arXiv subjects

Peng Zhang

Publications and source records attributed to Peng Zhang.

At least 91 records · Page 5Linked to original sources

arXiv2026

UCM: Unified Modeling of Camera Control and Memory with Time-aware Positional Encoding Warping for World Models

World models based on video generation demonstrate remarkable potential for simulating interactive environments yet suffer from persistent difficulties in two key areas: maintaining long-term content consistency when scenes are revisited and enabling precise camera control from user-specified inputs. Existing methods based on explicit 3D reconstruction often compromise flexibility in unbounded scenarios and struggle to preserve fine-grained structures. Alternative methods rely directly on previously generated frames without establishing explicit spatial correspondence, thereby limiting controllability and consistency. To address these limitations, we present UCM, a novel framework for unified modeling of long-term memory and precise camera control via a time-aware positional encoding warping mechanism. To reduce computational overhead, we design an efficient dual-stream diffusion transformer for high-fidelity generation. Moreover, we introduce a scalable data curation strategy that utilizes point-cloud-based rendering to simulate scene revisiting, enabling training on over 500K monocular videos. Extensive experiments on real-world and synthetic benchmarks demonstrate that UCM significantly outperforms state-of-the-art methods on long-term scene consistency, while achieving precise camera controllability in high-fidelity video generation.

cs.CV↗

arXiv2026

Wan-Streamer v0.1: End-to-end Real-time Interactive Foundation Models

We present Wan-Streamer, a native-streaming, end-to-end interactive foundation model designed from the ground up for real-time, low-latency, full-duplex audio-visual interaction. Wan-Streamer seamlessly models language, audio, and video as both input and output within a single Transformer, where the sequence is represented as interleaved visual, audio, and text input tokens together with visual, audio, and text output tokens, coordinated by block-causal attention for incremental streaming. Unlike cascaded interactive systems that rely on separate VAD, ASR, language, TTS, audio-driven animation, or video-generation modules, Wan-Streamer does not rely on external language, speech, avatar, or video-generation modules: perception, reasoning, generation, response timing, turn management, and cross-modal synchronization are learned jointly within one unified model, reducing pipeline latency and error accumulation. To support natural audio-visual responsiveness, we redesign the entire stack around streamability, including causal encoders, causal decoders, block-causal attention, and low-latency multimodal token scheduling, enabling streaming units as short as 160 ms at 25 fps. Wan-Streamer achieves approximately 200 ms model-side response latency and approximately 550 ms total interaction latency when combined with 350 ms bidirectional network latency, supporting sub-second duplex audio-visual communication. These results position Wan-Streamer as a unified, end-to-end, multimodal interactive foundation model for low-latency streaming interaction.

cs.CV↗

arXiv2026

Taming Text-to-Sounding Video Generation via Advanced Modality Condition and Interaction

This study focuses on a challenging yet promising task, Text-to-Sounding-Video (T2SV) generation, which aims to generate a video with synchronized audio from text conditions, meanwhile ensuring both modalities are aligned with text. Despite progress in joint audio-video training, two critical challenges still remain unaddressed: (1) a single, shared text caption where the text for video is equal to the text for audio often creates modal interference, confusing the pretrained backbones, and (2) the optimal mechanism for cross-modal feature interaction remains unclear. To address these challenges, we first propose the Hierarchical Visual-Grounded Captioning (HVGC) framework that generates pairs of disentangled captions, a video caption, and an audio caption, eliminating interference at the conditioning stage. Based on HVGC, we further introduce BridgeDiT, a novel dual-tower diffusion transformer, which employs a Dual CrossAttention (DCA) mechanism that acts as a robust ``bridge" to enable a symmetric, bidirectional exchange of information, achieving both semantic and temporal synchronization. Extensive experiments on three benchmark datasets, supported by human evaluations, demonstrate that our method achieves state-of-the-art results on most metrics. Comprehensive ablation studies further validate the effectiveness of our contributions, offering key insights for the future T2SV task. All the codes and checkpoints will be publicly released.

