arXiv subjects

Dong Li

Publications and source records attributed to Dong Li.

At least 163 records · Page 9Linked to original sources

arXiv2025

Modeling Hypergraph Using Large Language Models

Due to the advantages of hypergraphs in modeling high-order relationships in complex systems, they have been applied to higher-order clustering, hypergraph neural networks and computer vision. These applications rely heavily on access to high-quality, large-scale real-world hypergraph data. Yet, compared to traditional pairwise graphs, real hypergraph datasets remain scarce in both scale and diversity. This shortage significantly limits the development and evaluation of advanced hypergraph learning algorithms. Therefore, how to quickly generate large-scale hypergraphs that conform to the characteristics of real networks is a crucial task that has not received sufficient attention. Motivated by recent advances in large language models (LLMs), particularly their capabilities in semantic reasoning, structured generation, and simulating human behavior, we investigate whether LLMs can facilitate hypergraph generation from a fundamentally new perspective. We introduce HyperLLM, a novel LLM-driven hypergraph generator that simulates the formation and evolution of hypergraphs through a multi-agent collaboration. The framework integrates prompts and structural feedback mechanisms to ensure that the generated hypergraphs reflect key real-world patterns. Extensive experiments across diverse datasets demonstrate that HyperLLM achieves superior fidelity to structural and temporal hypergraph patterns, while requiring minimal statistical priors. Our findings suggest that LLM-based frameworks offer a promising new direction for hypergraph modeling.

cs.SI↗

arXiv2025

Omni-Thinker: Scaling Multi-Task RL in LLMs with Hybrid Reward and Task Scheduling

The pursuit of general-purpose artificial intelligence depends on large language models (LLMs) that can handle both structured reasoning and open-ended generation. We present Omni-Thinker, a unified reinforcement learning (RL) framework that scales LLMs across diverse tasks by combining hybrid rewards with backward-transfer-guided scheduling. Hybrid rewards integrate rule-based verifiable signals with preference-based evaluations from an LLM-as-a-Judge, enabling learning in both deterministic and subjective domains. Our scheduler orders tasks according to accuracy backward transfer (BWT), reducing forgetting and improving multi-task performance. Experiments across four domains show gains of 6.2% over joint training and 12.4% over model merging. Moreover, we demonstrate that simple assumptions on accuracy transfer yield accurate predictions of curriculum outcomes, with entropy dynamics explaining deviations due to generative tasks. These findings underscore the importance of BWT-aware scheduling and hybrid supervision for scaling RL-based post-training toward general-purpose LLMs.

cs.LG↗

arXiv2025

Extreme Ultraviolet Wave and Quasi-periodic Pulsations during an eruptive M-class Flare

In this paper, we report multiwavelength and multipoint observations of the prominence eruption originating from active region 11163, which generated an M3.5 class flare and a coronal mass ejection (CME) on 2011 February 24. The prominence lifts off and propagates nonradially in the southeast direction. Using the revised cone model, we carry out three-dimensional reconstructions of the icecream-like prominence. The latitudinal inclination angle decreases from $\sim$60$\degr$ to $\sim$37$\degr$, indicating that the prominence tends to propagate more radially. The longitudinal inclination angle almost keeps constant (-6$\degr$). The highly inclined prominence eruption and the related CME drive an extreme ultraviolet (EUV) wave, which propagates southward at speeds of $\sim$381.60 km s$^{-1}$ and $\sim$398.59 km s$^{-1}$ observed in 193 Å and 304 Å, respectively. The M3.5 class flare presents quasi-periodic pulsations (QPPs) in soft X-ray, hard X-ray, EUV, and radio wavelengths with periods of 80$-$120 s. Cotemporary with the flare QPPs, a thin current sheet and multiple plasmoids are observed following the eruptive prominence. Combining with the appearance of drifting pulsation structure, the QPPs are most probably generated by quasi-periodic magnetic reconnection and particle accelerations as a result of plasmoids in the current sheet.

