Search arXiv⌕ Search

arXiv subjects

Zheng Zhang

Publications and source records attributed to Zheng Zhang.

At least 127 records · Page 7Linked to original sources
arXiv2026

FourierPET: Deep Fourier-based Unrolled Network for Low-count PET Reconstruction

Low-count positron emission tomography (PET) reconstruction is a challenging inverse problem due to severe degradations arising from Poisson noise, photon scarcity, and attenuation correction errors. Existing deep learning methods typically address these in the spatial domain with an undifferentiated optimization objective, making it difficult to disentangle overlapping artifacts and limiting correction effectiveness. In this work, we perform a Fourier-domain analysis and reveal that these degradations are spectrally separable: Poisson noise and photon scarcity cause high-frequency phase perturbations, while attenuation errors suppress low-frequency amplitude components. Leveraging this insight, we propose FourierPET, a Fourier-based unrolled reconstruction framework grounded in the Alternating Direction Method of Multipliers. It consists of three tailored modules: a spectral consistency module that enforces global frequency alignment to maintain data fidelity, an amplitude-phase correction module that decouples and compensates for high-frequency phase distortions and low-frequency amplitude suppression, and a dual adjustment module that accelerates convergence during iterative reconstruction. Extensive experiments demonstrate that FourierPET achieves state-of-the-art performance with significantly fewer parameters, while offering enhanced interpretability through frequency-aware correction.

eess.IV↗
arXiv2026

Fatigue-Aware Learning to Defer via Constrained Optimisation

Learning to defer (L2D) enables human-AI cooperation by deciding when an AI system should act autonomously or defer to a human expert. Existing L2D methods, however, assume static human performance, contradicting well-established findings on fatigue-induced degradation. We propose Fatigue-Aware Learning to Defer via Constrained Optimisation (FALCON), which explicitly models workload-varying human performance using psychologically grounded fatigue curves. FALCON formulates L2D as a Constrained Markov Decision Process (CMDP) whose state includes both task features and cumulative human workload, and optimises accuracy under human-AI cooperation budgets via PPO-Lagrangian training. We further introduce FA-L2D, a benchmark that systematically varies fatigue dynamics from near-static to rapidly degrading regimes. Experiments across multiple datasets show that FALCON consistently outperforms state-of-the-art L2D methods across coverage levels, generalises zero-shot to unseen experts with different fatigue patterns, and demonstrates the advantage of adaptive human-AI collaboration over AI-only or human-only decision-making when coverage lies strictly between 0 and 1.

cs.LG↗
arXiv2026

Genesis: Evolving Attack Strategies for LLM Web Agent Red-Teaming

As large language model (LLM) agents increasingly automate complex web tasks, they boost productivity while simultaneously introducing new security risks. However, relevant studies on web agent attacks remain limited. Existing red-teaming approaches mainly rely on manually crafted attack strategies or static models trained offline. Such methods fail to capture the underlying behavioral patterns of web agents, making it difficult to generalize across diverse environments. In web agent attacks, success requires the continuous discovery and evolution of attack strategies. To this end, we propose Genesis, a novel agentic framework composed of three modules: Attacker, Scorer, and Strategist. The Attacker generates adversarial injections by integrating the genetic algorithm with a hybrid strategy representation. The Scorer evaluates the target web agent's responses to provide feedback. The Strategist dynamically uncovers effective strategies from interaction logs and compiles them into a continuously growing strategy library, which is then re-deployed to enhance the Attacker's effectiveness. Extensive experiments across various web tasks show that our framework discovers novel strategies and consistently outperforms existing attack baselines. Our code is available at https://github.com/CjangCjengh/web_agent_attack.

cs.AI↗
arXiv2026

AGFT: Alignment-Guided Fine-Tuning for Zero-Shot Adversarial Robustness of Vision-Language Models

