arXiv subjects

Zongyi Li

Publications and source records attributed to Zongyi Li.

At least 19 recordsLinked to original sources

arXiv2026

Enhancing Localized Reasoning for Long Video Understanding via Efficient Segment-to-Video Supervision

Though Multimodal Large Language Models (MLLMs) have shown impressive potential in video understanding, long video understanding (LVU) remains challenging since distracting noise in complex and lengthy contexts can obscure localized details, misleading MLLMs to produce incorrect answers. Recent works mitigate these issues by incentivizing deep reasoning to include relevant evidence. However, these methods have two main problems: First, the reinforcement fine-tuning framework (RFT) they leveraged incurs substantial training overheads, including high annotation costs and complicated reward designs. Second, the self-reflective and iterative-perception mechanism in some methods causes lengthy outputs and high inference latency. To alleviate these problems, we propose a novel Segment-to-Video Supervision} method (S2V) to efficiently enhance fine-grained reasoning in LVU. Specifically, we generate question answer pairs (VQA) based on localized segments, and then transfer these segment-based VQA back to the whole video for training. Due to focusing on short segments, segment-based VQA can naturally notice details which tend to be overlooked from a whole-video perspective. Training on such data can enforce MLLMs to correctly associate fine-grained details with QA while avoiding distracting noise in the whole video. The S2V training involves just reinforcement learning (RL) with a simple accuracy reward based on only 10K VQA samples and the resulting S2V model predicts answer using a single forward pass with limited output tokens. Experimental results demonstrate that S2V can consistently improve LVU performance across multiple LVU benchmarks, outperforming both general MLLMs and reasoning-based methods not only in LVU accuracy but also in training and inference efficiency.

cs.CV↗

arXiv2026

Three-dimensional evolution of a solar filament with multipoint observations

In this paper, we first devise a geometrical model, featuring a torus-like flux rope based on the shape of 3DCORE model. The global shape of the torus is an ellipse, while the cross sections are circular along the torus. The thinnest point is located between the Sun center and photosphere. Deflections and inclination are considered as well. Using multiwavelength observations from perspectives of Earth, Ahead-STEREO (STA), and Solar Orbiter, we apply the model to three-dimensional (3D) reconstructions and tracking of the filament eruption, which was associated with a flare and a coronal mass ejection (CME) on 2024 October 8. The morphology, direction, and true velocity ($\sim$433 km/s) of the eruptive filament are obtained. It is found that the filament propagates nonradially, deflecting slightly eastward by $\sim$10 degrees and significantly southward by $\sim$40 degrees. Trajectory of the filament in the ecliptic plane reveals that the filament moves toward STA. The true direction of the eruptive filament using imaging and spectral observations is mutually verified by 3D reconstructions. The heliocentric distance of the filament increases from $\sim$1.68 to $\sim$2.94 solar radii within 35 minutes. Based on the results of 3D reconstructions, the true speed of the CME leading edge is evaluated to be 1046$-$1145 km/s.

astro-ph.SR↗

arXiv2026

Pinching-Antenna Systems (PASS)-Based User-Side Navigation: An Anchor-Line-based Approach

Pinching-antenna systems (PASS) are capable of dynamically reconfiguring wireless channels by flexibly repositioning pinching antennas (PAs) along the waveguides to establish short-range line-of-sight links. In this paper, a user-side navigation framework for PASS is proposed, where mobile users determine their own positions using only downlink broadcast signals without any prior knowledge of the PA positions. First, a Lambert W function-based PA positioning and pseudorange estimation (LWF-PAP) algorithm is developed, in which the closed-form expressions for both the PA positions along the waveguide and the PA-user pseudoranges are derived. Second, a weighted least squares-based PASS navigation (WLS-PAN) algorithm is formulated, where the nonlinear PASS-based navigation equations are transformed into a closed-form linear system, and the optimal weight matrix is derived, achieving minimum-variance unbiased estimation. Third, the PA-derived position dilution of precision (PA-PDOP) metric is further defined to characterize the theoretical accuracy bound. Simulation results demonstrate that centimeter-level positioning accuracy is achieved for both PAs and users within the breakpoint distance. It is also shown that uniform PA deployment and a moderate increase in the number of PAs effectively improve navigation accuracy, thereby validating the effectiveness and robustness of the proposed framework for distributed real-time user-side self-navigation.

