arXiv subjects

Xu Wang

Publications and source records attributed to Xu Wang.

At least 181 records · Page 10Linked to original sources

arXiv2025

Diffusion LLMs Can Do Faster-Than-AR Inference via Discrete Diffusion Forcing

Diffusion Large Language Models (dLLMs) have emerged as a promising alternative to autoregressive (AR) LLMs for text generation, with the potential to decode multiple tokens in a single iteration. However, none of the existing open-source dLLMs have achieved superior inference speed over AR LLMs of similar size. This paper breaks this barrier based on a simple and effective strategy named discrete diffusion forcing (D2F). D2F equips dLLMs with two key capabilities: (1) block-wise autoregressive generation to enable KV cache utilization; (2) prediction of following tokens without requiring completion of prior blocks for inter-block parallel decoding. In this way, the vanilla dLLMs are refurbished into an AR-diffusion hybrid paradigm for efficient inference. D2F can be implemented with an asymmetric distillation process based on pre-trained dLLMs. We further propose a pipelined parallel decoding algorithm, which enables a trade-off between efficiency and efficacy. Empirically, D2F dLLMs achieve more than $\mathbf{2.5\times}$ inference speed than LLaMA3 and Qwen2.5 on GSM8K. Compared to vanilla dLLMs like LLaDA and Dream, the acceleration can be more than $\mathbf{50\times}$ while maintaining comparable output quality. The code is available at https://github.com/zhijie-group/Discrete-Diffusion-Forcing.

cs.LG↗

arXiv2025

TokenFlow: Unified Image Tokenizer for Multimodal Understanding and Generation

We present TokenFlow, a novel unified image tokenizer that bridges the long-standing gap between multimodal understanding and generation. Prior research attempt to employ a single reconstruction-targeted Vector Quantization (VQ) encoder for unifying these two tasks. We observe that understanding and generation require fundamentally different granularities of visual information. This leads to a critical trade-off, particularly compromising performance in multimodal understanding tasks. TokenFlow addresses this challenge through an innovative dual-codebook architecture that decouples semantic and pixel-level feature learning while maintaining their alignment via a shared mapping mechanism. This design enables direct access to both high-level semantic representations crucial for understanding tasks and fine-grained visual features essential for generation through shared indices. Our extensive experiments demonstrate TokenFlow's superiority across multiple dimensions. Leveraging TokenFlow, we demonstrate for the first time that discrete visual input can surpass LLaVA-1.5 13B in understanding performance, achieving a 7.2\% average improvement. For image reconstruction, we achieve a strong FID score of 0.63 at 384*384 resolution. Moreover, TokenFlow establishes state-of-the-art performance in autoregressive image generation with a GenEval score of 0.55 at 256*256 resolution, achieving comparable results to SDXL.

cs.CV↗

arXiv2025

Motion is the Choreographer: Learning Latent Pose Dynamics for Seamless Sign Language Generation

Sign language video generation requires producing natural signing motions with realistic appearances under precise semantic control, yet faces two critical challenges: excessive signer-specific data requirements and poor generalization. We propose a new paradigm for sign language video generation that decouples motion semantics from signer identity through a two-phase synthesis framework. First, we construct a signer-independent multimodal motion lexicon, where each gloss is stored as identity-agnostic pose, gesture, and 3D mesh sequences, requiring only one recording per sign. This compact representation enables our second key innovation: a discrete-to-continuous motion synthesis stage that transforms retrieved gloss sequences into temporally coherent motion trajectories, followed by identity-aware neural rendering to produce photorealistic videos of arbitrary signers. Unlike prior work constrained by signer-specific datasets, our method treats motion as a first-class citizen: the learned latent pose dynamics serve as a portable "choreography layer" that can be visually realized through different human appearances. Extensive experiments demonstrate that disentangling motion from identity is not just viable but advantageous - enabling both high-quality synthesis and unprecedented flexibility in signer personalization.

cs.CV↗

arXiv2025

Discovery of Itinerant Magnetic Domain Wall and Quasiparticle Boundary State in Spin-Density-Waves

