arXiv subjects

Zhi Xu

Publications and source records attributed to Zhi Xu.

At least 19 recordsLinked to original sources

arXiv2026

WA-JEPA: Rethinking the Video JEPA Paradigm for World-Action Modeling in Autonomous Driving

Video Joint Embedding Predictive Architecture (V-JEPA) learns powerful spatiotemporal representations from video through self-supervised latent feature prediction. However, V-JEPA is built around random-mask completion and deterministic regression, making it fundamentally ill-suited for autonomous driving planning that demands future-directed prediction tightly coupled with action. To address this, we rethink the V-JEPA paradigm and present WA-JEPA, a V-JEPA-native world-action model designed for autonomous driving planning. Instead of random spatiotemporal masking, WA-JEPA employs hybrid future-masked pre-training, where the model infers future latents from observed context. Departing from deterministic regression, we recast future prediction as conditional flow matching over latent futures, which substantially improves the model's ability to generate plausible future latents for downstream planning. Finally, a joint future-action predictor is proposed to denoise future scene tokens and ego trajectories together in a unified spatiotemporal latent space, allowing action supervision to directly shape planning-relevant world representations. Pre-trained on nuPlan videos and fine-tuned on NAVSIM, WA-JEPA reaches 91.7 EPDMS on NAVSIM-v2, surpassing the strongest end-to-end and world-action baselines by 1.6 and 1.3 EPDMS, and, without HUGSIM-specific fine-tuning, attains the best HD-Score of 0.4462 on the closed-loop HUGSIM benchmark under the same evaluation protocol. These results validate V-JEPA-native world-action modeling as a powerful and scalable paradigm for autonomous driving planning. Code is available at https://github.com/AFARI-Research/WA-JEPA.

cs.CV↗

arXiv2026

Research Progress on Solar Small-Scale Dynamo

The small-scale solar dynamo theory, as the core mechanism explaining the origin of the persistent, disperse weak magnetic field in the quiet Sun regions, has made significant progress over the past three decades in the fields of observation, theory, and simulation. Breakthrough observations from high-resolution space-based and ground-based telescopes have revealed that the quiet Sun is ubiquitously populated by highly dynamic, mixed-polarity and possibly predominantly horizontal magnetic structures with complex topology. These observations confirm that their total magnetic flux is substantial, with a high and widely distributed magnetic flux emergence rate, strongly suggesting a local dynamo effect independent of the solar cycle and driven by intense turbulence and convection. Theoretical studies indicate that even in the challenging low magnetic Prandtl number ( Pm << 1) environment of the solar photosphere and convection zone, turbulent motions can self-excitedly convert kinetic energy into magnetic energy through the stretching, folding, and twisting of magnetic field lines. MHD simulations have successfully reproduced observed features, demonstrating that a pure small-scale dynamo can operate efficiently and sustain magnetic fields even in an open, stratified solar environment incorporating realistic physical processes. Current research strongly suggests that small-scale magnetic fields constitute the majority of the magnetic energy in the quiet Sun and also influence coronal heating, solar wind acceleration, and radiation distribution. This article conducts a literature review centered on observations, theoretical models, and numerical simulations of the small-scale dynamo, organizing and discussing the relevant research history and progress. Finally, it summarizes the content and provides an outlook on future research from multiple perspectives.