astro-ph.SR↗

arXiv2025

SpecVLM: Fast Speculative Decoding in Vision-Language Models

Speculative decoding is a powerful way to accelerate autoregressive large language models (LLMs), but directly porting it to vision-language models (VLMs) faces unique systems constraints: the prefill stage is dominated by visual tokens whose count scales with image resolution and video length, inflating both compute and memory, especially the key-value (KV) cache. We study speculative decoding for VLMs and introduce SpecVLM, a practical system that (1) establishes a strong EAGLE-2-style baseline, EagleVLM, delivering 1.5--2.3x end-to-end speedups over full autoregressive inference, and (2) further accelerates VLM inference with an elastic visual compressor that adaptively selects among pruning, pooling, convolution, and resampler primitives to balance FLOPs/parameters and accuracy per input. To avoid costly offline distillation corpora, we propose an online-logit distillation protocol that trains the draft model with on-the-fly teacher logits and penultimate features using a combined cross-entropy and Smooth L1 objective, eliminating storage and preprocessing while remaining compute-efficient. This protocol reveals a training-time scaling effect: longer online training monotonically increases the draft model's average accepted length, improving speculative efficiency. Empirically, SpecVLM achieves additional acceleration, culminating in 2.5--2.9x end-to-end speedups within 5 epochs across LLaVA and MMMU, consistently over resolutions and task difficulties, while preserving the target model's output distribution (lossless decoding). Our code is available at https://github.com/haiduo/SpecVLM.

cs.CV↗

arXiv2025

Modelling time series of counts with hysteresis

In this article, we propose a novel model for time series of counts called the hysteretic Poisson autoregressive (HPART) model with thresholds by extending the linear Poisson autoregressive model into a nonlinear model. Unlike other approaches that bear the adjective ``hysteretic", our model incorporates a scientifically relevant controlling factor that produces genuine hysteresis. Further, we re-analyse the buffered Poisson autoregressive (BPART) model with thresholds. Although the two models share the convenient piecewise linear structure, the HPART model probes deeper into the intricate dynamics that governs regime switching. We study the maximum likelihood estimation of the parameters of both models and their asymptotic properties in a unified manner, establish tests of separate families of hypotheses for the non-nested case involving a BPART model and a HPART model, and demonstrate the finite-sample efficacy of parameter estimation and tests with Monte Carlo simulation. We showcase advantages of the HPART model with two real time series, including plausible interpretations and improved out-of-sample predictions.

stat.ME↗

arXiv2025

Superpose Task-specific Features for Model Merging

Model merging enables powerful capabilities in neural networks without requiring additional training. In this paper, we introduce a novel perspective on model merging by leveraging the fundamental mechanisms of neural network representation. Our approach is motivated by the linear representation hypothesis, which states that neural networks encode information through linear combinations of feature vectors. We propose a method that superposes task-specific features from individual models into a merged model. Our approach specifically targets linear transformation matrices, which are crucial for feature activation and extraction in deep networks. By formulating the merging process as a linear system, we can preserve task-specific features from individual models and create merged models that effectively maintain multi-task capabilities compared to existing methods. Extensive experiments across diverse benchmarks and models demonstrate that our method outperforms existing techniques. Code is available at https://github.com/LARS-research/STF.

cs.LG↗

arXiv2025

Face4FairShifts: A Large Image Benchmark for Fairness and Robust Learning across Visual Domains

Ensuring fairness and robustness in machine learning models remains a challenge, particularly under domain shifts. We present Face4FairShifts, a large-scale facial image benchmark designed to systematically evaluate fairness-aware learning and domain generalization. The dataset includes 100,000 images across four visually distinct domains with 39 annotations within 14 attributes covering demographic and facial features. Through extensive experiments, we analyze model performance under distribution shifts and identify significant gaps. Our findings emphasize the limitations of existing related datasets and the need for more effective fairness-aware domain adaptation techniques. Face4FairShifts provides a comprehensive testbed for advancing equitable and reliable AI systems. The dataset is available online at https://meviuslab.github.io/Face4FairShifts/.

cs.CV↗

arXiv2025

Accurate Measles Rash Detection via Vision Transformer Fine-Tuning

Measles, a highly contagious disease declared eliminated in the United States in 2000 after decades of successful vaccination campaigns, resurged in 2025, with 1,356 confirmed cases reported as of August 5, 2025. Given its rapid spread among susceptible individuals, fast and reliable diagnostic systems are critical for early prevention and containment. In this work, we applied transfer learning to fine-tune a pretrained Data-efficient Image Transformer (DeiT) model for distinguishing measles rashes from other skin conditions. After tuning the classification head on a diverse, curated skin rash image dataset, the DeiT model achieved an average classification accuracy of 95.17%, precision of 95.06%, recall of 95.17%, and an F1-score of 95.03%, demonstrating high effectiveness in accurate measles detection to aid outbreak control. We also compared the DeiT model with a convolutional neural network and discussed the directions for future research.