eess.SP↗

arXiv2026

GeoPT: Scaling Physics Simulation via Lifted Geometric Pre-Training

Neural simulators promise efficient surrogates for physics simulation, but scaling them is bottlenecked by the prohibitive cost of generating high-fidelity training data. Pre-training on abundant off-the-shelf geometries offers a natural alternative, yet faces a fundamental gap: supervision on static geometry alone ignores dynamics and can lead to negative transfer on physics tasks. We present GeoPT, a unified pre-trained model for general physics simulation based on lifted geometric pre-training. The core idea is to augment geometry with synthetic dynamics, enabling dynamics-aware self-supervision without physics labels. Pre-trained on over one million samples, GeoPT consistently improves industrial-fidelity benchmarks spanning fluid mechanics for cars, aircraft, and ships, and solid mechanics in crash simulation, reducing labeled data requirements by 20-60% and accelerating convergence by 2$\times$. These results show that lifting with synthetic dynamics bridges the geometry-physics gap, unlocking a scalable path for neural simulation and potentially beyond. Code is available at https://github.com/Physics-Scaling/GeoPT.

cs.LG↗

arXiv2025

Youtu-Agent: Scaling Agent Productivity with Automated Generation and Hybrid Policy Optimization

Existing Large Language Model (LLM) agent frameworks face two significant challenges: high configuration costs and static capabilities. Building a high-quality agent often requires extensive manual effort in tool integration and prompt engineering, while deployed agents struggle to adapt to dynamic environments without expensive fine-tuning. To address these issues, we propose \textbf{Youtu-Agent}, a modular framework designed for the automated generation and continuous evolution of LLM agents. Youtu-Agent features a structured configuration system that decouples execution environments, toolkits, and context management, enabling flexible reuse and automated synthesis. We introduce two generation paradigms: a \textbf{Workflow} mode for standard tasks and a \textbf{Meta-Agent} mode for complex, non-standard requirements, capable of automatically generating tool code, prompts, and configurations. Furthermore, Youtu-Agent establishes a hybrid policy optimization system: (1) an \textbf{Agent Practice} module that enables agents to accumulate experience and improve performance through in-context optimization without parameter updates; and (2) an \textbf{Agent RL} module that integrates with distributed training frameworks to enable scalable and stable reinforcement learning of any Youtu-Agents in an end-to-end, large-scale manner. Experiments demonstrate that Youtu-Agent achieves state-of-the-art performance on WebWalkerQA (71.47\%) and GAIA (72.8\%) using open-weight models. Our automated generation pipeline achieves over 81\% tool synthesis success rate, while the Practice module improves performance on AIME 2024/2025 by +2.7\% and +5.4\% respectively. Moreover, our Agent RL training achieves 40\% speedup with steady performance improvement on 7B LLMs, enhancing coding/reasoning and searching capabilities respectively up to 35\% and 21\% on Maths and general/multi-hop QA benchmarks.

cs.AI↗

arXiv2025

Youtu-LLM: Unlocking the Native Agentic Potential for Lightweight Large Language Models

We introduce Youtu-LLM, a lightweight yet powerful language model that harmonizes high computational efficiency with native agentic intelligence. Unlike typical small models that rely on distillation, Youtu-LLM (1.96B) is pre-trained from scratch to systematically cultivate reasoning and planning capabilities. The key technical advancements are as follows: (1) Compact Architecture with Long-Context Support: Built on a dense Multi-Latent Attention (MLA) architecture with a novel STEM-oriented vocabulary, Youtu-LLM supports a 128k context window. This design enables robust long-context reasoning and state tracking within a minimal memory footprint, making it ideal for long-horizon agent and reasoning tasks. (2) Principled "Commonsense-STEM-Agent" Curriculum: We curated a massive corpus of approximately 11T tokens and implemented a multi-stage training strategy. By progressively shifting the pre-training data distribution from general commonsense to complex STEM and agentic tasks, we ensure the model acquires deep cognitive abilities rather than superficial alignment. (3) Scalable Agentic Mid-training: Specifically for the agentic mid-training, we employ diverse data construction schemes to synthesize rich and varied trajectories across math, coding, and tool-use domains. This high-quality data enables the model to internalize planning and reflection behaviors effectively. Extensive evaluations show that Youtu-LLM sets a new state-of-the-art for sub-2B LLMs. On general benchmarks, it achieves competitive performance against larger models, while on agent-specific tasks, it significantly surpasses existing SOTA baselines, demonstrating that lightweight models can possess strong intrinsic agentic capabilities.