Pre-trained vision-language models (VLMs) exhibit strong zero-shot generalization but remain vulnerable to adversarial perturbations. Existing classification-guided adversarial fine-tuning methods often disrupt pre-trained cross-modal alignment, weakening visual-textual correspondence and degrading zero-shot performance. In this paper, we propose an Alignment-Guided Fine-Tuning (AGFT) framework that enhances zero-shot adversarial robustness while preserving the cross-modal semantic structure. Unlike label-based methods that rely on hard labels and fail to maintain the relative relationships between image and text, AGFT leverages the probabilistic predictions of the original model for text-guided adversarial training, which aligns adversarial visual features with textual embeddings via soft alignment distributions, improving zero-shot adversarial robustness. To address structural discrepancies introduced by fine-tuning, we introduce a distribution consistency calibration mechanism that adjusts the robust model output to match a temperature-scaled version of the pre-trained model predictions. Extensive experiments across multiple zero-shot benchmarks demonstrate that AGFT outperforms state-of-the-art methods while significantly improving zero-shot adversarial robustness.

cs.CV↗
arXiv2026

Rényi Entropy: A New Token Pruning Metric for Vision Transformers

Vision Transformers (ViTs) achieve state-of-the-art performance but suffer from the $O(N^2)$ complexity of self-attention, making inference costly for high-resolution inputs. To address this bottleneck, token pruning has emerged as a critical technique to accelerate inference. Most existing methods rely on the [CLS] token to estimate patch importance. However, we argue that the [CLS] token can be unreliable in early layers where semantic representations are still immature. As a result, pruning in the early layer often leads to inaccurate importance estimation and unnecessary information loss. In this work, we propose a training-free token importance metric, namely Col-Ln, which is derived from Rényi entropy that enables the identification of informative tokens from the first layer of the network, thereby enabling more reliable pruning in token reduction. Extensive experiments on ViTs and Large Vision-Language Models (LVLMs) demonstrate that our approach consistently outperforms state-of-the-art pruning methods across diverse benchmarks.

cs.CV↗
arXiv2026

FETTA: Flexible and Efficient Hardware Accelerator for Tensorized Neural Network Training

The increasing demand for on-device training of deep neural networks (DNNs) aims to leverage personal data for high-performance applications while addressing privacy concerns and reducing communication latency. However, resource-constrained platforms face significant challenges due to the intensive computational and memory demands of DNN training. Tensor decomposition emerges as a promising approach to compress model size without sacrificing accuracy. Nevertheless, training tensorized neural networks (TNNs) incurs non-trivial overhead and severe performance degradation on conventional accelerators due to complex tensor shaping requirements. To address these challenges, we propose FETTA, an algorithm and hardware co-optimization framework for efficient TNN training. On the algorithm side, we develop a contraction sequence search engine (CSSE) to identify the optimal contraction sequence with the minimal computational overhead. On the hardware side, FETTA features a flexible and efficient architecture equipped with a reconfigurable contraction engine (CE) array to support diverse dataflows. Furthermore, butterfly-based distribution and reduction networks are implemented to perform flexible tensor shaping operations during computation. Evaluation results demonstrate that FETTA achieves reductions of 20.5x/100.9x, 567.5x/45.03x, and 11609.7x/4544.8x in terms of processing latency, energy, and energy-delay product (EDP) over GPU and TPU, respectively. Moreover, working on the tensorized training, FETTA outperforms prior accelerators with a speedup of 3.87~14.63x, and an energy efficiency improvement of 1.41~2.73x on average.

cs.AR↗
arXiv2026

Enhancing Jailbreak Attacks on LLMs via Persona Prompts

Jailbreak attacks aim to exploit large language models (LLMs) by inducing them to generate harmful content, thereby revealing their vulnerabilities. Understanding and addressing these attacks is crucial for advancing the field of LLM safety. Previous jailbreak approaches have mainly focused on direct manipulations of harmful intent, with limited attention to the impact of persona prompts. In this study, we systematically explore the efficacy of persona prompts in compromising LLM defenses. We propose a genetic algorithm-based method that automatically crafts persona prompts to bypass LLM's safety mechanisms. Our experiments reveal that: (1) our evolved persona prompts reduce refusal rates by 50-70% across multiple LLMs, and (2) these prompts demonstrate synergistic effects when combined with existing attack methods, increasing success rates by 10-20%. Our code and data are available at https://github.com/CjangCjengh/Generic_Persona.

