieving 59.92% mIoU on the indoor Occ-ScanNet and 23.06% mIoU on the outdoor SurroundOcc-nuScenes.

cs.CV↗
arXiv2026

AgentPatch: Coarse-to-Fine Weak-Task Repair for Merging Agentic Multimodal Large Language Models

Agentic multimodal large language models (MLLMs) extend multimodal perception and reasoning with planning, tool use, and interaction in dynamic environments. Yet current models are specialized for particular tools or environments, complicating consolidation into a single generalist. We formulate Agentic MLLM Merging and identify two challenges: asymmetric capability preservation, whereby capabilities with different interaction complexity are retained unevenly, producing weak tasks after merging, and behavior-critical forgetting, whereby losing decisive actions can derail long-horizon execution. We propose AgentPatch, a training-free coarse-to-fine repair framework. It selects a stable merged backbone, restores diluted weak-task-specific signals through Weak-Task Unique Residual Recovery, and applies an Agent-Guided Behavior-Critical Patch that recovers decisive behaviors under explicit capability protection. AgentPatch produces a single static checkpoint without routing or ensembles. Experiments across six agentic and multimodal benchmarks show that AgentPatch improves diverse merged backbones, alleviates weak-task degradation, and better balances weak-task recovery with the preservation of complementary search and agentic visual processing capabilities. Code is available at https://github.com/ziboshao/AgentPatch.

cs.AI↗
arXiv2026

Pulse-Duration Control of Subcycle Multiband Electron Dynamics Extends the High-Harmonic Cutoff in a Light-Driven Insulator

We demonstrate pathway-selective control of extreme-ultraviolet high-harmonic generation by jointly tuning laser pulse duration ($5$ - $29$ fs) and intensity ($0.8$ - $74$ TW/cm$^2$). Many-cycle pulses at moderate intensities, $\sim 6$ TW/cm$^2$, promote cumulative carrier transfer over successive optical cycles, progressively accessing higher conduction bands. In contrast, few-cycle, high-intensity, $\sim 22$ TW/cm$^2$, pulses drive subcycle multiband dynamics that reach $25$ - $50$ eV photon energies before decoherence can suppress coherent emission. These results reveal pulse duration and intensity as decisive control knobs for high-harmonic emission, opening a route to band-structure-guided pulse design for higher energy extreme-ultraviolet light sources.

physics.optics↗
arXiv2026

Recompute or Reuse? Diagnosing and Mitigating Textual Shortcuts in VLM Self-Reflection

Vision-language models (VLMs) are expected to revise their reasoning when visual evidence changes. Failures to do so are often attributed to insufficient visual attention or contextual inertia, leaving unclear what models reuse instead of recomputing from the current image. We show that evidence-bearing reasoning in a prior chain of thought (CoT) can form a textual shortcut that competes behaviorally with visual recomputation. Across 16 VLMs, a matched counterfactual analysis identifies evidence-bearing content as the most robust carrier of prior-CoT influence. Removing this evidence-bearing content shifts answer preference more than removing length-matched non-evidence context or the final-answer span, with prior control weakening progressively as more stale evidence is removed. Reordering this evidence also weakens prior control, showing that its organization modulates shortcut strength. Beyond the immediate answer, the shortcut can retain residual influence after answer correction: weakening current-image support shifts preference back toward the prior answer, while repeated prior answers and reused premises arise mainly when the shortcut remains active. To limit this influence, we introduce Fresh-State Attention Firewall (FSAF), a training-free intervention that isolates fresh computation from the prior CoT. Across five VLMs, FSAF raises visual update rate from 35.28% to 53.61% and reduces prior-answer rate from 39.22% to 3.67%. Reliable VLM self-reflection therefore requires more than looking again: fresh visual recomputation must be protected from stale textual reuse.

cs.CV↗
arXiv2026

PACE-QAOA: Physics-Constrained Quantum Optimization for Qubit-Efficient Power System Islanding

Increasing renewable-energy penetration heightens power-system variability and complicates disturbance containment. Controlled islanding mitigates cascading failures by partitioning a stressed network to limit disrupted power transfer while preserving each island's operational integrity, but this constrained partitioning problem is NP-hard. Although QAOA offers a complementary search strategy, limited near-term qubit capacity restricts conventional formulations. This paper presents a qubit-efficient hybrid quantum framework combining a physics-informed compact encoding with Lagrangian constraint handling and classical feasibility refinement. The encoding exploits grid structure while formally preserving the original feasible solution space and objective. For a fixed island count on sparse working graphs, the formulation reduces phase-separator and per-layer gate complexity from quadratic to linear scaling with system size. Tests on eight IEEE systems ranging from 9 to 89 buses and multiple quantum-provider backends produce feasible, high-quality islanding solutions under practical circuit and sampling budgets. Factorial ablation attributes resource and runtime improvements to the complementary effects of compact encoding and qubit-efficient constraint handling. Noise analysis shows stable solution quality under tested device noise, while landscape diagnostics reveal smoother, more consistently scaled QAOA cost surfaces and improved parameter-optimization behavior. These results offer a transferable approach for scaling constrained quantum optimization toward larger real-world applications on near-term hardware.

quant-ph↗
arXiv2026

A Provable Oracle-Free Quantum Algorithm for Nonlinear Dynamics on Hybrid Oscillator-Qubit Processors

We develop a hybrid qubit--qumode algorithm for nonlinear ordinary differential equations of the form $\dot{\mathbf{x}}=\mathbf{f}(\mathbf{x})$ with drift of polynomial degree~$L$. Following the Fokker--Planck route of Tennie and Magri, the algorithm propagates the state density and returns the deterministic trajectory as the peak of that density in the small-noise limit. The discretised generator is carried into a parametrised family of Schr\"{o}dinger equations by the warped-phase transformation of Jin, Liu, and Yu, and the Fourier-mode parameter of that family is placed on a single continuous-variable qumode. Our central structural result is that the Hermitian parts $H_{1}$ and $H_{2}$ of the discretised generator admit a bipartite Pauli decomposition that sorts the non-zero Pauli strings into $\mathcal{O}(\log N)$ mutually commuting families and factorises each family into a diagonal of degree at most $L$ tensored with a fixed rank-two bond operator. The factorisation renders each family exponential an exact product of $\mathcal{O}(n^{L})$ monomial-controlled momentum displacements, with no intra-family Trotter error. On a $d$-dimensional grid of $N=2^{n}$ points per axis the circuit costs $\mathcal{O}(d^{L+1}n^{L+2})$ gates per Trotter step. No sparse-access oracle and no block encoding is invoked: every gate is fixed in closed form by the polynomial coefficients of the drift. We also prove a bound on the numerical abscissa $\lambda_{\max}(H_{1})$ that fixes the recovery domain of the warped-phase transform and the post-selection cost. A classical simulation on two nonlinear benchmarks confirms the structural theorems, the shifted recovery, and the accuracy-per-resource advantage of the continuous-variable coupling over a discretised mode register.

quant-ph↗