Searcharxiv⌕ Search

arXiv subjects

Hao Zhang

Publications and source records attributed to Hao Zhang.

At least 253 records · Page 14Linked to original sources
arXiv2026

ConsRoute:Consistency-Aware Adaptive Query Routing for Cloud-Edge-Device Large Language Models

Large language models (LLMs) deliver impressive capabilities but incur substantial inference latency and cost, which hinders their deployment in latency-sensitive and resource-constrained scenarios. Cloud-edge-device collaborative inference has emerged as a promising paradigm by dynamically routing queries to models of different capacities across tiers. In this paper, we propose ConsRoute, a lightweight, semantic-aware, and adaptive routing framework that significantly improves inference efficiency while minimizing impact on response quality. Unlike prior routing methods that rely on predicting coarse-grained output quality gaps, ConsRoute leverages a reranker to directly assess the semantic consistency between responses generated by models at different tiers, yielding fine-grained soft supervision signals for routing. To minimize device-side overhead, ConsRoute reuses hidden states from the LLM prefilling stage as compact query representations, avoiding additional encoders or inference passes. Furthermore, these representations are clustered, and Bayesian optimization is employed to learn cluster-specific routing thresholds that dynamically balance quality, latency, and cost under heterogeneous query distributions. Extensive experiments demonstrate that ConsRoute achieves near-cloud performance (>=95%) while reducing end-to-end latency and inference cost by nearly 40%, consistently outperforming existing routing baselines in both response quality and system efficiency.

cs.AI↗
arXiv2026

Strategic Navigation or Stochastic Search? How Agents and Humans Reason Over Document Collections

Multimodal agents offer a promising path to automating complex document-intensive workflows. Yet, a critical question remains: do these agents demonstrate genuine strategic reasoning, or merely stochastic trial-and-error search? To address this, we introduce MADQA, a benchmark of 2,250 human-authored questions grounded in 800 heterogeneous PDF documents. Guided by Classical Test Theory, we design it to maximize discriminative power across varying levels of agentic abilities. To evaluate agentic behaviour, we introduce a novel evaluation protocol measuring the accuracy-effort trade-off. Using this framework, we show that while the best agents can match human searchers in raw accuracy, they succeed on largely different questions and rely on brute-force search to compensate for weak strategic planning. They fail to close the nearly 20% gap to oracle performance, persisting in unproductive loops. We release the dataset and evaluation harness to help facilitate the transition from brute-force retrieval to calibrated, efficient reasoning.

cs.CL↗
arXiv2026

Programmable spectral phase transfer to the ultraviolet by gas-filled-fibre four-wave mixing

Programmable shaping of femtosecond ultraviolet (UV) pulses is still much less flexible than at visible and near-infrared wavelengths, mainly because direct UV modulators remain limited in bandwidth, throughput and damage threshold. Here we show that dispersive four-wave mixing (DFWM) in a gas-filled hollow-cappillary fibre (HCF) can transfer programmed spectral phase from the near infrared (NIR) to the UV without relying a narrowband pump. A shaped NIR signal at 1032 nm and a chirped 516-nm pump generate a 344-nm idler, which is characterized with transient-grating frequency-resolved optical gating (TG FROG). As a benchmark, second-order dispersion (SOD) applied to the signal is quantitatively reproduced in the idler. We then demonstrate the transfer of two nontrivial phase patterns: a localized nominal π-step and a moderate sinusoidal modulation. In the π-step case, a step imposed on the long-wavelength side of the signal appears on the short-wavelength side of the idler, consistent with the 2*pump - 1*signal mixing relation. In the sinusoidal case, the periodic phase produces a split temporal waveform in both signal and idler. These results show that gas-filled HCF DFWM can act as a practical spectral-phase transducer from the NIR to the UV, while also revealing a trade-off between conversion efficiency and phase-transfer fidelity.