cs.CL↗

arXiv2025

SmartSnap: Proactive Evidence Seeking for Self-Verifying Agents

Agentic reinforcement learning (RL) holds great promise for the development of autonomous agents under complex GUI tasks, but its scalability remains severely hampered by the verification of task completion. Existing task verification is treated as a passive, post-hoc process: a verifier (i.e., rule-based scoring script, reward or critic model, and LLM-as-a-Judge) analyzes the agent's entire interaction trajectory to determine if the agent succeeds. Such processing of verbose context that contains irrelevant, noisy history poses challenges to the verification protocols and therefore leads to prohibitive cost and low reliability. To overcome this bottleneck, we propose SmartSnap, a paradigm shift from this passive, post-hoc verification to proactive, in-situ self-verification by the agent itself. We introduce the Self-Verifying Agent, a new type of agent designed with dual missions: to not only complete a task but also to prove its accomplishment with curated snapshot evidences. Guided by our proposed 3C Principles (Completeness, Conciseness, and Creativity), the agent leverages its accessibility to the online environment to perform self-verification on a minimal, decisive set of snapshots. Such evidences are provided as the sole materials for a general LLM-as-a-Judge verifier to determine their validity and relevance. Experiments on mobile tasks across model families and scales demonstrate that our SmartSnap paradigm allows training LLM-driven agents in a scalable manner, bringing performance gains up to 26.08% and 16.66% respectively to 8B and 30B models. The synergizing between solution finding and evidence seeking facilitates the cultivation of efficient, self-verifying agents with competitive performance against DeepSeek V3.1 and Qwen3-235B-A22B. Code is available at: https://github.com/TencentYoutuResearch/SmartSnap

cs.CL↗

arXiv2025

Counterfactual LLM-based Framework for Measuring Rhetorical Style

The rise of AI has fueled growing concerns about ``hype'' in machine learning papers, yet a reliable way to quantify rhetorical style independently of substantive content has remained elusive. Because bold language can stem from either strong empirical results or mere rhetorical style, it is often difficult to distinguish between the two. To disentangle rhetorical style from substantive content, we introduce a counterfactual, LLM-based framework: multiple LLM rhetorical personas generate counterfactual writings from the same substantive content, an LLM judge compares them through pairwise evaluations, and the outcomes are aggregated using a Bradley--Terry model. Applying this method to 8,485 ICLR submissions sampled from 2017 to 2025, we generate more than 250,000 counterfactual writings and provide a large-scale quantification of rhetorical style in ML papers. We find that visionary framing significantly predicts downstream attention, including citations and media attention, even after controlling for peer-review evaluations. We also observe a sharp rise in rhetorical strength after 2023, and provide empirical evidence showing that this increase is largely driven by the adoption of LLM-based writing assistance. The reliability of our framework is validated by its robustness to the choice of personas and the high correlation between LLM judgments and human annotations. Our work demonstrates that LLMs can serve as instruments to measure and improve scientific evaluation.