cs.CR↗
arXiv2026

GeoDiffMM: Geometry-Guided Conditional Diffusion for Motion Magnification

Video Motion Magnification (VMM) amplifies subtle macroscopic motions to a perceptible level. Recently, existing mainstream Eulerian approaches address amplification-induced noise via decoupling representation learning such as texture, shape and frequency schemes, but they still struggle to mitigate the interference of photon noise on true micro-motion when motion displacements are very small. We propose GeoDiffMM, a novel diffusion-based Lagrangian VMM framework conditioned on optical flow as a geometric cue, enabling structurally consistent motion magnification. Specifically, we design a Noise-Free Optical Flow Augmentation strategy that synthesizes diverse nonrigid motion fields without photon noise as supervision, helping the model learn more accurate geometry-aware optical flow and generalize better. Next, we develop a Diffusion Motion Magnifier that conditions the denoising process on (i) optical flow as a geometry prior and (ii) a learnable magnification factor controlling magnitude, thereby selectively amplifying motion components consistent with scene semantics and structure. Finally, we perform Flow-based Video Synthesis to map the amplified motion back to the image domain with high fidelity. Extensive experiments on real and synthetic datasets show that GeoDiffMM outperforms state-of-the-art methods and significantly improves motion magnification.

cs.CV↗
arXiv2026

PROBE: Diagnosing Residual Concept Capacity in Erased Text-to-Video Diffusion Models

Concept erasure techniques for text-to-video (T2V) diffusion models report substantial suppression of sensitive content, yet current evaluation is limited to checking whether the target concept is absent from generated frames, treating output-level suppression as evidence of representational removal. We introduce PROBE, a diagnostic protocol that quantifies the \textit{reactivation potential} of erased concepts in T2V models. With all model parameters frozen, PROBE optimizes a lightweight pseudo-token embedding through a denoising reconstruction objective combined with a novel latent alignment constraint that anchors recovery to the spatiotemporal structure of the original concept. We make three contributions: (1) a multi-level evaluation framework spanning classifier-based detection, semantic similarity, temporal reactivation analysis, and human validation; (2) systematic experiments across three T2V architectures, three concept categories, and three erasure strategies revealing that all tested methods leave measurable residual capacity whose robustness correlates with intervention depth; and (3) the identification of temporal re-emergence, a video-specific failure mode where suppressed concepts progressively resurface across frames, invisible to frame-level metrics. These findings suggest that current erasure methods achieve output-level suppression rather than representational removal. We release our protocol to support reproducible safety auditing. Our code is available at https://github.com/YiweiXie/PRObingBasedEvaluation.

cs.CV↗
arXiv2026

Speed by Simplicity: A Single-Stream Architecture for Fast Audio-Video Generative Foundation Model

We present daVinci-MagiHuman, an open-source audio-video generative foundation model for human-centric generation. daVinci-MagiHuman jointly generates synchronized video and audio using a single-stream Transformer that processes text, video, and audio within a unified token sequence via self-attention only. This single-stream design avoids the complexity of multi-stream or cross-attention architectures while remaining easy to optimize with standard training and inference infrastructure. The model is particularly strong in human-centric scenarios, producing expressive facial performance, natural speech-expression coordination, realistic body motion, and precise audio-video synchronization. It supports multilingual spoken generation across Chinese (Mandarin and Cantonese), English, Japanese, Korean, German, and French. For efficient inference, we combine the single-stream backbone with model distillation, latent-space super-resolution, and a Turbo VAE decoder, enabling generation of a 5-second 256p video in 2 seconds on a single H100 GPU. In automatic evaluation, daVinci-MagiHuman achieves the highest visual quality and text alignment among leading open models, along with the lowest word error rate (14.60%) for speech intelligibility. In pairwise human evaluation, it achieves win rates of 80.0% against Ovi 1.1 and 60.9% against LTX 2.3 over 2000 comparisons. We open-source the complete model stack, including the base model, the distilled model, the super-resolution model, and the inference codebase.