astro-ph.SR↗

arXiv2026

Lung-R1: A Knowledge Graph-Guided LLM for Pulmonary Diagnostic Reasoning

Diagnosing pulmonary diseases requires integrating heterogeneous evidence amid phenotypic variability and cross-disease overlap. Although large language models (LLMs) have shown progress on pulmonary knowledge question answering (QA) and information-processing tasks, reliable pulmonary diagnosis requires patient-specific, relation-aware reasoning over electronic medical record (EMR) evidence rather than isolated knowledge recall. We define this gap between pulmonary knowledge and case-level diagnostic reasoning as the Pulmonary Knowledge-to-Diagnosis Gap. To address it, we introduce LungKG, the first structured pulmonary knowledge graph for diagnostic knowledge organization and record-grounded reasoning. LungKG contains 59,038 nodes and 164,308 edges across 15 entity types and 112 relation types, serving as both a reusable pulmonary knowledge resource and the foundation for LungKG-guided model adaptation. Built on LungKG, we propose Lung-R1, a LungKG-guided pulmonary LLM trained through KG-constrained reasoning-chain construction and KG-guided reinforcement learning. In a 20-system evaluation, Lung-R1-14B achieves state-of-the-art performance across Choice, Pulmonary-QA, and EMR Diagnosis, reaching an EMR Diagnosis score of 4.3583 and surpassing the strongest non-Lung-R1 baseline by 0.1476 points. These results demonstrate the value of LungKG-guided training for EMR-based pulmonary diagnosis.

cs.AI↗

arXiv2026

SEP-Attack: A Simple and Effective Paradigm for Transfer-Based Textual Adversarial Attack

Despite the strong performance of deep neural networks in modern Web and language applications, they remain vulnerable to adversarial attacks, especially transferable attacks that generate adversarial examples using surrogate models without accessing the victim model. Transferable attacks in the text domain are still under-explored, with only a few studies addressing this challenging issue, often with suboptimal results due to equal treatment of submodels or inaccurate estimation of importance scores. To address these challenges, we propose a simple yet effective paradigm for transfer-based textual adversarial attack, named SEP-Attack. Specifically, we employ the Determinantal Point Process (DPP) to generate diverse surrogate ensemble weights, representing the transferability of submodels. Using these weights, we introduce a new metric to evaluate prediction confidence scores, which in turn are used to calculate word importance scores and generate adversarial candidates. Finally, we quantify the transferability score for each candidate and select the top ones as the final transferable adversarial examples. Experiments conducted on four datasets and two real-world APIs validate the efficacy of SEP-Attack, significantly outperforming state-of-the-art baselines.

cs.CL↗

arXiv2026

ChainFlow-VLA: Causal Flow Planning with Vision-Language Models

Current end-to-end autonomous driving systems are fundamentally limited by a mismatch between temporal causal reasoning and global trajectory consistency. Autoregressive (AR) models capture interaction-aware temporal dependencies via causal factorization, but their step-wise decoding leads to error accumulation and suboptimal global structure. In contrast, diffusion models optimize trajectories globally but lack explicit causal constraints, making them unreliable in interactive and safety-critical scenarios. This dichotomy reveals a deeper issue: existing methods treat causal modeling and global optimization as separate paradigms, without a principled way to unify them within a single trajectory distribution. To address this, we propose ChainFlow-VLA, which unifies causal generation and global refinement within a unified probabilistic framework. We formulate planning as a mixture over AR-induced modes and learn Vision-Language Model (VLM)-conditioned residual distributions over these modes. An autoregressive generator (Chain) produces a discrete set of causal trajectory modes, followed by a diffusion-based refiner (Flow) that leverages VLM hidden states as semantic priors to perform mode-conditioned correction in residual space while preserving causal structure. This straightforward conditioning seamlessly injects high-level scene understanding into fine-grained trajectory adjustments. Experiments demonstrate that ChainFlow-VLA achieves robust planning in ambiguous and long-tail scenarios, achieving a state-of-the-art score of 94.85 on the NAVSIM v1 leaderboard, matching human-level performance (94.8). Code will be available at https://github.com/AFARI-Research/ChainFlow-VLA.