cs.CV↗

arXiv2026

Grammar-Guided Hierarchical Parsing for Long-form Audio Activity Recognition

Long-form audio exhibits an inherent hierarchy: fine-grained events form sub-activities, which in turn constitute higher-level activities. Prior work often models these levels separately, leading to cross-level inconsistencies and requiring supervision at multiple levels. We formulate the problem as hierarchical parsing from event-level evidence: given detected event segments with class posteriors, we infer an order-consistent Act-Sub-Event parse tree. We propose Hierarchical Activity Grammar, encoding hierarchical composition and temporal-order constraints, and perform grammar-guided decoding that combines event evidence with a grammar prior. This yields a temporally grounded parse tree from which sub-activity segmentation and activity classification are derived, without requiring sub-activity or activity labels for training. Experiments on the long-form MultiAct audio dataset demonstrate improved temporal-order consistency (Edit score) and produces interpretable hierarchies.

cs.SD↗

arXiv2026

SharQ: Bridging Activation Sparsity and FP4 Quantization for LLM Inference

Low-bit floating-point formats and semi-structured sparsity are increasingly supported by modern accelerators, yet combining them for LLM activation compression remains challenging: activations contain input-dependent outliers that dominate block scales in FP4 quantization, and directly applying N:M sparsity masks discards moderate values, coupling sparsification loss with quantization error. We introduce SharQ, a training-free inference method that bridges activation sparsity and FP4 quantization through an online sparse--dense decomposition. For each activation tensor, SharQ generates an input-adaptive N:M mask to extract an outlier-dominated sparse backbone, quantizes it to FP4, and defines a dense residual relative to the quantized sparse backbone rather than the unquantized sparse values. A sparse FP4 GEMM processes the backbone while a dense FP4 GEMM compensates for both mask-induced activation loss and sparse-path quantization error. The two paths share a single FP4 weight payload with path-specific scale views, and a fused preparation kernel absorbs mask generation, residual construction, and layer normalization into one operator. SharQ requires no calibration data, retraining, or model-specific tuning. Evaluated on Llama-3.1-8B, Qwen2.5-7B, Qwen3-30B-A3B, and Qwen3-VL-8B, SharQ recovers 43--63% of the NVFP4-to-FP16 accuracy gap across language and vision-language tasks, and generalizes across NVFP4, HiF4, and MXFP4 formats. On an RTX 5090, SharQ delivers 2.2--2.4$\times$ latency reduction over FP16 and 1.2--1.4$\times$ throughput improvement over FP8 in language model serving, and up to 1.58$\times$ speedup on Wan2.2-T2V-A14B video generation when combined with SageAttention. Our code is available at https://github.com/actypedef/SharQ.

cs.LG↗

arXiv2026

Astragalus: Automatic Configuration Repair for Production Networks

Network configurations are prone to errors, which can lead to catastrophic service outages. A tool that can achieve automatic configuration repair (ACR) is highly desired by operators. Existing tools for ACR follow a \textit{semantics-driven approach}: they model network semantics as a set of SMT constraints, and solve them for a location or fix of the error. Due to the complex semantics of networks, constructing and solving these constraints can be prohibitively expensive, making these tools neither general nor scalable. Inspired by automatic program repair (APR), we explore another direction, i.e., a \textit{syntax-driven approach}, which generates and validates syntactically-valid candidate updates without modeling program semantics, often drawing on existing code in the same repository. Following this direction, we propose Astragalus, a syntax-driven method for ACR. It uses multiple iterations of a "localize-fix-validate" pipeline to search for repairs, and proves quite effective on configurations of our production network. Specifically, we show that Astragalus can repair every incident in multiple sizes of a synthesized network, and 97.5% of the incidents on a real network, both with 15 types of errors injected, within an average time of 6.93 seconds. It has also provided valid repairs in under 6 minutes for 7 recent network incidents or undesired changes, in a real production network with O(1,000)~O(10,000) devices.