cs.CV↗
arXiv2026

Test-Time 3D Occupancy Prediction

Self-supervised 3D occupancy prediction offers a promising solution for understanding complex driving scenes without requiring costly 3D annotations. However, training dense occupancy decoders to capture fine-grained geometry and semantics can demand hundreds of GPU hours, and once trained, such models struggle to adapt to varying voxel resolutions or novel object categories without extensive retraining. To overcome these limitations, we propose a practical and flexible test-time occupancy prediction framework termed TT-Occ. Our method incrementally constructs, optimizes, and voxelizes time-aware 3D Gaussians from raw sensor streams by integrating vision foundation models (VFMs) at runtime. The flexible representation of 3D Gaussians enables voxelization at arbitrary user-specified resolutions, while the strong generalization capability of VFMs supports accurate perception and open-vocabulary recognition without requiring any network training or fine-tuning. To validate the generality and effectiveness of our framework, we present two variants: a LiDAR-based version and a vision-centric version, and conduct extensive experiments on the Occ3D-nuScenes and nuCraft benchmarks under varying voxel resolutions. Experimental results show that TT-Occ significantly outperforms existing computationally expensive pretrained self-supervised counterparts. Code is available at https://github.com/Xian-Bei/TT-Occ.

cs.CV↗
arXiv2026

TALO: Pushing 3D Vision Foundation Models Towards Globally Consistent Online Reconstruction

3D vision foundation models have shown strong generalization in reconstructing key 3D attributes from uncalibrated images through a single feed-forward pass. However, when deployed in online settings such as driving scenarios, predictions are made over temporal windows, making it non-trivial to maintain consistency across time. Recent strategies align consecutive predictions by solving global transformation, yet our analysis reveals their fundamental limitations in assumption validity, local alignment scope, and robustness under noisy geometry. In this work, we propose a higher-DOF and long-term alignment framework based on Thin Plate Spline, leveraging globally propagated control points to correct spatially varying inconsistencies. In addition, we adopt a point-agnostic submap registration design that is inherently robust to noisy geometry predictions. The proposed framework is fully plug-and-play, compatible with diverse 3D foundation models and camera configurations (e.g., monocular or surround-view). Extensive experiments demonstrate that our method consistently yields more coherent geometry and lower trajectory errors across multiple datasets, backbone models, and camera setups, highlighting its robustness and generality. Code is available at https://github.com/Xian-Bei/TALO.

cs.CV↗
arXiv2026

MV2UV: Generating High-quality UV Texture Maps with Multiview Prompts

Generating high-quality textures for 3D assets is a challenging task. Existing multiview texture generation methods suffer from the multiview inconsistency and missing textures on unseen parts, while UV inpainting texture methods do not generalize well due to insufficient UV data and cannot well utilize 2D image diffusion priors. In this paper, we propose a new method called MV2UV that combines 2D generative priors from multiview generation and the inpainting ability of UV refinement to get high-quality texture maps. Our key idea is to adopt a UV space generative model that simultaneously inpaints unseen parts of multiview images while resolving the inconsistency of multiview images. Experiments show that our method enables a better texture generation quality than existing methods, especially in unseen occluded and multiview-inconsistent parts.

cs.CV↗
arXiv2026

Lessons from Real-World Deployment of a Cognition-Preserving Writing Tool: Students Actively Engage with Critical Thinking and Planning Affordances

AI-supported writing tools show strong potential for scaffolding students' learning of argumentative writing. Prior work has demonstrated the benefits of AI-supported cognitive scaffolds, such as idea exploration and argument refinement, but how these features function in authentic classroom settings remains underexplored. In this paper, we investigate the classroom integration of an AI-supported writing tool, VISAR. We deployed VISAR in an undergraduate writing course across three sections for one week each over two semesters (49 students total). Using a mixed-methods approach that combines interaction logs, writing artifact analysis, surveys, and interviews, we examine how students used VISAR features in authentic writing tasks. Our findings confirm that students appropriated AI-supported cognitive scaffolds for writing learning and achieved measurable learning gains. While prior studies suggest that students may bypass important cognitive processes when using AI writing assistants, our classroom deployment shows that when systems provide structured supports for planning and targeted generation, students naturally choose to engage with these cognition-preserving scaffolds. These learning-oriented interaction patterns were positively associated with argumentative writing quality, improved conceptual understanding, and emerging critical AI literacy, highlighting the design value of cognition-preserving features in AI writing tools. Together, these findings provide empirical evidence of how AI-supported writing scaffolds operate in authentic classroom contexts and offer design insights for future learning-oriented AI writing tools.

cs.HC↗
arXiv2026

Spectral Signatures of Spinning Dust from Grain Ensembles in Diverse Environments: A Combined Theoretical and Observational Study