cs.CV↗

arXiv2026

CoWorld-VLA: Thinking in a Multi-Expert World Model for Autonomous Driving

Vision-Language-Action (VLA) models have emerged as a promising paradigm for end-to-end autonomous driving. However, existing reasoning mechanisms still struggle to provide planning-oriented intermediate representations: textual Chain-of-Thought (CoT) fails to preserve continuous spatiotemporal structure, while latent world reasoning remains difficult to use as a direct condition for action generation. In this paper, we propose CoWorld-VLA, a multi-expert world reasoning framework for autonomous driving, where world representations serve as explicit conditions to guide action planning. CoWorld-VLA extracts complementary world information through multi-source supervision and encodes it into expert tokens within the VLA, thereby providing planner-accessible conditioning signals. Specifically, we construct four types of tokens: semantic interaction, geometric structure, dynamic evolution, and ego trajectory tokens, which respectively model interaction intent, spatial structure, future temporal dynamics, and behavioral goals. During action generation, CoWorld-VLA employs a diffusion-based hierarchical multi-expert fusion planner, which is coupled with scene context throughout the joint denoising process to generate continuous ego trajectories. Experiments on NAVSIM v1 & v2 demonstrate future-scene modeling capability and strong planning performance, including collision avoidance and trajectory accuracy. Ablation studies further validate the complementarity of expert tokens and their effectiveness as planning conditions for action generation. Code will be available at https://github.com/AFARI-Research/CoWorld-VLA.

cs.CV↗

arXiv2026

NoisyCausal: A Benchmark for Evaluating Causal Reasoning Under Structured Noise

Causal reasoning in natural language requires identifying relevant variables, understanding their interactions, and reasoning about effects and interventions, often under noisy or ambiguous conditions. While large language models (LLMs) exhibit strong general reasoning abilities, they struggle to disentangle correlation from causation, particularly when observations are partially incorrect or irrelevant information is present. In this work, we introduce NoisyCausal, a new benchmark designed to evaluate causal reasoning under structured noise. Each instance is generated from a ground-truth causal graph and contextualized with a natural language scenario by injecting controllable forms of noise, such as irrelevant distractors, value perturbations, confounding, and partial observability. Moreover, we propose a modular reasoning framework that combines LLMs with explicit causal structure to address these challenges. Our method prompts the LLM to extract variables, construct a causal graph from context, and then reformulates the reasoning task as a structured prompt grounded in this graph. Rather than relying on statistical patterns alone, the LLM is guided by symbolic structure, enabling more interpretable and robust inference. Experimental results show that our method significantly outperforms standard prompting and reasoning baselines on NoisyCausal. Furthermore, it generalizes well to external benchmarks such as Cladder without task-specific tuning. Our findings highlight the importance of combining causal abstractions with language-driven reasoning to achieve faithful and robust causal understanding in LLMs.

cs.CL↗

arXiv2026

HQA-VLAttack: Towards High Quality Adversarial Attack on Vision-Language Pre-Trained Models

Black-box adversarial attack on vision-language pre-trained models is a practical and challenging task, as text and image perturbations need to be considered simultaneously, and only the predicted results are accessible. Research on this problem is in its infancy, and only a handful of methods are available. Nevertheless, existing methods either rely on a complex iterative cross-search strategy, which inevitably consumes numerous queries, or only consider reducing the similarity of positive image-text pairs but ignore that of negative ones, which will also be implicitly diminished, thus inevitably affecting the attack performance. To alleviate the above issues, we propose a simple yet effective framework to generate high-quality adversarial examples on vision-language pre-trained models, named HQA-VLAttack, which consists of text and image attack stages. For text perturbation generation, it leverages the counter-fitting word vector to generate the substitute word set, thus guaranteeing the semantic consistency between the substitute word and the original word. For image perturbation generation, it first initializes the image adversarial example via the layer-importance guided strategy, and then utilizes contrastive learning to optimize the image adversarial perturbation, which ensures that the similarity of positive image-text pairs is decreased while that of negative image-text pairs is increased. In this way, the optimized adversarial images and texts are more likely to retrieve negative examples, thereby enhancing the attack success rate. Experimental results on three benchmark datasets demonstrate that HQA-VLAttack significantly outperforms strong baselines in terms of attack success rate.

cs.CV↗

arXiv2026

Superconductivity and competing orders in honeycomb $t$-$J$ model: interplay of lattice geometry and next-nearest-neighbor hopping

