arXiv subjects

Cong Wang

Publications and source records attributed to Cong Wang.

At least 145 records · Page 8Linked to original sources

arXiv2026

MeGA: Hybrid Mesh-Gaussian Head Avatar for High-Fidelity Rendering and Head Editing

Creating high-fidelity head avatars from multi-view videos is a core issue for many AR/VR applications. However, existing methods usually struggle to obtain high-quality renderings for all different head components simultaneously since they use one single representation to model components with drastically different characteristics (e.g., skin vs. hair). In this paper, we propose a Hybrid Mesh-Gaussian Head Avatar (MeGA) that models different head components with more suitable representations. Specifically, we select an enhanced FLAME mesh as our facial representation and predict a UV displacement map to provide per-vertex offsets for improved personalized geometric details. To achieve photorealistic renderings, we obtain facial colors using deferred neural rendering and disentangle neural textures into three meaningful parts. For hair modeling, we first build a static canonical hair using 3D Gaussian Splatting. A rigid transformation and an MLP-based deformation field are further applied to handle complex dynamic expressions. Combined with our occlusion-aware blending, MeGA generates higher-fidelity renderings for the whole head and naturally supports more downstream tasks. Experiments on the NeRSemble dataset demonstrate the effectiveness of our designs, outperforming previous state-of-the-art methods and supporting various editing functionalities, including hairstyle alteration and texture editing.

cs.CV↗

arXiv2026

RRPO: Robust Reward Policy Optimization for LLM-based Emotional TTS

Differentiable reinforcement learning (RL) frameworks like DiffRO offer a powerful approach for controllable text-to-speech (TTS), but are vulnerable to reward hacking, particularly for nuanced tasks like emotion control. The policy model can exploit a vanilla Reward Model (RM) by generating acoustic artifacts to achieve spurious rewards, but at the cost of degrading perceptual quality. To address this, we propose Robust Reward Policy Optimization (RRPO), a novel framework that employs a hybrid regularization scheme. This scheme develops a robust RM whose reward signal is more reliably aligned with human perception, compelling the policy to abandon detrimental shortcuts and instead learn the complex features of genuine emotions. Our ablation study confirms the enhanced robustness of our RM, as evidenced by its strong cross-lingual generalization. The subjective evaluation demonstrates that this robust RM effectively mitigates reward hacking, leading to significant improvements in both emotional expressiveness and naturalness over all baselines. Demo page: https://lrwinr.github.io/RRPO-CosyVoice.

cs.SD↗

arXiv2026

UniFit: Towards Universal Virtual Try-on with MLLM-Guided Semantic Alignment

Image-based virtual try-on (VTON) aims to synthesize photorealistic images of a person wearing specified garments. Despite significant progress, building a universal VTON framework that can flexibly handle diverse and complex tasks remains a major challenge. Recent methods explore multi-task VTON frameworks guided by textual instructions, yet they still face two key limitations: (1) semantic gap between text instructions and reference images, and (2) data scarcity in complex scenarios. To address these challenges, we propose UniFit, a universal VTON framework driven by a Multimodal Large Language Model (MLLM). Specifically, we introduce an MLLM-Guided Semantic Alignment Module (MGSA), which integrates multimodal inputs using an MLLM and a set of learnable queries. By imposing a semantic alignment loss, MGSA captures cross-modal semantic relationships and provides coherent and explicit semantic guidance for the generative process, thereby reducing the semantic gap. Moreover, by devising a two-stage progressive training strategy with a self-synthesis pipeline, UniFit is able to learn complex tasks from limited data. Extensive experiments show that UniFit not only supports a wide range of VTON tasks, including multi-garment and model-to-model try-on, but also achieves state-of-the-art performance. The source code and pretrained models are available at https://github.com/zwplus/UniFit.

cs.CV↗

arXiv2026

Hemlet: A Heterogeneous Compute-in-Memory Chiplet Architecture for Vision Transformers with Group-Level Parallelism

Vision Transformers (ViTs) have established new performance benchmarks in vision tasks such as image recognition and object detection. However, these advancements come with significant demands for memory and computational resources, presenting challenges for hardware deployment. Heterogeneous compute-in-memory (CIM) accelerators have emerged as a promising solution for enabling energy-efficient deployment of ViTs. Despite this potential, monolithic CIM-based designs face scalability issues due to the size limitations of a single chip. To address this challenge, emerging chiplet-based techniques offer a more scalable alternative. However, chiplet designs come with their own costs, as they introduce expensive communication, which can hinder improvements in throughput. This work introduces Hemlet, a heterogeneous CIM chiplet system designed to accelerate ViT workloads. Hemlet enables flexible resource scaling through the integration of heterogeneous analog CIM (ACIM), digital CIM (DCIM), and Intermediate Data Process (IDP) chiplets. To improve throughput while reducing communication overhead, it employs a group-level parallelism (GLP) mapping strategy and system-level dataflow optimization, achieving speedups ranging from 2.41x to 5.74x across various hardware configurations within the chiplet system. Our evaluation results show that Hemlet can reach a throughput of 9.56 TOPS with an energy efficiency of 4.98 TOPS/W.