cs.NI↗

arXiv2026

GRB 250706B/C: Insight-HXMT Discovery of a High-Luminosity Burst as a Candidate for Fallback-Regulated Accretion in the Prompt Emission

Fallback accretion in collapsar models is often associated with underluminous gamma-ray bursts (GRBs), leading to the widespread view that fallback-fed engines may be intrinsically inefficient at producing high-luminosity events. In this Letter, we present GRB 250706B/C, a luminous long GRB observed by \textit{Insight}-HXMT that exhibits an unusual combination of extreme short-timescale variability and coherent large-scale temporal evolution. The prompt emission contains at least 79 resolved pulses and a minimum variability timescale of $\sim11$ ms. The pulse widths are nearly independent of photon energy and span a broad distribution with a median FWHM of $\sim0.30$ s, while the waiting times between adjacent pulses have a median of $\sim0.38$ s. The prompt-emission envelope exhibits a prolonged rise described by $F(t)\propto (t-t_0)^{0.47\pm0.01}$ followed by a rapid decline. Despite substantial pulse-to-pulse fluctuations, neither the pulse widths nor the waiting times show significant secular evolution during the main emission episode. These features indicate the coexistence of two distinct temporal components, including a slow evolving rising luminosity envelope and rapid stochastic variability. Such behavior is consistent with scenarios in which a time-dependent engine-feeding history regulates the large-scale emission while internal dissipation within the relativistic outflow produces the pulse structure. Within this context, GRB~250706B/C may represent a fallback-fed collapsar operating on a high-luminosity branch, suggesting that fallback itself does not necessarily limit the luminosity scale of GRBs.

astro-ph.HE↗

arXiv2026

Beyond Pairwise Interactions: Equivariant Hypergraph Diffusion for Crystal Structure Prediction

Crystal Structure Prediction (CSP) remains a fundamental challenge with significant implications for materials discovery and the advancement of various scientific disciplines. Recent advances have demonstrated that generative models, particularly diffusion models, are especially promising for CSP. However, traditional graph-based representations, where atomic bonds are modeled as pairwise graph edges, fail to capture the intricate high-order interactions essential for accurately describing crystal structures. To address this limitation, we propose leveraging hypergraphs to represent crystal structures, enabling more expressive modeling of multi-way atomic interactions. Hypergraphs naturally encode complex high-order relationships and respect key symmetries -- such as permutation and periodic translation invariance -- that are crucial for characterizing crystalline materials. Building on this representation, we propose the \textbf{E}quivariant \textbf{H}ypergraph \textbf{Diff}usion Model (\textbf{EH-Diff}), a generative framework designed to exploit the symmetry-preserving properties of hypergraphs. EH-Diff provides an efficient and accurate method for predicting crystal structures, with rigorous theoretical guarantees on invariance preservation. Empirically, we conduct extensive experiments on four benchmark datasets, and the results demonstrate that EH-Diff outperforms state-of-the-art CSP methods even with a single diffusion sample.

cs.CE↗

arXiv2026

Evolution of the intertwining correlated topological phases in iron-based superconductor Fe(Te,Se)

Multiple topological electronic phases can coexist within a single quantum material and induce different topological superconducting states, offering deeper insights into interplay of topological superconducting states and Majorana modes, which may also be influenced and modified by correlation effect. Iron-based superconductors, with both topological states and correlation effect, is an ideal platform to study these phenomena. Here, with high resolution angle resolved photoelectron spectroscopy, we directly resolve two distinct intertwining topological states in iron-based superconductor Co-doped Fe(Te,Se), and study their evolution with electron doping. We identify a region where both topological insulator surface states and topological Dirac semimetal states intersect the Fermi level. The topological states are affected by the strong correlation effect and are isolated from trivial bulk states. The evolution between distinct topological phases offers a good opportunity to study various Majorana modes from different superconducting phases according to theoretical analysis. Our findings establish an ideal platform for exploring the interaction between multiple topological superconducting states and the related Majorana modes.