We investigate the extended $t$-$J$ model on honeycomb lattices with next-nearest-neighbor (NNN) electron hopping $t'$ and superexchange coupling $J'=(t'/t)^2 J$ using large-scale density-matrix renormalization group (DMRG) simulations and slave-boson mean-field theory (SBMFT). By systematically varying $t'$ and cylinder geometries, our DMRG results reveal several competing phases with distinct charge and superconducting (SC) properties. On YC4-0 cylinders possessing bonds lying along $\vec{e}_y$ direction, the ground state of doped models exhibits pronounced quasi-long-range $d$-wave SC with coexisting armchair-oriented stripes (a-stripe) across a broad range of $t'$. Notably, the SC Luttinger exponent has a non-monotonic dependence on $t'$, showing an optimal $t'_{op}\sim0.4$ for dominant SC. Conversely, XC cylinders host a competing long-range zigzag stripes phase without SC for $t'>0.5$, highlighting the role of boundary geometry in stabilizing distinct competing phases in DMRG. To elucidate the stability of all these competing phases in 2D limit, we employ SBMFT and identify the a-stripe as the stable configuration across most of phase diagram, with a transition to uniform nematic $d$-wave SC at large $t'$ for $\delta=1/8$. The combined results from two complementary approaches suggest a robust $t'$-induced SC phase that might remain stable in doped extended $t$-$J$ model on the honeycomb lattice.

cond-mat.str-el↗

arXiv2026

TAO-Attack: Toward Advanced Optimization-Based Jailbreak Attacks for Large Language Models

Large language models (LLMs) have achieved remarkable success across diverse applications but remain vulnerable to jailbreak attacks, where attackers craft prompts that bypass safety alignment and elicit unsafe responses. Among existing approaches, optimization-based attacks have shown strong effectiveness, yet current methods often suffer from frequent refusals, pseudo-harmful outputs, and inefficient token-level updates. In this work, we propose TAO-Attack, a new optimization-based jailbreak method. TAO-Attack employs a two-stage loss function: the first stage suppresses refusals to ensure the model continues harmful prefixes, while the second stage penalizes pseudo-harmful outputs and encourages the model toward more harmful completions. In addition, we design a direction-priority token optimization (DPTO) strategy that improves efficiency by aligning candidates with the gradient direction before considering update magnitude. Extensive experiments on multiple LLMs demonstrate that TAO-Attack consistently outperforms state-of-the-art methods, achieving higher attack success rates and even reaching 100\% in certain scenarios.

cs.CL↗

arXiv2026

Kinetics of Stacking Order Evolution During Heterogeneous Ice Formation

The selection of stacking order in a broad range of close-packed polymorphic materials remains a challenging enigma. Using in situ cryogenic transmission electron microscopy, we uncover the atomistic mechanisms governing the vapour deposition growth of ice. We find that the heterogeneous ice nucleation and growth undergoes recrystallization accompanied by bifurcation, reflecting a coherent epitaxial transition from a cubic-ice embryonic core to hexagonal-ice prismatic dendrites, with intermediate stacking-disordered layers serving as a dynamic fluctuating bridge. Supported by molecular dynamics simulations, these phenomena are attributed to a surface-constrained, symmetry-breaking crystallization preference aligned with the principle of minimizing free energy. Our results highlight the critical role of the combined effects of surface and symmetry in shaping ice crystallization, providing fresh insights into crystal growth mechanisms and guiding principles for the design of advanced materials.

cond-mat.mtrl-sci↗

arXiv2025

Vision-G1: Towards General Vision Language Reasoning with Multi-Domain Data Curation

Despite their success, current training pipelines for reasoning VLMs focus on a limited range of tasks, such as mathematical and logical reasoning. As a result, these models face difficulties in generalizing their reasoning capabilities to a wide range of domains, primarily due to the scarcity of readily available and verifiable reward data beyond these narrowly defined areas. Moreover, integrating data from multiple domains is challenging, as the compatibility between domain-specific datasets remains uncertain. To address these limitations, we build a comprehensive RL-ready visual reasoning dataset from 46 data sources across 8 dimensions, covering a wide range of tasks such as infographic, mathematical, spatial, cross-image, graphic user interface, medical, common sense and general science. We propose an influence function based data selection and difficulty based filtering strategy to identify high-quality training samples from this dataset. Subsequently, we train the VLM, referred to as Vision-G1, using multi-round RL with a data curriculum to iteratively improve its visual reasoning capabilities. Our model achieves state-of-the-art performance across various visual reasoning benchmarks, outperforming similar-sized VLMs and even proprietary models like GPT-4o and Gemini-1.5 Flash. The model, code and dataset are publicly available at https://github.com/yuh-zha/Vision-G1.