cs.CL↗

arXiv2025

CUARewardBench: A Benchmark for Evaluating Reward Models on Computer-using Agent

Computer-using agents (CUAs) enable task completion through natural interaction with operating systems and software interfaces. While script-based verifiers are widely adopted for evaluation, they suffer from limited scalability and inability to provide step-wise assessment. Reward models offer promising alternatives, but their effectiveness on CUA evaluation remains largely underexplored. To address this gap, we present CUARewardBench, comprising four key contributions: (1) First-ever Comprehensive CUA Reward Benchmark: We introduce the first benchmark for evaluating both outcome reward models (ORM) and process reward models (PRM) on CUA tasks, enabling systematic assessment across trajectory-level and step-level evaluation. (2) Diverse, Practical and Reliable Dataset: CUARewardBench encompasses trajectories from 10 software categories and 7 agent architectures with varying performance levels (25.9%-50.8% success rates). All trajectories are expertly annotated through carefully designed protocols, with rigorous quality control to ensure reliability and practical applicability. (3) Comprehensive Analysis and Insights: Through extensive experiments across 7 vision-language models and 3 prompt templates, we reveal critical limitations of current CUA RMs, including insufficient visual reasoning capabilities, knowledge deficiencies, and the superiority of general VLMs over specialized CUA models for reward evaluation. (4) Unanimous Prompt Ensemble (UPE): Based on the insights from our comprehensive analysis, we propose UPE, a novel ensemble method that significantly enhances reward model reliability through strict unanimous voting and strategic prompt-template configurations. UPE achieves 89.8% precision and 93.3% NPV for ORM, and 81.7% precision and 85.1% NPV for PRM, substantially outperforming single VLMs and traditional ensemble approaches.

cs.SE↗

arXiv2025

Training-Free Group Relative Policy Optimization

Recent advances in Large Language Model (LLM) agents have demonstrated their promising general capabilities. However, their performance in specialized real-world domains often degrades due to challenges in effectively integrating external tools and specific prompting strategies. While methods like agentic reinforcement learning have been proposed to address this, they typically rely on costly parameter updates, for example, through a process that uses Supervised Fine-Tuning (SFT) followed by a Reinforcement Learning (RL) phase with Group Relative Policy Optimization (GRPO) to alter the output distribution. However, we argue that LLMs can achieve a similar effect on the output distribution by learning experiential knowledge as a token prior, which is a far more lightweight approach that not only addresses practical data scarcity but also avoids the common issue of overfitting. To this end, we propose Training-Free Group Relative Policy Optimization (Training-Free GRPO), a cost-effective solution that enhances LLM agent performance without any parameter updates. Our method leverages the group relative semantic advantage instead of numerical ones within each group of rollouts, iteratively distilling high-quality experiential knowledge during multi-epoch learning on a minimal ground-truth data. Such knowledge serves as the learned token prior, which is seamlessly integrated during LLM API calls to guide model behavior. Experiments on mathematical reasoning and web searching tasks demonstrate that Training-Free GRPO, when applied to DeepSeek-V3.1-Terminus, significantly improves out-of-domain performance. With just a few dozen training samples, Training-Free GRPO outperforms fine-tuned small LLMs with marginal training data and cost.

cs.CL↗

arXiv2025

Physics-informed Neural-operator Predictive Control for Drag Reduction in Turbulent Flows

Assessing turbulence control effects for wall friction numerically is a significant challenge since it requires expensive simulations of turbulent fluid dynamics. We instead propose an efficient deep reinforcement learning (RL) framework for modeling and control of turbulent flows. It is model-based RL for predictive control (PC), where both the policy and the observer models for turbulence control are learned jointly using Physics Informed Neural Operators (PINO), which are discretization invariant and can capture fine scales in turbulent flows accurately. Our PINO-PC outperforms prior model-free reinforcement learning methods in various challenging scenarios where the flows are of high Reynolds numbers and unseen, i.e., not provided during model training. We find that PINO-PC achieves a drag reduction of 39.0\% under a bulk-velocity Reynolds number of 15,000, outperforming previous fluid control methods by more than 32\%.

cs.LG↗

arXiv2025

Learn the Ropes, Then Trust the Wins: Self-imitation with Progressive Exploration for Agentic Reinforcement Learning