physics.optics↗
arXiv2026

ProRL Agent: Rollout-as-a-Service for RL Training of Multi-Turn LLM Agents

Multi-turn LLM agents are increasingly important for solving complex, interactive tasks, and reinforcement learning (RL) is a key ingredient for improving their long-horizon behavior. However, RL training requires generating large numbers of sandboxed rollout trajectories, and existing infrastructures often couple rollout orchestration with the training loop, making systems hard to migrate and maintain. Under the rollout-as-a-service philosophy, we present ProRL Agent , a scalable infrastructure that serves the full agentic rollout lifecycle through an API service. ProRL Agent also provides standardized and extensible sandbox environments that support diverse agentic tasks in rootless HPC settings. We validate ProRL Agent through RL training on software engineering, math, STEM, and coding tasks. ProRL Agent is open-sourced and integrated as part of NVIDIA NeMo Gym.

cs.AI↗
arXiv2026

SkyReels-V4: Multi-modal Video-Audio Generation, Inpainting and Editing model

SkyReels V4 is a unified multi modal video foundation model for joint video audio generation, inpainting, and editing. The model adopts a dual stream Multimodal Diffusion Transformer (MMDiT) architecture, where one branch synthesizes video and the other generates temporally aligned audio, while sharing a powerful text encoder based on the Multimodal Large Language Models (MLLM). SkyReels V4 accepts rich multi modal instructions, including text, images, video clips, masks, and audio references. By combining the MLLMs multi modal instruction following capability with in context learning in the video branch MMDiT, the model can inject fine grained visual guidance under complex conditioning, while the audio branch MMDiT simultaneously leverages audio references to guide sound generation. On the video side, we adopt a channel concatenation formulation that unifies a wide range of inpainting style tasks, such as image to video, video extension, and video editing under a single interface, and naturally extends to vision referenced inpainting and editing via multi modal prompts. SkyReels V4 supports up to 1080p resolution, 32 FPS, and 15 second duration, enabling high fidelity, multi shot, cinema level video generation with synchronized audio. To make such high resolution, long-duration generation computationally feasible, we introduce an efficiency strategy: Joint generation of low resolution full sequences and high-resolution keyframes, followed by dedicated super-resolution and frame interpolation models. To our knowledge, SkyReels V4 is the first video foundation model that simultaneously supports multi-modal input, joint video audio generation, and a unified treatment of generation, inpainting, and editing, while maintaining strong efficiency and quality at cinematic resolutions and durations.

cs.CV↗
arXiv2026

SPRITE: From Static Mockups to Engine-Ready Game UI

Game UI implementation requires translating stylized mockups into interactive engine entities. However, current "Screenshot-to-Code" tools often struggle with the irregular geometries and deep visual hierarchies typical of game interfaces. To bridge this gap, we introduce SPRITE, a pipeline that transforms static screenshots into editable engine assets. By integrating Vision-Language Models (VLMs) with a structured YAML intermediate representation, SPRITE explicitly captures complex container relationships and non-rectangular layouts. We evaluated SPRITE against a curated Game UI benchmark and conducted expert reviews with professional developers to assess reconstruction fidelity and prototyping efficiency. Our findings demonstrate that SPRITE streamlines development by automating tedious coding and resolving complex nesting. By facilitating rapid in-engine iteration, SPRITE effectively blurs the boundaries between artistic design and technical implementation in game development. Project page: https://baiyunshu.github.io/sprite.github.io/

cs.HC↗
arXiv2026

Internalizing Agency from Reflective Experience

Large language models are increasingly deployed as autonomous agents that must plan, act, and recover from mistakes through long-horizon interaction with environments that provide rich feedback. However, prevailing outcome-driven post-training methods (e.g., RL with verifiable rewards) primarily optimize final success signals, leaving rich environment feedback underutilized. Consequently, they often lead to distribution sharpening: the policy becomes better at reproducing a narrow set of already-successful behaviors, while failing to improve the feedback-grounded agency needed to expand problem-solving capacity (e.g., Pass@k) in long-horizon settings. To address this, we propose LEAFE (Learning Feedback-Grounded Agency from Reflective Experience), a framework that internalizes recovery agency from reflective experience. Specifically, during exploration, the agent summarizes environment feedback into actionable experience, backtracks to earlier decision points, and explores alternative branches with revised actions. We then distill these experience-guided corrections into the model through supervised fine-tuning, enabling the policy to recover more effectively in future interactions. Across a diverse set of interactive coding and agentic tasks under fixed interaction budgets, LEAFE consistently improves Pass@1 over the base model and achieves higher Pass@k than outcome-driven baselines (GRPO) and experience-based methods such as Early Experience, with gains of up to 14% on Pass@128.