cs.AR↗

arXiv2026

The LCLStream Ecosystem for Multi-Institutional Dataset Exploration

We describe a new end-to-end experimental data streaming framework designed from the ground up to support new types of applications -- AI training, extremely high-rate X-ray time-of-flight analysis, crystal structure determination with distributed processing, and custom data science applications and visualizers yet to be created. Throughout, we use design choices merging cloud microservices with traditional HPC batch execution models for security and flexibility. This project makes a unique contribution to the DOE Integrated Research Infrastructure (IRI) landscape. By creating a flexible, API-driven data request service, we address a significant need for high-speed data streaming sources for the X-ray science data analysis community. With the combination of data request API, mutual authentication web security framework, job queue system, high-rate data buffer, and complementary nature to facility infrastructure, the LCLStreamer framework has prototyped and implemented several new paradigms critical for future generation experiments.

cs.IR↗

arXiv2026

Winning the Pruning Gamble: A Unified Approach to Joint Sample and Token Pruning for Efficient Supervised Fine-Tuning

As supervised fine-tuning (SFT) evolves from a lightweight post-training step into a compute-intensive phase rivaling mid-training in scale, data efficiency has become critical for aligning large language models (LLMs) under tight budgets. Existing data pruning methods suffer from a fragmented design: they operate either at the sample level or the token level in isolation, failing to jointly optimize both dimensions. This disconnect leads to significant inefficiencies--high-value samples may still contain redundant tokens, while token-level pruning often discards crucial instructional or corrective signals embedded in individual examples. To address this bottleneck, we introduce the Error-Uncertainty (EU) Plane, a diagnostic framework that jointly characterizes the heterogeneous utility of training data across samples and tokens. Guided by this insight, we propose Quadrant-based Tuning (Q-Tuning), a unified framework that strategically coordinates sample pruning and token pruning. Q-Tuning employs a two-stage strategy: first, it performs sample-level triage to retain examples rich in informative misconceptions or calibration signals; second, it applies an asymmetric token-pruning policy, using a context-aware scoring mechanism to trim less salient tokens exclusively from misconception samples while preserving calibration samples in their entirety. Our method sets a new state of the art across five diverse benchmarks. Remarkably, on SmolLM2-1.7B, Q-Tuning achieves a +38\% average improvement over the full-data SFT baseline using only 12.5\% of the original training data. As the first dynamic pruning approach to consistently outperform full-data training, Q-Tuning provides a practical and scalable blueprint for maximizing data utilization in budget-constrained LLM SFT.

cs.CL↗

arXiv2026

vLLM-Omni: Fully Disaggregated Serving for Any-to-Any Multimodal Models

Any-to-any multimodal models that jointly handle text, images, video, and audio represent a significant advance in multimodal AI. However, their complex architectures (typically combining multiple autoregressive LLMs, diffusion transformers, and other specialized components) pose substantial challenges for efficient model serving. Existing serving systems are mainly tailored to a single paradigm, such as autoregressive LLMs for text generation or diffusion transformers for visual generation. They lack support for any-to-any pipelines that involve multiple interconnected model components. As a result, developers must manually handle cross-stage interactions, leading to huge performance degradation. We present vLLM-Omni, a fully disaggregated serving system for any-to-any models. vLLM-Omni features a novel stage abstraction that enables users to decompose complex any-to-any architectures into interconnected stages represented as a graph, and a disaggregated stage execution backend that optimizes resource utilization and throughput across stages. Each stage is independently served by an LLM or diffusion engine with per-stage request batching, flexible GPU allocation, and unified inter-stage connectors for data routing. Experimental results demonstrate that vLLM-Omni reduces job completion time (JCT) by up to 91.4% compared to baseline methods. The code is public available at https://github.com/vllm-project/vllm-omni.

cs.DC↗

arXiv2026

Who Transfers Safety? Identifying and Targeting Cross-Lingual Shared Safety Neurons

