arXiv subjects

Wei Sun

Publications and source records attributed to Wei Sun.

At least 55 records · Page 3Linked to original sources

arXiv2026

Wind-Aware Reinforcement Learning Control of a Small Quadrotor Using Learned Onboard Wind Estimation in Simulated Atmospheric Turbulence

Small multirotor aircraft are increasingly tasked with operations in the atmospheric boundary layer, where turbulent winds comparable to the vehicle's airspeed degrade trajectory tracking and can defeat conventional feedback control. This work illustrates a two-stage learning pipeline that first estimates the local wind from onboard kinematics and dynamics and then exploits that estimate inside a reinforcement learning (RL) flight controller. The wind estimator, an attention-augmented gated recurrent network trained on thousands of simulated flights through von Karman turbulence with power-law shear and veer, recovers the horizontal wind vector with a per-flight root-mean-square error of 0.40 m/s and a direction error of 3.2 degrees on unseen wind regimes, an accuracy near the floor imposed by unresolved turbulence, and generalizes to vertical ascent profiles with a skill score of 0.861 over a constant-wind reference. A proximal policy optimization controller receiving the frozen estimator's output reduces horizontal trajectory tracking error by 48% relative to a wind-blind proportional-derivative baseline across mean winds of 4 m/s to 12 m/s, winning on 100% of evaluation episodes. A three-way ablation decomposes this improvement into a kinematic component, available without wind information, and a wind-perception component; the perception share rises with wind speed, from small in light winds toward roughly half the total benefit in strong winds, consistent with the quadratic scaling of aerodynamic drag. The controller degrades gracefully on out-of-distribution winds of 13 m/s to 15 m/s, where the baseline fails catastrophically.

cs.LG↗

arXiv2026

LEIQ-Assessor: Multi-dimensional Quality Assessment of Low-light Enhanced Images via Multi-task Learning

Low-light image enhancement algorithms (LIEAs) aim to improve the visibility of images captured under poor illumination. However, the enhancement process often introduces artifacts such as noise amplification, color shift, structural damage, and over-exposure, which degrade the perceptual quality of the enhanced images. Therefore, a reliable image quality assessment (IQA) metric for evaluating enhancement effects is of great importance for both the development of LIEAs and their practical applications. In this paper, we present \textbf{LEIQ-Assessor}, a multi-dimensional quality assessment model for low-light image enhancement based on multi-task learning, developed for the QoMEX 2026 Grand Challenge on Low-light Enhanced Image Quality Assessment. Specifically, our method leverages a pre-trained SigLIP2 Vision Transformer as the backbone and simultaneously predicts the overall Mean Opinion Score (MOS) together with six perceptual sub-attributes: lightness, color fidelity, noise level, exposure quality, naturalness, and content recovery. By jointly optimizing these correlated objectives via the PLCC loss, the shared representation captures richer quality-aware features than its single-task counterpart. Experiments on the MLE benchmark demonstrate that LEIQ-Assessor significantly outperforms existing no-reference IQA models and hand-crafted quality descriptors. Our method achieved second place in the QoMEX 2026 Grand Challenge on Low-light Enhanced Image Quality Assessment. The code is available at https://github.com/sunwei925/LEIQ-Assessor.

cs.CV↗

arXiv2026

EEmo-Logic: A Unified Dataset and Multi-Stage Framework for Comprehensive Image-Evoked Emotion Assessment

Understanding the multi-dimensional attributes and intensity nuances of image-evoked emotions is pivotal for advancing machine empathy and empowering diverse human-computer interaction applications. However, existing models are still limited to coarse-grained emotion perception or deficient reasoning capabilities. To bridge this gap, we introduce EEmoDB, the largest image-evoked emotion understanding dataset to date. It features $5$ analysis dimensions spanning $5$ distinct task categories, facilitating comprehensive interpretation. Specifically, we compile $1.2M$ question-answering (QA) pairs (EEmoDB-QA) from $125K$ images via automated generation, alongside a $36K$ dataset (EEmoDB-Assess) curated from $25K$ images for fine-grained assessment. Furthermore, we propose EEmo-Logic, an all-in-one multimodal large language model (MLLM) developed via instruction fine-tuning and task-customized group relative preference optimization (GRPO) with novel reward design. Extensive experiments demonstrate that EEmo-Logic achieves robust performance in in-domain and cross-domain datasets, excelling in emotion QA and fine-grained assessment. The dataset and code are available at https://github.com/workerred/EEmo-Logic.