cs.AI↗
arXiv2026

Covo-Audio Technical Report

In this work, we present Covo-Audio, a 7B-parameter end-to-end LALM that directly processes continuous audio inputs and generates audio outputs within a single unified architecture. Through large-scale curated pretraining and targeted post-training, Covo-Audio achieves state-of-the-art or competitive performance among models of comparable scale across a broad spectrum of tasks, including speech-text modeling, spoken dialogue, speech understanding, audio understanding, and full-duplex voice interaction. Extensive evaluations demonstrate that the pretrained foundation model exhibits strong speech-text comprehension and semantic reasoning capabilities on multiple benchmarks, outperforming representative open-source models of comparable scale. Furthermore, Covo-Audio-Chat, the dialogue-oriented variant, demonstrates strong spoken conversational abilities, including understanding, contextual reasoning, instruction following, and generating contextually appropriate and empathetic responses, validating its applicability to real-world conversational assistant scenarios. Covo-Audio-Chat-FD, the evolved full-duplex model, achieves substantially superior performance on both spoken dialogue capabilities and full-duplex interaction behaviors, demonstrating its competence in practical robustness. To mitigate the high cost of deploying end-to-end LALMs for natural conversational systems, we propose an intelligence-speaker decoupling strategy that separates dialogue intelligence from voice rendering, enabling flexible voice customization with minimal text-to-speech (TTS) data while preserving dialogue performance. Overall, our results highlight the strong potential of 7B-scale models to integrate sophisticated audio intelligence with high-level semantic reasoning, and suggest a scalable path toward more capable and versatile LALMs.

cs.SD↗
arXiv2026

Preserving Continuous Symmetry in Discrete Spaces: Geometric-Aware Quantization for SO(3)-Equivariant GNNs

Equivariant Graph Neural Networks (GNNs) are essential for physically consistent molecular simulations but suffer from high computational costs and memory bottlenecks, especially with high-order representations. While low-bit quantization offers a solution, applying it naively to rotation-sensitive features destroys the SO(3)-equivariant structure, leading to significant errors and violations of conservation laws. To address this issue, in this work, we propose a Geometric-Aware Quantization (GAQ) framework that compresses and accelerates equivariant models while rigorously preserving continuous symmetry in discrete spaces. Our approach introduces three key contributions: (1) a Magnitude-Direction Decoupled Quantization (MDDQ) scheme that separates invariant lengths from equivariant orientations to maintain geometric fidelity; (2) a symmetry-aware training strategy that treats scalar and vector features with distinct quantization schedules; and (3) a robust attention normalization mechanism to stabilize gradients in low-bit regimes. Experiments on the rMD17 benchmark demonstrate that our W4A8 models match the accuracy of FP32 baselines (9.31 meV vs. 23.20 meV) while reducing Local Equivariance Error (LEE) by over 30x compared to naive quantization. On consumer hardware, GAQ achieves 2.39x inference speedup and 4x memory reduction, enabling stable, energy-conserving molecular dynamics simulations for nanosecond timescales.

cs.LG↗
arXiv2026

Hybrid Tribo/piezoelectic Electrospun Nanofibers for Energy Harvesting Enhancement in Flexible Electronics