Conventional magnetic domain walls are characterized by reorientation of local spins. However, what occurs at the boundary of itinerant magnets is largely unknown. Here using spin-sensitive scanning tunneling microscopy, we investigated the microscopic domain wall structure of the spin-density-wave (SDW) state in a prototypical itinerant antiferromagnet - chromium (Cr). At the boundary of two incommensurate SDW domains, we found the spins undergo finite-scale decay rather than reorientation. This generates a double-Q SDW state, which is further evidenced by an accompanying second-order charge modulation. In the commensurate SDW domains, a clear SDW energy gap is observed. Interestingly, the screw dislocations induced half vortex and anti-vortex of SDW, paired by antiphase domain wall. The spin density vanished at such antiphase domain walls. Remarkably, for the first time we observed the SDW quasiparticle states at the boundary, resembling the Andreev bound states in superconductors. These unique SDW boundary structures can be viewed as consequences of local interference of two SDWs, either with different Q or reversed phases. Our findings thus reveal a new type of domain wall distinct to that of local moment magnetism, with a mechanism rooted in the itinerant nature of SDW.

cond-mat.str-el↗

arXiv2025

Text2Lip: Progressive Lip-Synced Talking Face Generation from Text via Viseme-Guided Rendering

Generating semantically coherent and visually accurate talking faces requires bridging the gap between linguistic meaning and facial articulation. Although audio-driven methods remain prevalent, their reliance on high-quality paired audio visual data and the inherent ambiguity in mapping acoustics to lip motion pose significant challenges in terms of scalability and robustness. To address these issues, we propose Text2Lip, a viseme-centric framework that constructs an interpretable phonetic-visual bridge by embedding textual input into structured viseme sequences. These mid-level units serve as a linguistically grounded prior for lip motion prediction. Furthermore, we design a progressive viseme-audio replacement strategy based on curriculum learning, enabling the model to gradually transition from real audio to pseudo-audio reconstructed from enhanced viseme features via cross-modal attention. This allows for robust generation in both audio-present and audio-free scenarios. Finally, a landmark-guided renderer synthesizes photorealistic facial videos with accurate lip synchronization. Extensive evaluations show that Text2Lip outperforms existing approaches in semantic fidelity, visual realism, and modality robustness, establishing a new paradigm for controllable and flexible talking face generation. Our project homepage is https://plyon1.github.io/Text2Lip/.

cs.CV↗

arXiv2025

YOLO-FireAD: Efficient Fire Detection via Attention-Guided Inverted Residual Learning and Dual-Pooling Feature Preservation

Fire detection in dynamic environments faces continuous challenges, including the interference of illumination changes, many false detections or missed detections, and it is difficult to achieve both efficiency and accuracy. To address the problem of feature extraction limitation and information loss in the existing YOLO-based models, this study propose You Only Look Once for Fire Detection with Attention-guided Inverted Residual and Dual-pooling Downscale Fusion (YOLO-FireAD) with two core innovations: (1) Attention-guided Inverted Residual Block (AIR) integrates hybrid channel-spatial attention with inverted residuals to adaptively enhance fire features and suppress environmental noise; (2) Dual Pool Downscale Fusion Block (DPDF) preserves multi-scale fire patterns through learnable fusion of max-average pooling outputs, mitigating small-fire detection failures. Extensive evaluation on two public datasets shows the efficient performance of our model. Our proposed model keeps the sum amount of parameters (1.45M, 51.8% lower than YOLOv8n) (4.6G, 43.2% lower than YOLOv8n), and mAP75 is higher than the mainstream real-time object detection models YOLOv8n, YOL-Ov9t, YOLOv10n, YOLO11n, YOLOv12n and other YOLOv8 variants 1.3-5.5%. For more details, please visit our repository: https://github.com/JEFfersusu/YOLO-FireAD

cs.CV↗

arXiv2025

Precise Measurement of $^{216}$Po Half-life with Exact Parent-daughter Pairing in PandaX-4T

We report a precise measurement of $^{216}\rm Po$ half-life using the PandaX-4T liquid xenon time projection chamber (TPC). $^{220}\rm Rn $, emanating from a $^{228}\rm Th $ calibration source, is injected to the detector and undergoes successive $α$ decays, first to $^{216}\rm Po$ and then to $^{212}\rm Pb$. PandaX-4T detector measures the 5-dimensional (5D) information of each decay, including time, energy, and 3-dimensional positions. Therefore, we can identify the $^{220}\rm Rn $ and $^{216}\rm Po$ decay events and pair them exactly to extract the lifetime of each $^{216}\rm Po$. With a large data set and high-precision $^{220}\rm $Rn-$^{216}\rm $Po pairing technique, we measure the $^{216}\rm Po$ half-life to be $143.7\pm0.5$ ms, which is the most precise result to date and agrees with previously published values. The leading precision of this measurement demonstrates the power of 5D calorimeter and the potential of exact parent-daughter pairing in the xenon TPC.