cs.CV↗

arXiv2026

Dynamical models for the two-parameter Poisson-Dirichlet distribution and the Pitman-Yor process

In this paper, we introduce and study a family of diffusion processes associated with the Pitman-Yor process and the two-parameter Poisson-Dirichlet distribution. The diffusion coefficients indexed by a non-negative parameter $γ$ are smaller than the corresponding one-parameter models in terms of quadratic forms or bilinear forms when $γ$ is positive. The well known Petrov's diffusion corresponds to $γ=0$ among the unlabelled diffusions. If $γ$ is the same as the stable parameter $α$ in the Pitman-Yor process, we obtain both labelled and unlabelled reversible diffusion processes with the Pitman-Yor process and the two-parameter Poisson-Dirichlet distribution as the corresponding reversible measures. We construct these processes analytically through Dirichlet forms. In comparison with existing models in the literature, our models possess two fundamental new features. Firstly, our labelled model is characterized by an explicit generator, which is the first among all models studied so far. This makes it possible to establish the crucial integration-by-parts formula. Secondly, a novel foundational structure of the two-parameter distributions is the existence of a diversity index (a multiple of local time) for the positive stable parameter. By slowing down the diffusion in our model, the essential role of the diversity index is revealed in the evolution of the population. Additionally, we also obtain properties including ergodicity, path behaviour, and finite dimensional approximations.

math.PR↗

arXiv2026

YouZhi: Towards High-Concurrency Financial LLMs via Adaptive GQA-to-MLA Transition

Large language models (LLMs) drive significant financial innovations, yet their high-concurrency deployment is severely bottlenecked by KV cache memory overhead, which inflates infrastructure costs and throttles scalability. To address this, we propose YouZhi-LLM, a highly efficient financial LLM empowered by a comprehensive structural transition and training pipeline natively built on the Huawei Ascend ecosystem. At its algorithmic core, YouZhi-LLM features a layer-adaptive GQA-to-MLA transition framework that dynamically assigns per-layer FreqFold sizes, maximizing KV-cache compression while minimizing perplexity degradation. To recover representation capacity and inject domain expertise, the Ascend-based training pipeline seamlessly integrates generalized knowledge distillation with financial-specific supervised fine-tuning. Evaluations demonstrate the superiority of this systematic approach, with the adaptive transition reducing perplexity degradation by up to 35% over uniform baselines. Crucially, when evaluated on Ascend NPUs via vLLM-Ascend, the massive KV-cache reduction translates directly into deployment efficiency. Compared to their respective base models, YouZhi-7B yields a 12.3% improvement in average financial benchmark score alongside a 2.69$\times$ increase in maximum concurrency; similarly, YouZhi-14B achieves a 7.0% accuracy gain and a 2.43$\times$ concurrency boost, establishing a new paradigm for cost-effective, high-throughput financial inference.

cs.CL↗

arXiv2026

Principled Synthetic Data Enables the First Scaling Laws for LLMs in Recommendation

Large Language Models (LLMs) represent a promising frontier for recommender systems, yet their development has been impeded by the absence of predictable scaling laws, which are crucial for guiding research and optimizing resource allocation. We hypothesize that this may be attributed to the inherent noise, bias, and incompleteness of raw user interaction data in prior continual pre-training (CPT) efforts. This paper introduces a novel, layered framework for generating high-quality synthetic data that circumvents such issues by creating a curated, pedagogical curriculum for the LLM. We provide powerful, direct evidence for the utility of our curriculum by showing that standard sequential models trained on our principled synthetic data significantly outperform ($+130\%$ on recall@100 for SasRec) models trained on real data in downstream ranking tasks, demonstrating its superiority for learning generalizable user preference patterns. Building on this, we empirically demonstrate, for the first time, robust power-law scaling for an LLM that is continually pre-trained on our high-quality, recommendation-specific data. Our experiments reveal consistent and predictable perplexity reduction across multiple synthetic data modalities. These findings establish a foundational methodology for reliable scaling LLM capabilities in the recommendation domain, thereby shifting the research focus from mitigating data deficiencies to leveraging high-quality, structured information.