Triboelectric nanogenerators or TENGs and piezoelectric nanogenerators or PENGs have emerged as promising platforms for harvesting mechanical energy and converting it into electrical energy for powering flexible electronic devices. However, the material selection and structure design of such hybrid nanogenerator, and mechanisms of energy output still remain challenges. In this work, electrospinning is employed for the fabrication of nanofibers, particularly polyvinylidene fluoride or PVDF based nanofibers, due to its capability to generate high beta phase contents that effectively increase the piezoelectric performance of the PVDF friction layer, thereby enhancing the overall electrical performance for flexible electronics by merging tribo-piezoelectric power. Furthermore, various concentrations carbon nanotubes or CNT or graphene nanosheets or GNS are individually incorporated into the PVDF solution as nanofillers or NF to enhance the piezoelectric responses of the PVDF based nanofibers. The introduction of nanofillers is found to not only alter the fiber diameter but also modify the surface roughness of the electrospun nanofibers, and thus, enhancing the triboelectric effect. In addition, the output performance of the fabricated nanogenerator is predominantly governed by the piezoelectric effect rather than triboelectric effect, as the electrical output shows a strong positive correlation with the beta phase content of PVDF based nanofibers: the highest beta phase content reached to 85.3 percent and consistently resulted in the optimal energy output of 1.133 watt per meter square. Notably, the power density achieved by the prototype device reaches to the level of watt/m2, representing a substantial improvement compared with that of the conventional TENGs or PENGs reported to date, providing expanded opportunities for flexible electronic devices

cond-mat.mtrl-sci↗
arXiv2026

Advanced Control of Electron Beams: Tailoring X-ray Production with Programmable Laser Shaping

Leveraging the full scientific capabilities of next-generation high-repetition-rate free-electron lasers requires programmable control over electron-beam properties at their source. The photoinjector drive laser defines the electron beam's initial six-dimensional phase-space distribution, yet has historically been limited to Gaussian or static flat-top profiles, with most manipulation occurring downstream. Here we demonstrate software-programmable ultraviolet pulse shaping at the LCLS-II photoinjector as a source-level actuator that complements traditional accelerator controls. Using a coupled architecture combining dispersion-controlled nonlinear frequency conversion with spatial-light-modulator spectral shaping, we generate user-defined temporal structures and observe their imprint on electron bunches through high-resolution time-domain diagnostics. Laser-imposed multi-peaked modulation persists through acceleration, magnetic compression, and undulator transport with shot-to-shot repeatability, producing clearly resolved current structure in the compressed beam. Variance-based reconstruction from transverse deflecting cavity measurements reveals structured X-ray emission profiles exhibiting temporal features consistent with the programmed laser waveform. By providing rapid, software-controlled reconfiguration of electron-beam initial conditions, this source-level control approach establishes a programmable upstream actuator for future adaptive optimization and autonomous facility operation at high-repetition-rate light sources.

physics.acc-ph↗
arXiv2026

GA-Drive: Geometry-Appearance Decoupled Modeling for Free-viewpoint Driving Scene Generation

A free-viewpoint, editable, and high-fidelity driving simulator is crucial for training and evaluating end-to-end autonomous driving systems. In this paper, we present GA-Drive, a novel simulation framework capable of generating camera views along user-specified novel trajectories through Geometry-Appearance Decoupling and Diffusion-Based Generation. Given a set of images captured along a recorded trajectory and the corresponding scene geometry, GA-Drive synthesizes novel pseudo-views using geometry information. These pseudo-views are then transformed into photorealistic views using a trained video diffusion model. In this way, we decouple the geometry and appearance of scenes. An advantage of such decoupling is its support for appearance editing via state-of-the-art video-to-video editing techniques, while preserving the underlying geometry, enabling consistent edits across both original and novel trajectories. Extensive experiments demonstrate that GA-Drive substantially outperforms existing methods in terms of NTA-IoU, NTL-IoU, and FID scores.

cs.CV↗
arXiv2026

Bridging Sequential and Contextual Features with a Dual-View of Fine-grained Core-Behaviors and Global Interest-Distribution