nucl-ex↗

arXiv2025

SoK: Semantic Privacy in Large Language Models

As Large Language Models (LLMs) are increasingly deployed in sensitive domains, traditional data privacy measures prove inadequate for protecting information that is implicit, contextual, or inferable - what we define as semantic privacy. This Systematization of Knowledge (SoK) introduces a lifecycle-centric framework to analyze how semantic privacy risks emerge across input processing, pretraining, fine-tuning, and alignment stages of LLMs. We categorize key attack vectors and assess how current defenses, such as differential privacy, embedding encryption, edge computing, and unlearning, address these threats. Our analysis reveals critical gaps in semantic-level protection, especially against contextual inference and latent representation leakage. We conclude by outlining open challenges, including quantifying semantic leakage, protecting multimodal inputs, balancing de-identification with generation quality, and ensuring transparency in privacy enforcement. This work aims to inform future research on designing robust, semantically aware privacy-preserving techniques for LLMs.

cs.CR↗

arXiv2025

The Giant Radio Array for Neutrino Detection (GRAND) Collaboration -- Contributions to the 39th International Cosmic Ray Conference (ICRC 2025)

The Giant Radio Array for Neutrino Detection (GRAND) is an envisioned observatory of ultra-high-energy particles of cosmic origin, with energies in excess of 100 PeV. GRAND uses large surface arrays of antennas to look for the radio emission from extensive air showers that are triggered by the interaction of ultra-high-energy cosmic rays, gamma rays, and neutrinos in the atmosphere or underground. In particular, for ultra-high-energy neutrinos, the future final phase of GRAND aims to be sensitive enough to detect them in spite of their plausibly tiny flux. Three prototype GRAND radio arrays have been in operation since 2023: GRANDProto300, in China, GRAND@Auger, in Argentina, and GRAND@Nançay, in France. Their goals are to field-test the GRAND detection units, understand the radio background to which they are exposed, and develop tools for diagnostic, data gathering, and data analysis. This list of contributions to the 39th International Cosmic Ray Conference (ICRC 2025) presents an overview of GRAND, in its present and future incarnations, and a first look at data collected by GRANDProto300 and GRAND@Auger, including the first cosmic-ray candidates detected by them.

astro-ph.IM↗

arXiv2025

CAN-Trace Attack: Exploit CAN Messages to Uncover Driving Trajectories

Driving trajectory data remains vulnerable to privacy breaches despite existing mitigation measures. Traditional methods for detecting driving trajectories typically rely on map-matching the path using Global Positioning System (GPS) data, which is susceptible to GPS data outage. This paper introduces CAN-Trace, a novel privacy attack mechanism that leverages Controller Area Network (CAN) messages to uncover driving trajectories, posing a significant risk to drivers' long-term privacy. A new trajectory reconstruction algorithm is proposed to transform the CAN messages, specifically vehicle speed and accelerator pedal position, into weighted graphs accommodating various driving statuses. CAN-Trace identifies driving trajectories using graph-matching algorithms applied to the created graphs in comparison to road networks. We also design a new metric to evaluate matched candidates, which allows for potential data gaps and matching inaccuracies. Empirical validation under various real-world conditions, encompassing different vehicles and driving regions, demonstrates the efficacy of CAN-Trace: it achieves an attack success rate of up to 90.59% in the urban region, and 99.41% in the suburban region.

cs.CR↗

arXiv2025

Feature Extraction and Steering for Enhanced Chain-of-Thought Reasoning in Language Models

Large Language Models (LLMs) demonstrate the ability to solve reasoning and mathematical problems using the Chain-of-Thought (CoT) technique. Expanding CoT length, as seen in models such as DeepSeek-R1, significantly enhances this reasoning for complex problems, but requires costly and high-quality long CoT data and fine-tuning. This work, inspired by the deep thinking paradigm of DeepSeek-R1, utilizes a steering technique to enhance the reasoning ability of an LLM without external datasets. Our method first employs Sparse Autoencoders (SAEs) to extract interpretable features from vanilla CoT. These features are then used to steer the LLM's internal states during generation. Recognizing that many LLMs do not have corresponding pre-trained SAEs, we further introduce a novel SAE-free steering algorithm, which directly computes steering directions from the residual activations of an LLM, obviating the need for an explicit SAE. Experimental results demonstrate that both our SAE-based and subsequent SAE-free steering algorithms significantly enhance the reasoning capabilities of LLMs.