Multilingual safety remains significantly imbalanced, leaving non-high-resource (NHR) languages vulnerable compared to robust high-resource (HR) ones. Moreover, the neural mechanisms driving safety alignment remain unclear despite observed cross-lingual representation transfer. In this paper, we find that LLMs contain a set of cross-lingual shared safety neurons (SS-Neurons), a remarkably small yet critical neuronal subset that jointly regulates safety behavior across languages. We first identify monolingual safety neurons (MS-Neurons) and validate their causal role in safety refusal behavior through targeted activation and suppression. Our cross-lingual analyses then identify SS-Neurons as the subset of MS-Neurons shared between HR and NHR languages, serving as a bridge to transfer safety capabilities from HR to NHR domains. We observe that suppressing these neurons causes concurrent safety drops across NHR languages, whereas reinforcing them improves cross-lingual defensive consistency. Building on these insights, we propose a simple neuron-oriented training strategy that targets SS-Neurons based on language resource distribution and model architecture. Experiments demonstrate that fine-tuning this tiny neuronal subset outperforms state-of-the-art methods, significantly enhancing NHR safety while maintaining the model's general capabilities. The code and dataset will be available athttps://github.com/1518630367/SS-Neuron-Expansion.

cs.CV↗

arXiv2026

TraceRouter: Robust Safety for Large Foundation Models via Path-Level Intervention

Despite their capabilities, large foundation models (LFMs) remain susceptible to adversarial manipulation. Current defenses predominantly rely on the "locality hypothesis", suppressing isolated neurons or features. However, harmful semantics act as distributed, cross-layer circuits, rendering such localized interventions brittle and detrimental to utility. To bridge this gap, we propose \textbf{TraceRouter}, a path-level framework that traces and disconnects the causal propagation circuits of illicit semantics. TraceRouter operates in three stages: (1) it pinpoints a sensitive onset layer by analyzing attention divergence; (2) it leverages sparse autoencoders (SAEs) and differential activation analysis to disentangle and isolate malicious features; and (3) it maps these features to downstream causal pathways via feature influence scores (FIS) derived from zero-out interventions. By selectively suppressing these causal chains, TraceRouter physically severs the flow of harmful information while leaving orthogonal computation routes intact. Extensive experiments demonstrate that TraceRouter significantly outperforms state-of-the-art baselines, achieving a superior trade-off between adversarial robustness and general utility. Our code will be publicly released. WARNING: This paper contains unsafe model responses.

cs.CV↗

arXiv2026

Post-Norm can Resharpen Attention

Length Generalization is the essential capacity of autonomous agents to perform tasks in longer contexts than those encountered during training. To systematically study this feat, we test how well models can approximate the next token distributions in algorithmic tasks. This is to take into account the realistic possibility of multiple next tokens being legal. We present a prototypical benchmark for this line of study: in the Set Complement Task, the model needs to output a uniform distribution over tokens not in the input. We prove a theorem that states simple transformers can length generalize on this task, however, with performance degradation due to attention dispersion. A mechanistic reading of how dispersion takes effect lets us discover a remedy: Post-Norm can Resharpen Attention. We present experimental evidence to support this idea. We also show that Exponential Moving Averages can help the issue of noisy gradients that arises when many next tokens are legal. We validate the general applicability of our proposed methods on a suite of formal language experiments. Our source code will be available upon publication.

cs.LG↗

arXiv2026

Innovator-VL: A Multimodal Large Language Model for Scientific Discovery

We present Innovator-VL, a scientific multimodal large language model designed to advance understanding and reasoning across diverse scientific domains while maintaining excellent performance on general vision tasks. Contrary to the trend of relying on massive domain-specific pretraining and opaque pipelines, our work demonstrates that principled training design and transparent methodology can yield strong scientific intelligence with substantially reduced data requirements. (i) First, we provide a fully transparent, end-to-end reproducible training pipeline, covering data collection, cleaning, preprocessing, supervised fine-tuning, reinforcement learning, and evaluation, along with detailed optimization recipes. This facilitates systematic extension by the community. (ii) Second, Innovator-VL exhibits remarkable data efficiency, achieving competitive performance on various scientific tasks using fewer than five million curated samples without large-scale pretraining. These results highlight that effective reasoning can be achieved through principled data selection rather than indiscriminate scaling. (iii) Third, Innovator-VL demonstrates strong generalization, achieving competitive performance on general vision, multimodal reasoning, and scientific benchmarks. This indicates that scientific alignment can be integrated into a unified model without compromising general-purpose capabilities. Our practices suggest that efficient, reproducible, and high-performing scientific multimodal models can be built even without large-scale data, providing a practical foundation for future research.