cs.IR↗

arXiv2026

Evaluating Dataset Watermarking for Fine-tuning Traceability of Customized Diffusion Models: A Comprehensive Benchmark and Removal Approach

Recent fine-tuning techniques for diffusion models enable them to reproduce specific image sets, such as particular faces or artistic styles, but also introduce copyright and security risks. Dataset watermarking has been proposed to ensure traceability by embedding imperceptible watermarks into training images, which remain detectable in outputs even after fine-tuning. However, current methods lack a unified evaluation framework. To address this, this paper establishes a general threat model and introduces a comprehensive evaluation framework encompassing Universality, Transmissibility, and Robustness. Experiments show that existing methods perform well in universality and transmissibility, and exhibit some robustness against common image processing operations, yet still fall short under real-world threat scenarios. To reveal these vulnerabilities, the paper further proposes a practical watermark removal method that fully eliminates dataset watermarks without affecting fine-tuning, highlighting a key challenge for future research.

cs.CV↗

arXiv2026

Representation Collapse in Sequential Post-Training of Large Language Models

Large language models are now adapted through chains of post-training stages rather than through a single instruction-tuning pass. This paper studies whether such sequential post-training gradually compresses internal representations into low-rank, anisotropic, and homogeneous feature spaces. We define a measurement suite for hidden states, logits, token trajectories, and LoRA updates, and we use it to analyze supervised fine-tuning, preference optimization, safety/refusal tuning, math and code specialization, and long chain-of-thought tuning under controlled stage orderings. The central hypothesis is that excessive representation concentration is not merely a geometric curiosity: it predicts reduced plasticity during later adaptation, weaker out-of-domain generalization, and poorer calibration. We further evaluate lightweight interventions, including mixed-domain replay, feature refresh, representation diversity regularization, and LoRA update decorrelation, as ways to preserve future learnability without giving up the behavioral gains of post-training.

cs.LG↗

arXiv2026

A Scalable Benchmark Test Suite for Dynamic Multi-Objective Optimization with a Changing Number of Objectives

Dynamic multi-objective optimization with a changing number of objectives has recently attracted increasing attention due to its relevance to real-world problems whose evaluation criteria may evolve over time. However, existing benchmark test suites for this problem setting suffer from a fundamental limitation: when the number of objectives changes, the objective functions themselves also change implicitly. This makes it difficult to isolate and evaluate an algorithm's capability to handle dynamics in the number of objectives alone. In this paper, we analyze this issue in detail and show that several theoretical properties claimed in prior studies rely on an assumption that is violated by commonly used test suites. To address this problem, we propose a scalable benchmark test suite in which the objective functions are fixed throughout the optimization process, while the number of active objectives changes over time. Our benchmark is constructed by defining a maximum-objective problem and dynamically selecting subsets of objectives. To avoid degeneracy issues in classical DTLZ and WFG problems, we adopt Minus-DTLZ and Minus-WFG formulations, in which all objectives are mutually conflicting. Extensive benchmark studies using representative algorithms from the literature demonstrate the usefulness and flexibility of the proposed test suite.