cs.CL↗

arXiv2025

Finite Sample Analysis of Subspace Identification for Stochastic Systems

The subspace identification method (SIM) has become a widely adopted approach for the identification of discrete-time linear time-invariant (LTI) systems. In this paper, we derive finite sample high-probability error bounds for the system matrices $A,C$, the Kalman filter gain $K$ and the estimation of system poles. Specifically, we demonstrate that, ignoring the logarithmic factors, for an $n$-dimensional LTI system with no external inputs, the estimation error of these matrices decreases at a rate of at least $ \mathcal{O}(\sqrt{1/N}) $, while the estimation error of the system poles decays at a rate of at least $ \mathcal{O}(N^{-1/2n}) $, where $ N $ represents the number of sample trajectories. Furthermore, we reveal that achieving a constant estimation error requires a super-polynomial sample size in $n/m $, where $n/m$ denotes the state-to-output dimension ratio. Finally, numerical experiments are conducted to validate the non-asymptotic results.

eess.SY↗

arXiv2025

Inverse random potential scattering for the polyharmonic wave equation using far-field patterns

This paper addresses the inverse scattering problem of a random potential associated with the polyharmonic wave equation in two and three dimensions. The random potential is represented as a centered complex-valued generalized microlocally isotropic Gaussian random field, where its covariance and relation operators are characterized as conventional pseudo-differential operators. Regarding the direct scattering problem, the well-posedness is established in the distribution sense for sufficiently large wavenumbers through analysis of the corresponding Lippmann--Schwinger integral equation. Furthermore, in the context of the inverse scattering problem, the uniqueness is attained in recovering the microlocal strengths of both the covariance and relation operators of the random potential. Notably, this is accomplished with only a single realization of the backscattering far-field patterns averaged over the high-frequency band.

math.AP↗

arXiv2025

Deep Learning based Joint Geometry and Attribute Up-sampling for Large-Scale Colored Point Clouds

Colored point cloud, which includes geometry and attribute components, is a mainstream representation enabling realistic and immersive 3D applications. To generate large-scale and denser colored point clouds, we propose a deep learning-based Joint Geometry and Attribute Up-sampling (JGAU) method that learns to model both geometry and attribute patterns while leveraging spatial attribute correlations. First, we establish and release a large-scale dataset for colored point cloud up-sampling called SYSU-PCUD, containing 121 large-scale colored point clouds with diverse geometry and attribute complexities across six categories and four sampling rates. Second, to improve the quality of up-sampled point clouds, we propose a deep learning-based JGAU framework that jointly up-samples geometry and attributes. It consists of a geometry up-sampling network and an attribute up-sampling network, where the latter leverages the up-sampled auxiliary geometry to model neighborhood correlations of the attributes. Third, we propose two coarse attribute up-sampling methods, Geometric Distance Weighted Attribute Interpolation (GDWAI) and Deep Learning-based Attribute Interpolation (DLAI), to generate coarse up-sampled attributes for each point. Then, an attribute enhancement module is introduced to refine these up-sampled attributes and produce high-quality point clouds by further exploiting intrinsic attribute and geometry patterns. Extensive experiments show that the Peak Signal-to-Noise Ratio (PSNR) achieved by the proposed JGAU method is 33.90 decibels, 32.10 decibels, 31.10 decibels, and 30.39 decibels for up-sampling rates of 4 times, 8 times, 12 times, and 16 times, respectively. Compared to state-of-the-art methods, JGAU achieves average PSNR gains of 2.32 decibels, 2.47 decibels, 2.28 decibels, and 2.11 decibels at these four up-sampling rates, demonstrating significant improvement.

cs.CV↗

arXiv2025

XVerse: Consistent Multi-Subject Control of Identity and Semantic Attributes via DiT Modulation