eess.IV↗

arXiv2025

DeepAP: Deep Learning-based Aperture Photometry Feasibility Assessment and Aperture Size Prediction

Aperture photometry is a fundamental technique widely used to obtain high-precision light curves in optical survey projects like Tianyu. However, its effectiveness is limited in crowded fields, and the choice of aperture size critically impacts photometric precision. To address these challenges, we propose DeepAP, an efficient and accurate two-stage deep learning framework for aperture photometry. Specifically, for a given source, we first train a Vision Transformer (ViT) model to assess its feasibility of aperture photometry. We then train the Residual Neural Network (ResNet) to predict its optimal aperture size. For aperture photometry feasibility assessment, the ViT model yields an ROC AUC value of 0.96, and achieves a precision of 0.974, a recall of 0.930, and an F1 score of 0.952 on the test set. For aperture size prediction, the ResNet model effectively mitigates biases inherent in classical growth curve methods by adaptively selecting apertures appropriate for sources of varying brightness, thereby enhancing the signal-to-noise ratio (SNR) across a wide range of targets. Meanwhile, some samples in the test set have a higher SNR than those obtained by exhaustive aperture size enumeration because of the finer granularity of aperture size estimation. By integrating ResNet with the ViT network, the DeepAP framework achieves a median total processing time of 18 milliseconds for a batch of 10 images, representing a speed-up of approximately 59000 times compared to exhaustive aperture size enumeration. This work paves the way for the automatic application of aperture photometry in future high-precision surveys such as Tianyu and LSST. The source code and model are available at https://github.com/ruiyicheng/DeepAP.

astro-ph.IM↗

arXiv2025

T-GRAG: A Dynamic GraphRAG Framework for Resolving Temporal Conflicts and Redundancy in Knowledge Retrieval

Large language models (LLMs) have demonstrated strong performance in natural language generation but remain limited in knowle- dge-intensive tasks due to outdated or incomplete internal knowledge. Retrieval-Augmented Generation (RAG) addresses this by incorporating external retrieval, with GraphRAG further enhancing performance through structured knowledge graphs and multi-hop reasoning. However, existing GraphRAG methods largely ignore the temporal dynamics of knowledge, leading to issues such as temporal ambiguity, time-insensitive retrieval, and semantic redundancy. To overcome these limitations, we propose Temporal GraphRAG (T-GRAG), a dynamic, temporally-aware RAG framework that models the evolution of knowledge over time. T-GRAG consists of five key components: (1) a Temporal Knowledge Graph Generator that creates time-stamped, evolving graph structures; (2) a Temporal Query Decomposition mechanism that breaks complex temporal queries into manageable sub-queries; (3) a Three-layer Interactive Retriever that progressively filters and refines retrieval across temporal subgraphs; (4) a Source Text Extractor to mitigate noise; and (5) a LLM-based Generator that synthesizes contextually and temporally accurate responses. We also introduce Time-LongQA, a novel benchmark dataset based on real-world corporate annual reports, designed to test temporal reasoning across evolving knowledge. Extensive experiments show that T-GRAG significantly outperforms prior RAG and GraphRAG baselines in both retrieval accuracy and response relevance under temporal constraints, highlighting the necessity of modeling knowledge evolution for robust long-text question answering. Our code is publicly available on the T-GRAG

cs.AI↗

arXiv2025

Geak: Introducing Triton Kernel AI Agent & Evaluation Benchmarks

The demand for AI-generated GPU kernels is rapidly growing, influenced by the need for scalable, hardware-optimized solutions in both industry and academia. As deep learning workloads grow in complexity and diversity, it is imperative to automate low-level kernel development to meet performance and productivity demands. Major cloud providers, semiconductor companies, and research institutions are now investing heavily in AI-driven code generation for GPUs, aiming to reduce manual optimization efforts while achieving near-expert performance on hardware like AMD MI300X. The Triton language, a Python-based DSL for GPU programming, has emerged as a popular target for such AI-generated kernels due to its balance of performance and ease-of-coding. In this work, we present an evaluation suite for Triton-based GPU kernels and GEAK (Generating Efficient AI-centric GPU Kernels)-a framework that leverages cutting-edge LLMs to generate performant Triton code specifically for AMD GPUs, including the AMD MI300X and MI250. GEAK leverages inference-time compute scaling to produce Triton-based GPU kernels using a reasoning loop adapted from Reflexion-style feedback mechanisms. On two evaluation benchmarks, GEAK significantly outperformed the baselines of directly prompting frontier LLMs as well as Reflexion-based generation pipelines by achieving correctness up to $63$% and execution speed up of up to $2.59$X. These results highlight the promise of GEAK-like agentic code generation for accelerating the adoption of diverse hardware platforms and democratizing access to expert-level kernel performance.