cond-mat.supr-con↗

arXiv2026

Skin Lesion Classification Based on ResNet-50 Enhanced With Adaptive Spatial Feature Fusion

Skin cancer classification is challenging due to high inter-class similarity, intra-class variability, and artifacts in dermoscopic images. To address these issues, we propose an improved ResNet-50 with Adaptive Spatial Feature Fusion (ASFF), which adaptively integrates multi-scale semantic and surface features to refine representations and reduce overfitting. The ResNet-50 model is enhanced with an adaptive feature fusion mechanism to achieve more effective multi-scale feature extraction and improve overall performance. Specifically, a dual-branch design fuses high-level semantic and mid-level detail features which use global average pooling and fully connected layers to produce spatial weights, and emphasizes lesion-relevant regions. Evaluated on a balanced subset of ISIC 2020 (3,297 images, randomly selected from the original dataset), the ASFF-based ResNet-50 outperforms multiple CNN baselines, achieving 93.182% accuracy with superior precision, recall, specificity, and F1. It also reaches 0.9670 AUC (P-R) and 0.9717 AUC (ROC). Grad-CAM visualizations show more accurate focus on lesion areas.The proposed model also generalizes well to ISIC 2019 external validation, outperforming the ResNet-50 baseline. These findings demonstrate that the proposed approach provides a more effective and efficient solution for computer-aided skin cancer diagnosis. The generation codes, weights and confusion matrices are open sourced in https://github.com/Grapesea/ASFF-ResNet50-enhanced.

cs.CV↗

arXiv2026

AbstainGNN: Teaching Graph Neural Networks to Abstain for Graph Classification

Graph classification is a core task in graph data mining with widespread real-world applications. Recent advances in graph neural networks (GNNs) have led to substantial performance improvements for graph classification. However, existing GNNs are typically forced to make predictions even under high uncertainty or unknown conditions, resulting in unreliable decisions that can severely impact downstream tasks, particularly in safety-critical scenarios. To address this critical limitation, we propose AbstainGNN, a novel and theory-driven framework for graph classification with abstention, which enables GNNs to reject uncertain predictions instead of producing incorrect decisions. Specifically, AbstainGNN explicitly models both the predictive function and the abstention function, allowing for effective utilization of graph structural information. Moreover, unlike existing heuristic abstention methods, we theoretically characterize the trade-off between classification errors and rejection costs from a PAC-Bayesian generalization perspective, and derive a unified learning objective for model optimization. Guided by this theoretical insight, we further develop an efficient two-stage training strategy consisting of predictive function warm-start and abstention function calibration. Extensive experiments on five benchmark datasets show that AbstainGNN outperforms existing abstention methods, achieving superior classification performance under the same rejection rates.

cs.LG↗

arXiv2026

V2I Work Zone Geometry Reconstruction with Pose-Conditioned UWB Range Denoising

Reliable work zone mapping is important for connected and autonomous vehicles (CAVs) to navigate safely and smoothly through work zone areas. Cone-mounted ultra-wideband (UWB) roadside units (RSU) offer a cost-effective way for work zone layout inference, as roadside anchors and vehicle tags provide direct vehicle-to-infrastructure (V2I) range constraints for work zone geometry reconstruction. However, UWB range estimation is degraded by bursty outliers, non-line-of-sight (NLOS) errors, arbitrary anchor-ordering issues, and vehicle pose uncertainties in practical field deployments. To address these challenges, this study proposes a pose-conditioned, permutation-equivariant predictive denoiser for multi-anchor UWB ranging. The model employs shared anchor-wise temporal prediction to capture range dynamics, symmetric set aggregation to handle unordered and missing anchors, and pose-conditioned residual decoding to incorporate vehicle motion as a geometric prior. A two-stage training strategy first learns prediction from observed ranges, and then fine-tunes the denoiser with NLOS-weighted supervision. The method is evaluated on rare real-world V2I UWB field data collected with a CAV, as well as on controlled large-scale simulation benchmarks for ablative insights. Results show that the proposed method substantially improves range accuracy, cone localization, and work zone geometry reconstruction in challenging NLOS-dominated regimes, remains robust to anchor re-indexing and moderate anchor dropout, and reduces measurement-weighted field MSE by 66.9% relative to the raw input.