Recent observations of anomalous microwave emission (AME) reveal spectral features that are not readily reproduced by spinning dust models. We examine how dust grain distributions and environmental parameters determine the peak frequency and spectral width of AME spectral energy distribution (SED). Using Monte Carlo sampling and global sensitivity analysis, we find that AME features are dominantly controlled by three parameters: grain size, shape, and a phase-dependent environmental parameter. We also quantify the effects of SED broadening from ensembles of these dominant parameters, finding that the level of tension with observations is strongly phase dependent: Molecular Cloud (MC) is fully consistent, Dark Cloud (DC) shows minor deviations, and HII regions exhibit significant offsets in peak frequency. The discrepancy in HII echoes the observed depletion of small dust grains, particularly polycyclic aromatic hydrocarbons (PAHs), in HII regions. However, an observational HII region may still have AME originating from nearby non-HII clouds. In this case, model calculations for HII regions would be inappropriate. Reproducing MC and DC AME features requires ensemble variations in both grain size and environmental parameters are required to reproduce the observed spread in peak frequency and spectral width. We further propose moment expansion and emulation-based inference methods for future AME spectral analysis.

astro-ph.GA↗
arXiv2026

Two results regarding the variation of K-moduli

In this note, we prove two results regarding the variation of K-moduli. The first one reveals the relationship between the chamber decomposition for K-semistable domains and the variation of GIT. The second one presents the relationship between the K-moduli generically parametrizing K-semistable smooth Fano complete intersections of the form $S_1\cap...\cap S_k$ and the K-moduli generically parametrizing K-semistable log Fano manifolds of the form $(\mathbb{P}^n, \sum_{j=1}^kx_jS_j)$, where $x_j\in (0,1)\cap \mathbb{Q}$ and $S_j\subset \mathbb{P}^n$ is a hypersurface of degree $d_j$ for each $1\leq j\leq k$.

math.AG↗
arXiv2026

Disentangling the anisotropic radio sky: Fisher forecasts for 21cm arrays

The existence of a radio synchrotron background (RSB) excess is implied by a number of measurements, including excess emission seen by the ARCADE~2 and LWA experiments. Highly sensitive wideband radio arrays, of the kind used to measure the cosmic 21cm signal, provide a promising way to further constrain the RSB excess through its anisotropy, providing additional insight into its origin. We present a framework for evaluating the potential of 21cm arrays to disentangle different components of the diffuse radio sky based on the combination of their frequency spectrum and angular power spectrum (APS). The formalism is designed to calculate uncertainties due to the intrinsic cosmic variance alone or together with instrumental noise. In particular, we predict the potential for measuring the anisotropy of a broad generalised class of excess radio background models using the low-frequency HERA array as an example. We find that a HERA-like array can distinguish an RSB excess from other sky components based on its angular clustering and spectral dependence, even if these are quite similar to one or more of the other components -- but only in the case that the RSB excess is relatively bright.

astro-ph.CO↗
arXiv2026

Integrating Heterogeneous Information in Randomized Experiments: A Unified Calibration Framework

In modern randomized experiments, large-scale data collection increasingly yields rich baseline covariates and auxiliary information from multiple sources. Such information offers opportunities for more precise treatment effect estimation, but it also raises the challenge of integrating heterogeneous information coherently without compromising validity. Covariate-adaptive randomization (CAR) is widely used to improve covariate balance at the design stage, but it typically balances only a small set of covariates used to form strata, making covariate adjustment at the analysis stage essential for more efficient estimation of treatment effects. Beyond standard covariate adjustment, it is often desirable to incorporate auxiliary information, including cross-stratum information, predictions from various machine learning models, and external data from historical trials or real-world sources. While this auxiliary information is widely available, existing covariate adjustment methods under CAR primarily exploit within-stratum covariates and do not provide a coherent mechanism for integrating it. We propose a unified calibration framework that integrates such information through an information proxy vector and calibration weights defined by a convex optimization problem. The resulting estimator recovers many recent covariate adjustment procedures as special cases while providing a systematic mechanism for both internal and external information borrowing within a single framework. We establish large-sample validity and a no-harm efficiency guarantee, showing that incorporating additional information sources cannot increase asymptotic variance, and we extend the theory to settings in which both the number of strata and the number of information sources grow with the sample size.

stat.ME↗