Click-through rate (CTR) prediction tasks typically estimate the probability of a user clicking on a candidate item by modeling both user behavior sequence features and the item's contextual features, where the user behavior sequence is particularly critical as it dynamically reflects real-time shifts in user interest. Traditional CTR models often aggregate this dynamic sequence into a single vector before interacting it with contextual features. This approach, however, not only leads to behavior information loss during aggregation but also severely limits the model's capacity to capture interactions between contextual features and specific user behaviors, ultimately impairing its ability to capture fine-grained behavioral details and hindering models' prediction accuracy. Conversely, a naive approach of directly interacting with each user action with contextual features is computationally expensive and introduces significant noise from behaviors irrelevant to the candidate item. This noise tends to overwhelm the valuable signals arising from interactions involving more behaviors relevant to the candidate item. Therefore, to resolve the above issue, we propose a Core-Behaviors and Distributional-Compensation Dual-View Interaction Network (CDNet), which bridges the gap between sequential and contextual feature interactions from two complementary angles: a fine-grained interaction involving the most relevant behaviors and contextual features, and a coarse-grained interaction that models the user's overall interest distribution against the contextual features. By simultaneously capturing important behavioral details without forgoing the holistic user interest, CDNet effectively models the interplay between sequential and contextual features without imposing a significant computational burden. Ultimately, extensive experiments validate the effectiveness of CDNet.

cs.IR↗
arXiv2026

Sema: A High-performance System for LLM-based Semantic Query Processing

The integration of Large Language Models (LLMs) into data analytics has unlocked powerful capabilities for reasoning over bulk structured and unstructured data. However, existing systems typically rely on either DataFrame primitives, which lack the efficient execution infrastructure of modern DBMSs, or SQL User-Defined Functions (UDFs), which isolate semantic logic from the query optimizer and burden users with implementation complexities. The LLM-powered semantic operators also bring new challenges due to the high cost and non-deterministic nature of LLM invocation, where conventional optimization rules and cost models are inapplicable for their optimization. To bridge these gaps, we present Sema, a high-performance semantic query engine built on DuckDB that treats LLM-powered semantic operators as first-class citizens. Sema introduces SemaSQL, a declarative dialect that allows users seamlessly inject natural language expressions into standard SQL clauses, enabling end-to-end optimization and execution. At the logical level, the optimizer of Sema compresses natural language expressions and deduces relational constraints from semantic operators. At runtime, Sema employs Adaptive Query Execution (AQE) to dynamically reorder operators, fuse semantic operations, and apply prompt batching. This approach seeks a Pareto-optimal execution path balancing token consumption and latency under accuracy constraints. We evaluate Sema on 20 semantic queries across classification, summarization, and extraction tasks. Experimental results demonstrate that Sema achieves $2-10 \times$ speedup against three baseline systems while achieving competitive result quality.

cs.DB↗
arXiv2026

Physics-Guided Inverse Design of Optical Waveforms for Nonlinear Electromagnetic Dynamics

Structured optical waveforms are emerging as powerful control fields for the next generation of complex photonic and electromagnetic systems, where the temporal structure of light can determine the ultimate performance of scientific instruments. However, identifying optimal optical drive fields in strongly nonlinear regimes remains challenging because the mapping between optical inputs and system response is high-dimensional and typically accessible only through computationally expensive simulations. Here, we present a physics-guided deep learning framework for the inverse design of optical temporal waveforms. By training a light-weighted surrogate model on simulations, the method enables gradient-based synthesis of optical profiles that compensate nonlinear field distortions in driven particle-field systems. As a representative application, we apply the approach to the generation of electron beams used in advanced photon and particle sources. The learned optical waveform actively suppresses extrinsic emittance growth by more than 52% compared with conventional Gaussian operation and by approximately 9% relative to the theoretical flattop limit in simulation. We further demonstrate experimental feasibility by synthesizing the predicted waveform using a programmable pulse-shaping platform; incorporating the measured optical profile into beamline simulations yields a 31% reduction in the extrinsic emittance contribution. Beyond accelerator applications, this work establishes a general way for physics-guided inverse design of optical control fields, enabling structured light to approach fundamental performance limits in nonlinear photonic and high-frequency electromagnetic systems.

physics.optics↗
arXiv2026

Phase-Interface Instance Segmentation as a Visual Sensor for Laboratory Process Monitoring