Reinforcement learning (RL) is the dominant paradigm for sharpening strategic tool use capabilities of LLMs on long-horizon, sparsely-rewarded agent tasks, yet it faces a fundamental challenge of exploration-exploitation trade-off. Existing studies stimulate exploration through the lens of policy entropy, but such mechanical entropy maximization is prone to RL instability due to the multi-turn distribution shifting. In this paper, we target the progressive exploration-exploitation balance under the guidance of the agent's own experiences without succumbing to either entropy collapsing or runaway divergence. We propose SPEAR, a self-imitation learning (SIL) recipe for training agentic LLMs. It extends the vanilla SIL, where a replay buffer stores good experience for off-policy update, by gradually steering the policy entropy across stages. Specifically, the proposed curriculum scheduling harmonizes intrinsic reward shaping and self-imitation to 1) expedite exploration via frequent tool interactions at the beginning, and 2) strengthen exploitation of successful tactics upon convergence towards familiarity with the environment. We also combine bag-of-tricks of industrial RL optimizations for a strong baseline Dr.BoT to demonstrate our effectiveness. In ALFWorld and WebShop, SPEAR increases the success rates of GRPO/GiGPO/Dr.BoT by up to 16.1%/5.1%/8.6% and 20.7%/11.8%/13.9%, respectively. In AIME24 and AIME25, SPEAR boosts Dr.BoT by up to 3.8% and 6.1%, respectively. Such gains incur only 10%-25% extra theoretical complexity and negligible runtime overhead in practice, demonstrating the plug-and-play scalability of SPEAR.

cs.LG↗

arXiv2025

InconVAD: A Two-Stage Dual-Tower Framework for Multimodal Emotion Inconsistency Detection

Detecting emotional inconsistency across modalities is a key challenge in affective computing, as speech and text often convey conflicting cues. Existing approaches generally rely on incomplete emotion representations and employ unconditional fusion, which weakens performance when modalities are inconsistent. Moreover, little prior work explicitly addresses inconsistency detection itself. We propose InconVAD, a two-stage framework grounded in the Valence/Arousal/Dominance (VAD) space. In the first stage, independent uncertainty-aware models yield robust unimodal predictions. In the second stage, a classifier identifies cross-modal inconsistency and selectively integrates consistent signals. Extensive experiments show that InconVAD surpasses existing methods in both multimodal emotion inconsistency detection and modeling, offering a more reliable and interpretable solution for emotion analysis.

cs.MM↗

arXiv2025

Harnessing Multimodal Large Language Models for Personalized Product Search with Query-aware Refinement

Personalized product search (PPS) aims to retrieve products relevant to the given query considering user preferences within their purchase histories. Since large language models (LLM) exhibit impressive potential in content understanding and reasoning, current methods explore to leverage LLM to comprehend the complicated relationships among user, query and product to improve the search performance of PPS. Despite the progress, LLM-based PPS solutions merely take textual contents into consideration, neglecting multimodal contents which play a critical role for product search. Motivated by this, we propose a novel framework, HMPPS, for \textbf{H}arnessing \textbf{M}ultimodal large language models (MLLM) to deal with \textbf{P}ersonalized \textbf{P}roduct \textbf{S}earch based on multimodal contents. Nevertheless, the redundancy and noise in PPS input stand for a great challenge to apply MLLM for PPS, which not only misleads MLLM to generate inaccurate search results but also increases the computation expense of MLLM. To deal with this problem, we additionally design two query-aware refinement modules for HMPPS: 1) a perspective-guided summarization module that generates refined product descriptions around core perspectives relevant to search query, reducing noise and redundancy within textual contents; and 2) a two-stage training paradigm that introduces search query for user history filtering based on multimodal representations, capturing precise user preferences and decreasing the inference cost. Extensive experiments are conducted on four public datasets to demonstrate the effectiveness of HMPPS. Furthermore, HMPPS is deployed on an online search system with billion-level daily active users and achieves an evident gain in A/B testing.

cs.MM↗

arXiv2025

Physics-Aware Neural Operators for Direct Inversion in 3D Photoacoustic Tomography