cs.NE↗

arXiv2026

Generalizable Video Quality Assessment via Weak-to-Strong Learning

Video quality assessment (VQA) seeks to predict the perceptual quality of a video in alignment with human visual perception, serving as a fundamental tool for quantifying quality degradation across video processing workflows. The dominant VQA paradigm relies on supervised training with human-labeled datasets, which, despite substantial progress, still suffers from poor generalization to unseen video content. In this work, we explore weak-to-strong (W2S) learning as a new paradigm for advancing VQA without reliance on human-labeled datasets. We first provide empirical evidence that a straightforward W2S strategy allows a strong student model to not only match its weak teacher on in-domain benchmarks but also surpass it on out-of-distribution (OOD) benchmarks, revealing a distinct weak-to-strong effect in VQA. Building on this insight, we propose a novel framework that enhances W2S learning from two aspects: (1) integrating homogeneous and heterogeneous supervision signals from diverse VQA teachers -- including off-the-shelf VQA models and synthetic distortion simulators -- via a learn-to-rank formulation, and (2) iterative W2S training, where each strong student is recycled as the teacher in subsequent cycles, progressively focusing on challenging cases. Extensive experiments show that our method achieves state-of-the-art results across both in-domain and OOD benchmarks, with especially strong gains in OOD scenarios. Our findings highlight W2S learning as a principled route to break annotation barriers and achieve scalable generalization in video quality assessment. Our data and code will be available at https://github.com/clh124/W2S-VQA.

cs.CV↗

arXiv2026

PCDM: A Diffusion-Based Data Poisoning Attack Against Federated Learning Systems

Federated learning (FL) is vulnerable to data poisoning attacks due to its distributed nature. Although recent GAN-based data poisoning methods have indicated the potential of using generative AI to generate seemingly legitimate poisoned data, the inherent consistency of GAN outputs can still reveal a sign of data poisoning. In this paper, we propose a diffusion-based data poisoning framework against FL systems, which leverages a Poisoning-Oriented Conditional Diffusion Model (PCDM) to enable fine-grained control over the local generation of poisoned data while ensuring both attack effectiveness and stealthiness. Our PCDM incorporates an adjustable poisoning vector within the global context to precisely control the generation of poisoned data, with theoretical guarantees on attack performance. Furthermore, it employs a novel jumping diffusion strategy for lightweight and efficient poisoned data generation. We conduct the most systematic and broad experimental evaluation for FL poisoning attacks against various defenses, including advanced Byzantine robust aggregation mechanisms, on four open datasets: MNIST, Fashion-MNIST, CIFAR-10, CIFAR-100, and a real-world wireless-specific dataset VRAI. Our results demonstrate that PCDM is less likely to exhibit statistical anomalies compared with the state-of-the-art methods while more effectively degrading global FL performance, which poses a significant risk to data security in FL.

cs.CR↗

arXiv2026

When Thinking Pays Off: Incentive Alignment for Human-AI Collaboration

Collaboration with artificial intelligence (AI) has improved human decision-making across various domains by leveraging the complementary capabilities of humans and AI. Yet, humans systematically overrely on AI advice, even when their independent judgment would yield superior outcomes, fundamentally undermining the potential of human-AI complementarity. Building on prior work, we identify prevailing incentive structures in human-AI decision-making as a structural driver of this overreliance. To address this misalignment, we propose an alternative incentive mechanism designed to counteract systemic overreliance. We empirically evaluate this approach through a behavioral experiment with 180 participants, finding that the proposed mechanism significantly reduces overreliance. We also show that while appropriately designed incentives can enhance collaboration and decision quality, poorly designed incentives may distort behavior, introduce unintended consequences, and ultimately degrade performance. These findings underscore the importance of aligning incentives with task context and human-AI complementarities, and suggest that effective collaboration requires a shift toward context-sensitive incentive design.