cs.RO↗

arXiv2026

Liquid-fueled Oblique Detonation Stabilized by a Transverse Jet

The role of a transverse liquid n-heptane jet in initiating and stabilizing liquid n-heptane oblique detonation waves (ODWs) in a confined model combustor was computationally investigated in the present work. The jet-to-inflow momentum ratio, J, was identified as the primary control parameter. Under steady inflow pressures, a weak jet with a small J fails to initiate an ODW; a slightly stronger jet ignites only a local near-normal detonation between the OSW and the separation shock wave without forming a developed ODW branch; a moderate jet establishes a standing detonation wave system consisting of an ODW, a near-normal detonation branch, and a separation shock wave; a large but still admissible J produces a wall-coupled ODW-Mach-stem configuration; and an excessive jet momentum destabilizes the ODW by pushing it out of the combustor into the external compression region. Under oscillatory inlet pressure, the standing ODW remains dynamically stabilized within the combustor through bounded, phase-dependent transitions between distinct combustion modes. At sufficiently large J, the transverse jet ceases to act as an effective stabilization actuator. The resulting dynamic-stabilization map reveals a finite operating window governed jointly by jet momentum and inlet-pressure fluctuation.

physics.flu-dyn↗

arXiv2026

DynFrame: Adaptive Reasoning-Driven Multimodal Framework with Dynamic Frame Augmentation for Complex Video Understanding

Recent video multimodal large language models (MLLMs) increasingly couple step-by-step reasoning with on-demand visual evidence retrieval, allowing models to revisit relevant video segments during inference. However, two structural gaps remain in existing thinking-with-video systems. (i) Sampling density is not a learnable decision: existing methods may let the model decide where to look, but the per-window frame rate is largely fixed. As a result, fine-grained evidence is often recovered through repeated retrieval calls, which increases inference context length and training difficulty. (ii) Retrieval and answer generation are usually optimized with a single trajectory-level advantage, so the "where to look" tokens and the "how to answer" tokens receive the same credit even when one is correct and the other is not. To address these gaps, we present DynFrame, a framework that emits the temporal window and the sampling density as native tokens within a single autoregressive pass. This learnable span-density retrieval enables acquiring multi-granularity evidence with a single retrieval step. Based on the above tokenized retrieval interface, we further introduce Segment-Decoupled GRPO (SD-GRPO), which splits each rollout at the retrieval boundary and assigns role-specific token-level advantages, separately crediting the sampling decision and the answer. Trained on the curated DM-CoT-74k and DM-RL-45k, DynFrame-4B is competitive with strong 7B-8B baselines across six benchmarks (NExT-GQA, Charades-STA, ActivityNet-MR, Video-MME, MLVU, LVBench), and DynFrame-8B sets new state-of-the-art on most metrics. Code is available at https://github.com/zhangguanghao523/DynFrame.

cs.CV↗

arXiv2026

PixelWizard: Towards Efficient High-Fidelity Video Generation at Ultra-Large Spatial Resolution