Achieving fine-grained control over subject identity and semantic attributes (pose, style, lighting) in text-to-image generation, particularly for multiple subjects, often undermines the editability and coherence of Diffusion Transformers (DiTs). Many approaches introduce artifacts or suffer from attribute entanglement. To overcome these challenges, we propose a novel multi-subject controlled generation model XVerse. By transforming reference images into offsets for token-specific text-stream modulation, XVerse allows for precise and independent control for specific subject without disrupting image latents or features. Consequently, XVerse offers high-fidelity, editable multi-subject image synthesis with robust control over individual subject characteristics and semantic attributes. This advancement significantly improves personalized and complex scene generation capabilities.

cs.CV↗

arXiv2025

DeckFlow: Iterative Specification on a Multimodal Generative Canvas

Generative AI promises to allow people to create high-quality personalized media. Although powerful, we identify three fundamental design problems with existing tooling through a literature review. We introduce a multimodal generative AI tool, DeckFlow, to address these problems. First, DeckFlow supports task decomposition by allowing users to maintain multiple interconnected subtasks on an infinite canvas populated by cards connected through visual dataflow affordances. Second, DeckFlow supports a specification decomposition workflow where an initial goal is iteratively decomposed into smaller parts and combined using feature labels and clusters. Finally, DeckFlow supports generative space exploration by generating multiple prompt and output variations, presented in a grid, that can feed back recursively into the next design iteration. We evaluate DeckFlow for text-to-image generation against a state-of-practice conversational AI baseline for image generation tasks. We then add audio generation and investigate user behaviors in a more open-ended creative setting with text, image, and audio outputs.

cs.HC↗

arXiv2025

Towards Understanding Fine-Tuning Mechanisms of LLMs via Circuit Analysis

Fine-tuning significantly improves the performance of Large Language Models (LLMs), yet its underlying mechanisms remain poorly understood. This paper aims to provide an in-depth interpretation of the fine-tuning process through circuit analysis, a popular tool in Mechanistic Interpretability (MI). Unlike previous studies (Prakash et al. 2024; Chhabra et al. 2024) that focus on tasks where pre-trained models already perform well, we develop a set of mathematical tasks where fine-tuning yields substantial performance gains, which are closer to the practical setting. In our experiments, we identify circuits at various checkpoints during fine-tuning and examine the interplay between circuit analysis, fine-tuning methods, and task complexities. First, we find that while circuits maintain high node similarity before and after fine-tuning, their edges undergo significant changes, in contrast to prior work that shows circuits only add some additional components after fine-tuning. Based on these observations, we develop a circuit-aware Low-Rank Adaptation (LoRA) method, which assigns ranks to layers based on edge changes in the circuits. Experimental results demonstrate that our circuit-based LoRA algorithm achieves an average performance improvement of 2.46% over standard LoRA with similar parameter sizes. Furthermore, we explore how combining circuits from subtasks can enhance fine-tuning in compositional tasks, providing new insights into the design of such tasks and deepening the understanding of circuit dynamics and fine-tuning mechanisms.

cs.CL↗

arXiv2025

SignAligner: Harmonizing Complementary Pose Modalities for Coherent Sign Language Generation

Sign language generation aims to produce diverse sign representations based on spoken language. However, achieving realistic and naturalistic generation remains a significant challenge due to the complexity of sign language, which encompasses intricate hand gestures, facial expressions, and body movements. In this work, we introduce PHOENIX14T+, an extended version of the widely-used RWTH-PHOENIX-Weather 2014T dataset, featuring three new sign representations: Pose, Hamer and Smplerx. We also propose a novel method, SignAligner, for realistic sign language generation, consisting of three stages: text-driven pose modalities co-generation, online collaborative correction of multimodality, and realistic sign video synthesis. First, by incorporating text semantics, we design a joint sign language generator to simultaneously produce posture coordinates, gesture actions, and body movements. The text encoder, based on a Transformer architecture, extracts semantic features, while a cross-modal attention mechanism integrates these features to generate diverse sign language representations, ensuring accurate mapping and controlling the diversity of modal features. Next, online collaborative correction is introduced to refine the generated pose modalities using a dynamic loss weighting strategy and cross-modal attention, facilitating the complementarity of information across modalities, eliminating spatiotemporal conflicts, and ensuring semantic coherence and action consistency. Finally, the corrected pose modalities are fed into a pre-trained video generation network to produce high-fidelity sign language videos. Extensive experiments demonstrate that SignAligner significantly improves both the accuracy and expressiveness of the generated sign videos.

cs.CV↗