cs.HC↗

arXiv2026

Neuro-Symbolic Proof Generation for Scaling Systems Software Verification

Formal verification via interactive theorem proving is increasingly used to ensure the correctness of critical systems, yet constructing large proof scripts remains highly manual and limits scalability. Advances in large language models (LLMs), especially in mathematical reasoning, make their integration into software verification increasingly promising. This paper introduces a neuro-symbolic proof generation framework designed to automate proof search for system-level verification projects. The framework performs a best-first tree search over proof states, repeatedly querying an LLM for the next candidate proof step. On the neural side, we fine-tune LLMs using datasets of proof state-step pairs; on the symbolic side, we incorporate a range of ITP tools to repair rejected steps, filter and rank proof states, and automatically discharge subgoals when search progress stalls. This synergy enables data-efficient LLM adaptation and semantics-informed pruning of the search space. We implement the framework on a new Isabelle REPL that exposes fine-grained proof states and automation tools, and evaluate it on the FVEL seL4 benchmark and additional Isabelle developments. On seL4, the system proves up to 77.6\% of the theorems, substantially surpassing previous LLM-based approaches and standalone Sledgehammer, while solving significantly more multi-step proofs. Results across further Isabelle benchmarks demonstrate strong generalization, indicating a viable path toward scalable automated software verification.

cs.AI↗

arXiv2026

Subjective and Objective Quality-of-Experience Evaluation Study for Live Video Streaming

In recent years, live video streaming has gained widespread popularity across various social media platforms. Quality of experience (QoE), which reflects end-users' satisfaction and overall experience, plays a critical role for media service providers to optimize large-scale live compression and transmission strategies to achieve perceptually optimal rate-distortion trade-off. Although many QoE metrics for video-on-demand (VoD) have been proposed, there remain significant challenges in developing QoE metrics for live video streaming. To bridge this gap, we conduct a comprehensive study of subjective and objective QoE evaluations for live video streaming. For the subjective QoE study, we introduce the first live video streaming QoE dataset, TaoLive QoE, which consists of $42$ source videos collected from real live broadcasts and $1,155$ corresponding distorted ones degraded due to a variety of streaming distortions, including conventional streaming distortions such as compression, stalling, as well as live streaming-specific distortions like frame skipping, variable frame rate, etc. Subsequently, a human study was conducted to derive subjective QoE scores of videos in the TaoLive QoE dataset. For the objective QoE study, we benchmark existing QoE models on the TaoLive QoE dataset as well as publicly available QoE datasets for VoD scenarios, highlighting that current models struggle to accurately assess video QoE, particularly for live content. Hence, we propose an end-to-end QoE evaluation model, Tao-QoE, which integrates multi-scale semantic features and optical flow-based motion features to predicting a retrospective QoE score, eliminating reliance on statistical quality of service (QoS) features.

cs.MM↗

arXiv2026

Production Rate of Glueball-like $X(2370)$ in $J/ψ$ Radiative Decay

$X(2370)$ falls in the mass region of the lowest pseudoscalar glueball predicted by lattice QCD studies and its decay properties are similar to those of $η_c$. A previous lattice QCD study finds that the pseudoscalar glueball ($G_{0^-}$) and the lowest pseudoscalar charmonium ($c\bar{c}(0^-)$) can mix with a small mixing angle $\sinθ$. It is therefore possible that $X(2370)$ and $η_c$ are admixtures of $G_{0^-}$ and $c\bar{c}(0^-)$. In this picture, although $\mathrm{Br}(J/ψ\toγη_c)$ is insensitive to the small $\sinθ$, $\mathrm{Br}(J/ψ\to γX(2370))$ can be enlarged drastically by the mixing due to the much larger kinematic factor for $J/ψ\to γX(2370)$ and the much larger transition form factor for $J/ψ\to γ(c\bar{c}(0^{-}))$. Depending on the value of $\sinθ$, $\mathrm{Br}(J/ψ\to γX(2370))$ can be much larger than that of the pure pseudoscalar glueball, namely, $2.3(8)\times 10^{-4}$ that is predicted by a quenched lattice QCD calculation. Present results by BESIII favor a small mixing angle of $\mathcal{O}(1^\circ)$, which can be further constrained by more measurements of $X(2370)$ decays if the mixing picture applies here.