High-resolution video generation faces a coupled bottleneck of optimization instability and prohibitive computational costs. The massive expansion of the token sequence not only biases optimization toward local textures at the expense of global coherence, leading to structural collapse, but also imposes prohibitive training costs and severe inference latency. To address this, we propose PixelWizard, a framework that hierarchically decouples global structure modeling from fine-grained detail synthesis. PixelWizard first establishes a compact spatiotemporal anchor to concentrate dense structural priors, which then guides fine-grained generation at high resolution. This mitigates the local optimization bias to ensure structural stability without compromising high-frequency details. Leveraging this structural stability, we introduce Noise-Span Aligned Shortcut Training to break the inference bottleneck. By explicitly modeling the step size, this mechanism allows the model to traverse the generation trajectory with large steps. Crucially, we incorporate Exponential Index-Biased Sampling and Adaptive Noise-Span Calibration to align optimization with the shifted noise schedules of high-resolution grids, ensuring robust few-step inference without incurring the heavy overhead of distillation. Extensive experiments demonstrate that PixelWizard achieves superior visual quality while accelerating the generative sampling of native 2K/4K videos by over 10x.

cs.CV↗

arXiv2026

Probing scalar-neutrino and scalar-dark-matter interactions with PandaX-4T

Scalar-mediated interactions may exist among neutrinos, dark matter particles, or between the two. Double $β$-decay experiments provide a powerful tool to probe such exotic interactions. Using $^{136}$Xe double $β$-decay data from PandaX-4T, we perform the first direct spectral search in the energy range of 20 to 2800~keV, setting the most stringent limits to date on scalar-mediated neutrino self-interactions for mediator masses below 2~MeV$/c^2$. These results place significant constraints on models invoking such interactions to alleviate the Hubble Tension. Assuming the same scalar also mediates dark matter self-interactions, constraints on the dark matter-scalar interactions can be placed in conjunction with cosmological constraints.

hep-ex↗

arXiv2026

Oscillating Detonation of Liquid Ammonia

Liquid ammonia is a promising carbon-free energy carrier, but its high volatility and low reactivity lead to detonation dynamics that differ significantly from those of liquid hydrocarbons. Using Eulerian-Lagrangian simulations, we revealed an oscillating detonation phenomenon driven by the flash boiling of ammonia. Specifically, intense endothermic evaporation and exothermic combustion periodically weaken and then restore the coupling between the shock front and the reaction zone. A delay differential equation (DDE) model is developed to describe this oscillatory behaviour. In this model, the shock response constraint is derived from the positive characteristic compatibility relation behind the shock front, and it is closed using delay-augmented relaxation models for evaporation and reaction progress variables. Normal-mode stability analysis of the DDE system shows that an oscillatory solution emerges when the evaporation and reaction timescales are comparable. Simulation data across different evaporation models, droplet diameters, and ambient temperatures collapse onto the theoretical frequency band predicted by the model.

physics.flu-dyn↗

arXiv2026

Understanding Generalization through Decision Pattern Shift

Understanding why deep neural networks (DNNs) fail to generalize to unseen samples remains a long-standing challenge. Existing studies mainly examine changes in externally observable factors such as data, representations, or outputs, yet offer limited insight into how a model's internal decision mechanism evolves from training to test. To address this gap, we introduce Decision Pattern Shift (DPS), a new perspective that defines generalization through the stability of internal decision patterns and quantifies failure as their deviation from those learned during training. Specifically, we represent each sample's decision pattern as a GradCAM-based channel-contribution vector, which captures how feature channels collectively support a prediction, and we propose the DPS metric to measure its discrepancy from the class-average pattern. Empirical analyses across multiple datasets and architectures show that, (i) decision patterns form a highly structured, class-consistent space with strong intra-class cohesion and low inter-class confusion, enabling direct analysis of a model's decision logic; (ii) the DPS magnitude correlates linearly with the generalization gap (nearly all Pearson r > 0.8), revealing generalization as a systematic drift in the model's internal decision mechanism; (iii) the DPS spectrum organizes diverse generalization degradation scenarios (covering ideal generalization, in-distribution degradation, domain shift, out-of-distribution, and shortcut learning) into a continuous trajectory, providing a unified explanation of their failure modes. These findings open up new possibilities for early generalization-risk detection, failure-mode diagnosis, and channel-level defect localization.

cs.LG↗