cs.CL↗

arXiv2025

Exploring Textual Semantics Diversity for Image Transmission in Semantic Communication Systems using Visual Language Model

In recent years, the rapid development of machine learning has brought reforms and challenges to traditional communication systems. Semantic communication has appeared as an effective strategy to effectively extract relevant semantic signals semantic segmentation labels and image features for image transmission. However, the insufficient number of extracted semantic features of images will potentially result in a low reconstruction accuracy, which hinders the practical applications and still remains challenging for solving. In order to fill this gap, this letter proposes a multi-text transmission semantic communication (Multi-SC) system, which uses the visual language model (VLM) to assist in the transmission of image semantic signals. Unlike previous image transmission semantic communication systems, the proposed system divides the image into multiple blocks and extracts multiple text information from the image using a modified large language and visual assistant (LLaVA), and combines semantic segmentation tags with semantic text for image recovery. Simulation results show that the proposed text semantics diversity scheme can significantly improve the reconstruction accuracy compared with related works.

cs.CV↗

arXiv2025

Optimal Gevrey regularity for supercritical quasi-geostrophic equations

We consider the two dimensional surface quasi-geostrophic equations with super-critical dissipation. For large initial data in critical Sobolev and Besov spaces, we prove optimal Gevrey regularity with the same decay exponent as the linear part. This settles several open problems in \cite{Bis14, BMS15}.

math.AP↗

arXiv2025

Does Movable Antenna Present A Dual-edged Nature? From the Perspective of Physical Layer Security: A Joint Design of Fixed-position Antenna and Movable Antenna

In conventional artificial noise (AN)-aided physical-layer security systems, fixed-position antenna (FPA) arrays exhibit inherent vulnerability to coverage gaps due to their static spatial configuration. Adversarial eavesdroppers can strategically exploit their mobility to infiltrate these spatial nulls of AN radiation patterns, thereby evading interference suppression and successfully intercepting the confidential communication. To overcome this limitation, in this paper, we investigate a hybrid antenna deployment framework integrating FPA arrays and movable antenna (MA) arrays (denoted by FMA co-design) to address the security performance in dynamic wireless environments, based on the fact that MA arrays enable channel reconfiguration through localized antenna repositioning, achieving more higher spatial degree of freedom (DoF). Enabled by FMA co-design framework, FPA arrays ensure baseline connectivity for legitimate links while MA arrays function as dynamic security enhancers, replacing conventional static AN generation. Furthermore, we formulate a non-convex optimization problem of the secrecy rate maximization through jointly optimizing MA positioning, FPA beamforming, and MA beamforming under practical constraints. the solution employs a dual-algorithm approach: Nesterov momentum-based projected gradient ascent (NMPGA) accelerates convergence in continuous position optimization, while alternating optimization (AO) handles coupled beamforming design. Experimental evaluations demonstrate that the proposed FMA co-design framework achieves significant secrecy performance gains over individual optimization benchmarks, yielding 42.34% and 9.12% improvements in secrecy rate compared to isolated FPA for AN generation and MA for confidential information baselines, respectively.

cs.IT↗

arXiv2025

An Effective Equivalence Model of Analyzing PLS of Multiple Eavesdroppers Facing Low-altitude Communication Systems

In low-altitude wireless communications, the increased complexity of wireless channels and the uncertainty of eavesdroppers (Eves)--caused by diverse altitudes, speeds, and obstacles--pose significant challenges to physical layer security (PLS) technologies based on fixed-position antennas (FPAs), particularly in terms of beamforming capabilities and spatial efficiency. In contrast, movable antennas (MAs) offer a flexible solution by enabling channel reconstruction through antenna movement, effectively compensating for the limitations of FPAs. In this paper, we aim to derive a closed-form expression for the secrecy rate, a key metric in PLS, which is often unattainable in current studies due to the uncertainty of Eves. We construct an equivalent model that leverages the reconfigurable nature of MAs, equating the secrecy rates obtained by multiple Eves with single FPAs to those achieved by a single virtual Eve equipped with an MA array. To minimize the gap between these two types of secrecy rates, we formulate and solve an optimization problem by jointly designing the equivalent distance between the transmitter and the virtual Eve} and the antenna positions of MAs at the virtual Eve. Numerical simulations validate the effectiveness of the proposed equivalent model, offering a new perspective for PLS strategies. This work provides significant insights for network designers on how system parameters affect PLS performance.