hep-ph↗

arXiv2026

Scalable Unseen Objects 6-DoF Absolute Pose Estimation with Robotic Integration

Pose estimation-guided unseen object 6-DoF robotic manipulation is a key task in robotics. However, the scalability of current pose estimation methods to unseen objects remains a fundamental challenge, as they generally rely on CAD models or dense reference views of unseen objects, which are difficult to acquire, ultimately limit their scalability. In this paper, we introduce a novel task setup, referred to as SinRef-6D, which addresses 6-DoF absolute pose estimation for unseen objects using only a single pose-labeled reference RGB-D image captured during robotic manipulation. This setup is more scalable yet technically nontrivial due to large pose discrepancies and the limited geometric and spatial information contained in a single view. To address these issues, our key idea is to iteratively establish point-wise alignment in a common coordinate system with state space models (SSMs) as backbones. Specifically, to handle large pose discrepancies, we introduce an iterative object-space point-wise alignment strategy. Then, Point and RGB SSMs are proposed to capture long-range spatial dependencies from a single view, offering superior spatial modeling capability with linear complexity. Once pre-trained on synthetic data, SinRef-6D can estimate the 6-DoF absolute pose of an unseen object using only a single reference view. With the estimated pose, we further develop a hardware-software robotic system and integrate the proposed SinRef-6D into it in real-world settings. Extensive experiments on six benchmarks and in diverse real-world scenarios demonstrate that our SinRef-6D offers superior scalability. Additional robotic grasping experiments further validate the effectiveness of the developed robotic system. The code and robotic demos are available at https://paperreview99.github.io/SinRef-6DoF-Robotic.

cs.CV↗

arXiv2026

NTIRE 2026 The 3rd Restore Any Image Model (RAIM) Challenge: Professional Image Quality Assessment (Track 1)

In this paper, we present an overview of the NTIRE 2026 challenge on the 3rd Restore Any Image Model in the Wild, specifically focusing on Track 1: Professional Image Quality Assessment. Conventional Image Quality Assessment (IQA) typically relies on scalar scores. By compressing complex visual characteristics into a single number, these methods fundamentally struggle to distinguish subtle differences among uniformly high-quality images. Furthermore, they fail to articulate why one image is superior, lacking the reasoning capabilities required to provide guidance for vision tasks. To bridge this gap, recent advancements in Multimodal Large Language Models (MLLMs) offer a promising paradigm. Inspired by this potential, our challenge establishes a novel benchmark exploring the ability of MLLMs to mimic human expert cognition in evaluating high-quality image pairs. Participants were tasked with overcoming critical bottlenecks in professional scenarios, centering on two primary objectives: (1) Comparative Quality Selection: reliably identifying the visually superior image within a high-quality pair; and (2) Interpretative Reasoning: generating grounded, expert-level explanations that detail the rationale behind the selection. In total, the challenge attracted nearly 200 registrations and over 2,500 submissions. The top-performing methods significantly advanced the state of the art in professional IQA. The challenge dataset is available at https://github.com/narthchin/RAIM-PIQA, and the official homepage is accessible at https://www.codabench.org/competitions/12789/.

cs.CV↗

arXiv2026

SafeSci: Safety Evaluation of Large Language Models in Science Domains and Beyond

The success of large language models (LLMs) in scientific domains has heightened safety concerns, prompting numerous benchmarks to evaluate their scientific safety. Existing benchmarks often suffer from limited risk coverage and a reliance on subjective evaluation. To address these problems, we introduce SafeSci, a comprehensive framework for safety evaluation and enhancement in scientific contexts. SafeSci comprises SafeSciBench, a multi-disciplinary benchmark with 0.25M samples, and SafeSciTrain, a large-scale dataset containing 1.5M samples for safety enhancement. SafeSciBench distinguishes between safety knowledge and risk to cover extensive scopes and employs objective metrics such as deterministically answerable questions to mitigate evaluation bias. We evaluate 24 advanced LLMs, revealing critical vulnerabilities in current models. We also observe that LLMs exhibit varying degrees of excessive refusal behaviors on safety-related issues. For safety enhancement, we demonstrate that fine-tuning on SafeSciTrain significantly enhances the safety alignment of models. Finally, we argue that knowledge is a double-edged sword, and determining the safety of a scientific question should depend on specific context, rather than universally categorizing it as safe or unsafe. Our work provides both a diagnostic tool and a practical resource for building safer scientific AI systems.

cs.LG↗