cs.CV↗

arXiv2026

HYDRA: Breaking the Global Ordering Barrier in Multi-BFT Consensus

Multi-Byzantine Fault Tolerant (Multi-BFT) consensus, which runs multiple BFT instances in parallel, has recently emerged as a promising approach to overcome the leader bottleneck in classical BFT protocols. However, existing designs rely on a global ordering layer to serialize blocks across instances, an intuitive yet costly mechanism that constrains scalability, amplifies failure propagation, and complicates deployment. In this paper, we challenge this conventional wisdom. We present HYDRA, the first Multi-BFT consensus framework that eliminates global ordering altogether. HYDRA introduces an object-centric execution model that partitions transactions by their accessed objects, enabling concurrent yet deterministic execution across instances. To ensure consistency, HYDRA combines lightweight lock-based coordination with a deadlock resolution mechanism, achieving both scalability and correctness. We implement HYDRA and evaluate it on up to 128 replicas in both LAN and WAN environments. Experimental results show HYDRA outperforms several state-of-the-art Multi-BFT protocols in the presence of a straggler. These results demonstrate strong consistency and high performance by removing global ordering, opening a new direction toward scalable Multi-BFT consensus design.

cs.DC↗

arXiv2026

PrivTune: Efficient and Privacy-Preserving Fine-Tuning of Large Language Models via Device-Cloud Collaboration

With the rise of large language models, service providers offer language models as a service, enabling users to fine-tune customized models via uploaded private datasets. However, this raises concerns about sensitive data leakage. Prior methods, relying on differential privacy within device-cloud collaboration frameworks, struggle to balance privacy and utility, exposing users to inference attacks or degrading fine-tuning performance. To address this, we propose PrivTune, an efficient and privacy-preserving fine-tuning framework via Split Learning (SL). The key idea of PrivTune is to inject crafted noise into token representations from the SL bottom model, making each token resemble the $n$-hop indirect neighbors. PrivTune formulates this as an optimization problem to compute the optimal noise vector, aligning with defense-utility goals. On this basis, it then adjusts the parameters (i.e., mean) of the $d_χ$-Privacy noise distribution to align with the optimization direction and scales the noise according to token importance to minimize distortion. Experiments on five datasets (covering both classification and generation tasks) against three embedding inversion and three attribute inference attacks show that, using RoBERTa on the Stanford Sentiment Treebank dataset, PrivTune reduces the attack success rate to 10% with only a 3.33% drop in utility performance, outperforming state-of-the-art baselines.

cs.CR↗

arXiv2026

Multi-pathline flow visualization using PIV images

One of the oldest flow visualization techniques is through multiple pathlines generated by the movement of seeding particles spatially distributed in the flow. In the computerized era, particle images are used in quantitative measurements, such as particle image and particle tracking velocimetry (PIV and PTV). Here, we present several methods for post-processing raw particle images to generate enhanced flow visualization without a need for conducting additional experiments. Three post-processing methods will be shown: 1) controlling the exposure time, 2) color-coding temporal information, and 3) changing the frame of reference. We showcase how employing these three methods can highlight different flow features in three canonical flow cases: vortex ring, leading edge vortex, and turbulent boundary layer. In addition to the quantitative flow field, the multi-pathline visualization is expected to augment our ability to observe fluid flow from many different perspectives.

physics.flu-dyn↗

arXiv2025

FUSE-RSVLM: Feature Fusion Vision-Language Model for Remote Sensing

Large vision-language models (VLMs) exhibit strong performance across various tasks. However, these VLMs encounter significant challenges when applied to the remote sensing domain due to the inherent differences between remote sensing images and natural images. Existing remote sensing VLMs often fail to extract fine-grained visual features and suffer from visual forgetting during deep language processing. To address this, we introduce MF-RSVLM, a Multi-Feature Fusion Remote Sensing Vision--Language Model that effectively extracts and fuses visual features for RS understanding. MF-RSVLM learns multi-scale visual representations and combines global context with local details, improving the capture of small and complex structures in RS scenes. A recurrent visual feature injection scheme ensures the language model remains grounded in visual evidence and reduces visual forgetting during generation. Extensive experiments on diverse RS benchmarks show that MF-RSVLM achieves state-of-the-art or highly competitive performance across remote sensing classification, image captioning, and VQA tasks. Our code is publicly available at https://github.com/Yunkaidang/RSVLM.

cs.CV↗

arXiv2025

Kinetically accessible 1D magnetic chains of transition-metal chalcogenides and halides on van der Waals surfaces