cs.IT↗

arXiv2025

Image Generation with Supervised Selection Based on Multimodal Features for Semantic Communications

Semantic communication (SemCom) has emerged as a promising technique for the next-generation communication systems, in which the generation at the receiver side is allowed with semantic features' recovery. However, the majority of existing research predominantly utilizes a singular type of semantic information, such as text, images, or speech, to supervise and choose the generated source signals, which may not sufficiently encapsulate the comprehensive and accurate semantic information, and thus creating a performance bottleneck. In order to bridge this gap, in this paper, we propose and investigate a SemCom framework using multimodal information to supervise the generated image. To be specific, in this framework, we first extract semantic features at both the image and text levels utilizing the Convolutional Neural Network (CNN) architecture and the Contrastive Language-Image Pre-Training (CLIP) model before transmission. Then, we employ a generative diffusion model at the receiver to generate multiple images. In order to ensure the accurate extraction and facilitate high-fidelity image reconstruction, we select the "best" image with the minimum reconstruction errors by taking both the aided image and text semantic features into account. We further extend multimodal semantic communication (MMSemCom) system to the multiuser scenario for orthogonal transmission. Experimental results demonstrate that the proposed framework can not only achieve the enhanced fidelity and robustness in image transmission compared with existing communication systems but also sustain a high performance in the low signal-to-noise ratio (SNR) conditions.

eess.IV↗

arXiv2025

Joint Source-Channel Noise Adding with Adaptive Denoising for Diffusion-Based Semantic Communications

Semantic communication (SemCom) aims to convey the intended meaning of messages rather than merely transmitting bits, thereby offering greater efficiency and robustness, particularly in resource-constrained or noisy environments. In this paper, we propose a novel framework which is referred to as joint source-channel noise adding with adaptive denoising (JSCNA-AD) for SemCom based on a diffusion model (DM). Unlike conventional encoder-decoder designs, our approach intentionally incorporates the channel noise during transmission, effectively transforming the harmful channel noise into a constructive component of the diffusion-based semantic reconstruction process. Besides, we introduce an attention-based adaptive denoising mechanism, in which transmitted images are divided into multiple regions, and the number of denoising steps is dynamically allocated based on the semantic importance of each region. This design effectively balances the reception quality and the inference latency by prioritizing the critical semantic information. Extensive experiments demonstrate that our method significantly outperforms existing SemCom schemes under various noise conditions, underscoring the potential of diffusion-based models in next-generation communication systems.

cs.IT↗

arXiv2025

Contrastive Conditional Latent Diffusion for Audio-visual Segmentation

We propose a contrastive conditional latent diffusion model for audio-visual segmentation (AVS) to thoroughly investigate the impact of audio, where the correlation between audio and the final segmentation map is modeled to guarantee the strong correlation between them. To achieve semantic-correlated representation learning, our framework incorporates a latent diffusion model. The diffusion model learns the conditional generation process of the ground-truth segmentation map, resulting in ground-truth aware inference during the denoising process at the test stage. As our model is conditional, it is vital to ensure that the conditional variable contributes to the model output. We thus extensively model the contribution of the audio signal by minimizing the density ratio between the conditional probability of the multimodal data, e.g. conditioned on the audio-visual data, and that of the unimodal data, e.g. conditioned on the audio data only. In this way, our latent diffusion model via density ratio optimization explicitly maximizes the contribution of audio for AVS, which can then be achieved with contrastive learning as a constraint, where the diffusion part serves as the main objective to achieve maximum likelihood estimation, and the density ratio optimization part imposes the constraint. By adopting this latent diffusion model via contrastive learning, we effectively enhance the contribution of audio for AVS. The effectiveness of our solution is validated through experimental results on the benchmark dataset. Code and results are online via our project page: https://github.com/OpenNLPLab/DiffusionAVS.

cs.CV↗