Learning physics-constrained inverse operators-rather than post-processing physics-based reconstructions-is a broadly applicable strategy for problems with expensive forward models. We demonstrate this principle in three-dimensional photoacoustic computed tomography (3D PACT), where current systems demand dense transducer arrays and prolonged scans, restricting clinical translation. We introduce PANO (PACT imaging neural operator), an end-to-end physics-aware neural operator-a deep learning architecture that generalizes across input sampling densities without retraining-that directly learns the inverse mapping from raw sensor measurements to a 3D volumetric image. Unlike two-step methods that reconstruct then denoise, PANO performs direct inversion in a single pass, jointly embedding physics and data priors. It employs spherical discrete-continuous convolutions to respect hemispherical sensor geometry and Helmholtz equation constraints to ensure physical consistency. PANO reconstructs high-quality images from both simulated and real data across diverse sparse acquisition settings, achieves real-time inference and outperforms the widely-used UBP algorithm by approximately 33 percentage points in cosine similarity on simulated data and 14 percentage points on real phantom data. These results establish a pathway toward more accessible 3D PACT systems for preclinical research, and motivate future in-vivo validation for clinical translation.

eess.IV↗

arXiv2025

Geometric Operator Learning with Optimal Transport

We propose integrating optimal transport (OT) into operator learning for partial differential equations (PDEs) on complex geometries. Classical geometric learning methods typically represent domains as meshes, graphs, or point clouds. Our approach generalizes discretized meshes to mesh density functions, formulating geometry embedding as an OT problem that maps these functions to a uniform density in a reference space. Compared to previous methods relying on interpolation or shared deformation, our OT-based method employs instance-dependent deformation, offering enhanced flexibility and effectiveness. For 3D simulations focused on surfaces, our OT-based neural operator embeds the surface geometry into a 2D parameterized latent space. By performing computations directly on this 2D representation of the surface manifold, it achieves significant computational efficiency gains compared to volumetric simulation. Experiments with Reynolds-averaged Navier-Stokes equations (RANS) on the ShapeNet-Car and DrivAerNet-Car datasets show that our method achieves better accuracy and also reduces computational expenses in terms of both time and memory usage compared to existing machine learning models. Additionally, our model demonstrates significantly improved accuracy on the FlowBench dataset, underscoring the benefits of employing instance-dependent deformation for datasets with highly variable geometries.

cs.LG↗

arXiv2025

Scale-Consistent Learning for Partial Differential Equations

Machine learning (ML) models have emerged as a promising approach for solving partial differential equations (PDEs) in science and engineering. Previous ML models typically cannot generalize outside the training data; for example, a trained ML model for the Navier-Stokes equations only works for a fixed Reynolds number ($Re$) on a pre-defined domain. To overcome these limitations, we propose a data augmentation scheme based on scale-consistency properties of PDEs and design a scale-informed neural operator that can model a wide range of scales. Our formulation leverages the facts: (i) PDEs can be rescaled, or more concretely, a given domain can be re-scaled to unit size, and the parameters and the boundary conditions of the PDE can be appropriately adjusted to represent the original solution, and (ii) the solution operators on a given domain are consistent on the sub-domains. We leverage these facts to create a scale-consistency loss that encourages matching the solutions evaluated on a given domain and the solution obtained on its sub-domain from the rescaled PDE. Since neural operators can fit to multiple scales and resolutions, they are the natural choice for incorporating scale-consistency loss during training of neural PDE solvers. We experiment with scale-consistency loss and the scale-informed neural operator model on the Burgers' equation, Darcy Flow, Helmholtz equation, and Navier-Stokes equations. With scale-consistency, the model trained on $Re$ of 1000 can generalize to $Re$ ranging from 250 to 10000, and reduces the error by 34% on average of all datasets compared to baselines.

cs.LG↗

arXiv2025

High precision PINNs in unbounded domains: application to singularity formulation in PDEs

We investigate the high-precision training of Physics-Informed Neural Networks (PINNs) in unbounded domains, with a special focus on applications to singularity formulation in PDEs. We propose a modularized approach and study the choices of neural network ansatz, sampling strategy, and optimization algorithm. When combined with rigorous computer-assisted proofs and PDE analysis, the numerical solutions identified by PINNs, provided they are of high precision, can serve as a powerful tool for studying singularities in PDEs. For 1D Burgers equation, our framework can lead to a solution with very high precision, and for the 2D Boussinesq equation, which is directly related to the singularity formulation in 3D Euler and Navier-Stokes equations, we obtain a solution whose loss is $4$ digits smaller than that obtained in \cite{wang2023asymptotic} with fewer training steps. We also discuss potential directions for pushing towards machine precision for higher-dimensional problems.

cs.LG↗