cs.CV↗

arXiv2024

CusConcept: Customized Visual Concept Decomposition with Diffusion Models

Enabling generative models to decompose visual concepts from a single image is a complex and challenging problem. In this paper, we study a new and challenging task, customized concept decomposition, wherein the objective is to leverage diffusion models to decompose a single image and generate visual concepts from various perspectives. To address this challenge, we propose a two-stage framework, CusConcept (short for Customized Visual Concept Decomposition), to extract customized visual concept embedding vectors that can be embedded into prompts for text-to-image generation. In the first stage, CusConcept employs a vocabulary-guided concept decomposition mechanism to build vocabularies along human-specified conceptual axes. The decomposed concepts are obtained by retrieving corresponding vocabularies and learning anchor weights. In the second stage, joint concept refinement is performed to enhance the fidelity and quality of generated images. We further curate an evaluation benchmark for assessing the performance of the open-world concept decomposition task. Our approach can effectively generate high-quality images of the decomposed concepts and produce related lexical predictions as secondary outcomes. Extensive qualitative and quantitative experiments demonstrate the effectiveness of CusConcept.

cs.CV↗

arXiv2024

Pair density wave, infinite-length stripes, and holon Wigner crystal in single-band Hubbard model on diagonal square lattice

We employ large-scale density-matrix renormalization group (DMRG) simulations to investigate the quantum phase diagram of the hole-doped Hubbard model on square lattices. By implementing a diagonally oriented square lattice and GPU-accelerated DMRG with up to $48000$ states, we identify three distinct quantum phases across $\delta = 5\%$ to $15\%$ doping: (i) A diagonal stripe phase with short-range uniform superconductivity (SC) at lower doping $\delta\lesssim 9\%$; (ii) An intermediate holon Wigner crystal (WC*) phase exhibiting bidirectional charge-density order and short-range SC with spatial oscillating correlations; (iii) An unprecedented infinite-length stripe (i-stripe) phase at $\delta\gtrsim 12\%$ hosting long stripes spanning the whole lattice. Remarkably, as doping increases, the short-range SC in WC* phase evolves into a 2D-like pair density wave (PDW) with divergent susceptibility in the i-stripe phase, constituting probably the first controlled numerical evidence of dominant PDW in the single-band square-lattice Hubbard model. The established 2D-like PDW and its interplay with charge orders provide new perspectives on dynamical layer decoupling phenomena in cuprates and multifaceted relationships between charge, spin and SC orders in quantum materials.

cond-mat.str-el↗

arXiv2024

AdaptiveFusion: Adaptive Multi-Modal Multi-View Fusion for 3D Human Body Reconstruction

Recent advancements in sensor technology and deep learning have led to significant progress in 3D human body reconstruction. However, most existing approaches rely on data from a specific sensor, which can be unreliable due to the inherent limitations of individual sensing modalities. Additionally, existing multi-modal fusion methods generally require customized designs based on the specific sensor combinations or setups, which limits the flexibility and generality of these methods. Furthermore, conventional point-image projection-based and Transformer-based fusion networks are susceptible to the influence of noisy modalities and sensor poses. To address these limitations and achieve robust 3D human body reconstruction in various conditions, we propose AdaptiveFusion, a generic adaptive multi-modal multi-view fusion framework that can effectively incorporate arbitrary combinations of uncalibrated sensor inputs. By treating different modalities from various viewpoints as equal tokens, and our handcrafted modality sampling module by leveraging the inherent flexibility of Transformer models, AdaptiveFusion is able to cope with arbitrary numbers of inputs and accommodate noisy modalities with only a single training network. Extensive experiments on large-scale human datasets demonstrate the effectiveness of AdaptiveFusion in achieving high-quality 3D human body reconstruction in various environments. In addition, our method achieves superior accuracy compared to state-of-the-art fusion methods.