One-dimensional (1D) chains offer unique opportunities for nanoelectronics and spintronics, yet their experimental realization remains challenging because 1D motifs are often thermodynamically disfavored relative to higher-dimensional phases. Here we present a high-throughput first-principles exploration of 1D single-atomic transition-metal chalcogenide and halide chains, screening 6,832 candidates constructed from binary combinations of 28 metals and 8 non-metals. To assess kinetic accessibility, we compare the formation energetics of 1D chains with competing two-dimensional polymorphs at the nucleation stage across relevant chemical-potential windows, using nucleation-stage thermodynamic selectivity as a proxy. This workflow identifies 183 kinetically accessible 1D chains. Interpretable machine-learning analysis reveals two simple stability descriptors as key drivers of 1D stabilization. The accessible chains exhibit diverse magnetic configurations with different magnetic characters. We further uncover their pronounced magnetoelastic couplings, exemplified by CrTe with giant magnetostriction reaching 5.93%. Finally, we show that selected metallic ferromagnetic chains retain robust edge magnetism on superconducting substrates, laying the groundwork for proximity-induced topological superconductivity and Majorana zero modes.

cond-mat.mtrl-sci↗

arXiv2025

Anomalous Hall effect and rich magnetic phase diagram of Mn$_{100-x}$Rh$_{x}$ epitaxial films

A series of Mn$_{100-x}$Rh$_x$ ($20 \le x \le 50$) thin films were epitaxially grown on the MgO substrate using magnetron sputtering technique, and were systematically investigated by magnetization, longitudinal electrical resistivity, and transverse Hall resistivity. After optimizing the growth conditions, phase-pure Mn$_{100-x}$Rh$_x$ films with a cubic CsCl-type structure were obtained, and their magnetic phase diagram was built. The manipulation of Rh content leads to a rich magnetic phase diagram, where three different regimes can be identified: for $x < 40$, Mn$_{100-x}$Rh$_x$ films undergo a ferromagnetic (FM) transition below $T_\mathrm{C} \approx$ 330-350 K; for $40 \le x \le 45$, in addition to the FM transition at $T_\mathrm{C} \approx$ 200 K, Mn$_{100-x}$Rh$_x$ films undergo a FM-to-antiferromagnetic (AFM) transition at $T_\mathrm{N} \approx$ 120 K; finally for $x > 45$, only one AFM transition at $T_\mathrm{N} \approx$ 150 K can be tracked. All the Mn$_{100-x}$Rh$_x$ films exhibit distinct anomalous Hall effect in their magnetically ordered state, which is most likely due to the intrinsic Berry-curvature mechanism. In addition, all the anomalous Hall transport properties, including the resistivity, conductivity, and angle exhibit a strong correlation with the magnetic properties of Mn$_{100-x}$Rh$_x$ films, which become most evident for $x$ = 35. Our systematic investigations suggest a strong correlation between magnetic properties and electronic band topology in Mn$_{100-x}$Rh$_x$ films, highlighting their great potential for AFM spintronics.

cond-mat.str-el↗

arXiv2025

Amulet: Fast TEE-Shielded Inference for On-Device Model Protection

On-device machine learning (ML) introduces new security concerns about model privacy. Storing valuable trained ML models on user devices exposes them to potential extraction by adversaries. The current mainstream solution for on-device model protection is storing the weights and conducting inference within Trusted Execution Environments (TEEs). However, due to limited trusted memory that cannot accommodate the whole model, most existing approaches employ a partitioning strategy, dividing a model into multiple slices that are loaded into the TEE sequentially. This frequent interaction between untrusted and trusted worlds dramatically increases inference latency, sometimes by orders of magnitude. In this paper, we propose Amulet, a fast TEE-shielded on-device inference framework for ML model protection. Amulet incorporates a suite of obfuscation methods specifically designed for common neural network architectures. After obfuscation by the TEE, the entire transformed model can be securely stored in untrusted memory, allowing the inference process to execute directly in untrusted memory with GPU acceleration. For each inference request, only two rounds of minimal-overhead interaction between untrusted and trusted memory are required to process input samples and output results. We also provide theoretical proof from an information-theoretic perspective that the obfuscated model does not leak information about the original weights. We comprehensively evaluated Amulet using diverse model architectures ranging from ResNet-18 to GPT-2. Our approach incurs inference latency only 2.8-4.8x that of unprotected models with negligible accuracy loss, achieving an 8-9x speedup over baseline methods that execute inference entirely within TEEs, and performing approximately 2.2x faster than the state-of-the-art obfuscation-based method.

cs.CR↗