Reliable visual monitoring of chemical experiments remains challenging in transparent glassware, where weak phase boundaries and optical artifacts degrade conventional segmentation. We formulate laboratory phenomena as the time evolution of phase interfaces and introduce the Chemical Transparent Glasses dataset 2.0 (CTG 2.0), a vessel-aware benchmark with 3,668 images, 23 glassware categories, and five multiphase interface types for phase-interface instance segmentation. Building on YOLO11m-seg, we propose LGA-RCM-YOLO, which combines Local-Global Attention (LGA) for robust semantic representation and a Rectangular Self-Calibration Module (RCM) for boundary refinement of thin, elongated interfaces. On CTG 2.0, the proposed model achieves 84.4% AP@0.5 and 58.43% AP@0.5-0.95, improving over the YOLO11m baseline by 6.42 and 8.75 AP points, respectively, while maintaining near real-time inference (13.67 FPS, RTX 3060). An auxiliary color-attribute head further labels liquid instances as colored or colorless with 98.71% precision and 98.32% recall. Finally, we demonstrate continuous process monitoring in separatory-funnel phase separation and crystallization, showing that phase-interface instance segmentation can serve as a practical visual sensor for laboratory automation.

cs.CV↗
arXiv2026

Logos: An evolvable reasoning engine for rational molecular design

The discovery and design of functional molecules remain central challenges across chemistry,biology, and materials science. While recent advances in machine learning have accelerated molecular property prediction and candidate generation, existing models tend to excel either in physical fidelity without transparent reasoning, or in flexible reasoning without guarantees of chemical validity. This imbalance limits the reliability of artificial intelligence systems in real scientific design workflows.Here we present Logos, a compact molecular reasoning model that integrates multi-step logical reasoning with strict chemical consistency. Logos is trained using a staged strategy that first exposes the model to explicit reasoning examples linking molecular descriptions to structural decisions, and then progressively aligns these reasoning patterns with molecular representations. In a final training phase, chemical rules and invariants are incorporated directly into the optimization objective, guiding the model toward chemically valid outputs. Across multiple benchmark datasets, Logos achieves strong performance in both structural accuracy and chemical validity, matching or surpassing substantially larger general-purpose language models while operating with a fraction of their parameters. Beyond benchmark evaluation, the model exhibits stable behaviour in molecular optimization tasks involving multiple, potentially conflicting constraints. By explicitly exposing intermediate reasoning steps, Logos enables human inspection and assessment of the design logic underlying each generated structure. These results indicate that jointly optimizing for reasoning structure and physical consistency offers a practical pathway toward reliable and interpretable AI systems for molecular science, supporting closer integration of artificial intelligence into scientific discovery processes.

cs.AI↗
arXiv2026

Bayesian neural network with autoencoder for model-based description of $α$-particle preformation factor

$α$ decay is an important probe for studying the structure of heavy and superheavy nuclei, in which the $α$-particle preformation ($P_α$) is a key physical quantity for describing decay half-lives. This work develops a hybrid framework that integrates Bayesian neural networks with autoencoder (BNN-Auto), combined with the cosh potential (CPT), to systematically optimize the constraint and prediction of $P_α$. The model employs variational inference for probabilistic modeling of network weights, naturally providing robust uncertainty quantification for predictions, and utilizes an autoencoder to enhance the robustness of feature representation. Based on experimental data from 535 nuclei, the BNN-Auto method achieves relative improvements in the root mean square deviation ($σ_{\rm{RMS}}$) of $P_α$ prediction by $61.14\%$ on the training set and $54.49\%$ on the validation set. Further analysis reveals that the $P_α$ and half-life extracted by the model exhibit pronounced odd-even staggering and shell effects in isotopic chains with $Z=86-90$ and isotones with $N=124-128$ and $N=150-154$. Moreover, we successfully predict the $α$ decay half-lives of nuclei with $Z=120$ and observe a significant increase in the half-life near $N=184$, which verifies the shell effect of the predicted 'stable island'. This study not only provides a high-precision theoretical description for $α$ decay, but also offers a new machine learning perspective for exploring the structure of superheavy nuclei.

nucl-th↗