cs.CV↗

arXiv2024

Evolution of Interfacial Hydration Structure Induced by Ion Condensation and Correlation Effects

Interfacial hydration structures are crucial in wide-ranging applications, including battery, colloid, lubrication etc. Multivalent ions like Mg2+ and La3+ show irreplaceable roles in these applications, which are hypothesized due to their unique interfacial hydration structures. However, this hypothesis lacks experimental supports. Here, using three-dimensional atomic force microscopy (3D-AFM), we provide the first observation for their interfacial hydration structures with molecular resolution. We observed the evolution of layered hydration structures at La(NO3)3 solution-mica interfaces with concentration. As concentration increases from 25 mM to 2 M, the layer number varies from 2 to 1 and back to 2, and the interlayer thickness rises from 0.25 to 0.34 nm, with hydration force increasing from 0.27+-0.07 to 1.04+-0.24 nN. Theory and molecular simulation reveal that multivalence induces concentration-dependent ion condensation and correlation effects, resulting in compositional and structural evolution within interfacial hydration structures. Additional experiments with MgCl2-mica, La(NO3)3-graphite and Al(NO3)3-mica interfaces together with literature comparison confirm the universality of this mechanism for both multivalent and monovalent ions. New factors affecting interfacial hydration structures are revealed, including concentration and solvent dielectric constant. This insight provides guidance for designing interfacial hydration structures to optimize solid-liquid-interphase for battery life extension, modulate colloid stability and develop efficient lubricants.

physics.chem-ph↗

arXiv2024

Debiased Multimodal Understanding for Human Language Sequences

Human multimodal language understanding (MLU) is an indispensable component of expression analysis (e.g., sentiment or humor) from heterogeneous modalities, including visual postures, linguistic contents, and acoustic behaviours. Existing works invariably focus on designing sophisticated structures or fusion strategies to achieve impressive improvements. Unfortunately, they all suffer from the subject variation problem due to data distribution discrepancies among subjects. Concretely, MLU models are easily misled by distinct subjects with different expression customs and characteristics in the training data to learn subject-specific spurious correlations, limiting performance and generalizability across new subjects. Motivated by this observation, we introduce a recapitulative causal graph to formulate the MLU procedure and analyze the confounding effect of subjects. Then, we propose SuCI, a simple yet effective causal intervention module to disentangle the impact of subjects acting as unobserved confounders and achieve model training via true causal effects. As a plug-and-play component, SuCI can be widely applied to most methods that seek unbiased predictions. Comprehensive experiments on several MLU benchmarks clearly show the effectiveness of the proposed module.

cs.AI↗

arXiv2024

Superconductivity enhancement and particle-hole asymmetry: interplay with electron attraction in doped Hubbard model

The role of near-neighbor electron attraction $V$ in strongly correlated systems has been at the forefront of recent research of unconventional superconductivity. However, its implications in the doped Hubbard model on expansive systems remain predominantly unexplored. In this study, we employ the density-matrix renormalization group to examine its effect in the lightly doped $t$-$t'$-Hubbard model on six-leg square cylinders, where $t$ and $t'$ are the first and second neighbor electron hopping amplitudes. For positive $t'$ in the electron-doped case, our results show that the attractive $V$ can significantly enhance the superconducting correlations and drive the system into a pronounced superconducting phase when the attraction exceeds a modest value $V_c \approx 0.7t$. In contrast, in the hole-doped regime with negative $t' $, while heightened superconducting correlations have also been observed in the charge stripe phase, the systems remain insulating with pronounced charge density wave order. Our results demonstrate the importance of the electron attraction in boosting superconductivity in broader doped Hubbard systems and highlight the asymmetry between the electron and hole-doped regimes.

cond-mat.str-el↗