Searcharxiv⌕ Search

arXiv subjects

Qi Zhang

Publications and source records attributed to Qi Zhang.

At least 235 records · Page 13Linked to original sources
arXiv2026

SpeechRole: A Large-Scale Dataset and Benchmark for Evaluating Speech Role-Playing Agents

Speech is essential for realistic role-playing, yet existing work on role-playing agents largely centers on text, leaving Speech Role-Playing Agents (SRPAs) underexplored and without systematic evaluation. We introduce SpeechRole, a unified framework for developing and assessing SRPAs. SpeechRole-Data contains 98 roles and 111k speech-to-speech conversations with rich timbre and prosodic variation, providing large-scale resources for training SRPAs. SpeechRole-Eval offers a multidimensional benchmark that directly evaluates generated speech, preserving paralinguistic cues and measuring interaction ability, speech expressiveness, and role-playing fidelity. Experiments show that end-to-end SRPAs such as GPT-4o Audio achieve strong fluency and naturalness, but remain limited in prosody consistency and emotion appropriateness. In contrast, current open-source end-to-end models exhibit substantial performance gaps across multiple evaluation dimensions. Cascaded and end-to-end systems achieve comparable results in interaction ability and role-playing fidelity, suggesting that these aspects are still largely influenced by the underlying text-based language models. We release all data, code, and evaluation tools at https://github.com/yuhui1038/SpeechRole.

cs.CL↗
arXiv2026

Unblur-SLAM: Dense Neural SLAM for Blurry Inputs

We propose Unblur-SLAM, a novel RGB SLAM pipeline for sharp 3D reconstruction from blurred image inputs. In contrast to previous work, our approach is able to handle different types of blur and demonstrates state-of-the-art performance in the presence of both motion blur and defocus blur. Moreover, we adjust the computation effort with the amount of blur in the input image. As a first stage, our method uses a feed-forward image deblurring model for which we propose a suitable training scheme that can improve both tracking and mapping modules. Frames that are successfully deblurred by the feed-forward network obtain refined poses and depth through local-global multi-view optimization and loop closure. Frames that fail the first stage deblurring are directly modeled through the global 3DGS representation and an additional blur network to model multiple blurred sub-frames and simulate the blur formation process in 3D space, thereby learning sharp details and refined sub-frame poses. Experiments on several real-world datasets demonstrate consistent improvements in both pose estimation and sharp reconstruction results of geometry and texture.

cs.CV↗
arXiv2026

EnvSocial-Diff: A Diffusion-Based Crowd Simulation Model with Environmental Conditioning and Individual-Group Interaction

Modeling realistic pedestrian trajectories requires accounting for both social interactions and environmental context, yet most existing approaches largely emphasize social dynamics. We propose \textbf{EnvSocial-Diff}: a diffusion-based crowd simulation model informed by social physics and augmented with environmental conditioning and individual--group interaction. Our structured environmental conditioning module explicitly encodes obstacles, objects of interest, and lighting levels, providing interpretable signals that capture scene constraints and attractors. In parallel, the individual--group interaction module goes beyond individual-level modeling by capturing both fine-grained interpersonal relations and group-level conformity through a graph-based design. Experiments on multiple benchmark datasets demonstrate that EnvSocial-Diff outperforms the latest state-of-the-art methods, underscoring the importance of explicit environmental conditioning and multi-level social interaction for realistic crowd simulation. Code is here: https://github.com/zqyq/EnvSocial-Diff.

cs.CV↗
arXiv2026

SynMVCrowd: A Large Synthetic Benchmark for Multi-view Crowd Counting and Localization

Existing multi-view crowd counting and localization methods are evaluated under relatively small scenes with limited crowd numbers, camera views, and frames. This makes the evaluation and comparison of existing methods impractical, as small datasets are easily overfit by these methods. To avoid these issues, 3DROM proposes a data augmentation method. Instead, in this paper, we propose a large synthetic benchmark, SynMVCrowd, for more practical evaluation and comparison of multi-view crowd counting and localization tasks. The SynMVCrowd benchmark consists of 50 synthetic scenes with a large number of multi-view frames and camera views and a much larger crowd number (up to 1000), which is more suitable for large-scene multi-view crowd vision tasks. Besides, we propose strong multi-view crowd localization and counting baselines that outperform all comparison methods on the new SynMVCrowd benchmark. Moreover, we prove that better domain transferring multi-view and single-image counting performance could be achieved with the aid of the benchmark on novel new real scenes. As a result, the proposed benchmark could advance the research for multi-view and single-image crowd counting and localization to more practical applications. The codes and datasets are here: https://github.com/zqyq/SynMVCrowd.

cs.CV↗
arXiv2026

FinToolSyn: A forward synthesis Framework for Financial Tool-Use Dialogue Data with Dynamic Tool Retrieval

Tool-use capabilities are vital for Large Language Models (LLMs) in finance, a domain characterized by massive investment targets and data-intensive inquiries. However, existing data synthesis methods typically rely on a reverse synthesis paradigm, generating user queries from pre-sampled tools. This approach inevitably introduces artificial explicitness, yielding queries that fail to capture the implicit, event-driven nature of real-world needs. Moreover, its reliance on static tool sets overlooks the dynamic retrieval process required to navigate massive tool spaces. To address these challenges, we introduce \textit{FinToolSyn}, a forward synthesis framework designed to generate high-quality financial dialogues. Progressing from persona instruction and atomic tool synthesis to dynamic retrieval dialogue generation, our pipeline constructs a repository of 43,066 tools and synthesizes over 148k dialogue instances, incorporating dynamic retrieval to emulate the noisy candidate sets typical of massive tool spaces. We also establish a dedicated benchmark to evaluate tool-calling capabilities in realistic financial scenarios. Extensive experiments demonstrate that models trained on FinToolSyn achieve a 21.06\% improvement, providing a robust foundation for tool learning in financial scenarios.

cs.CL↗
arXiv2026

Delay Bound Relaxation with Deep Learning-based Haptic Estimation for Tactile Internet

Haptic teleoperation typically demands sub-millisecond latency and ultra-high reliability (99.999%) in Tactile Internet. At a 1 kHz haptic signal sampling rate, this translates into an extremely high packet transmission rate, posing significant challenges for timely delivery and introducing substantial complexity and overhead in radio resource allocation. To address this critical challenge, we introduce a novel DL modelthat estimates force feedback using multi-modal input, i.e. both force measurements from the remote side and local operator motion signals. The DL model can capture complex temporal features of haptic time-series with the use of CNN and LSTM layers, followed by a transformer encoder, and autoregressively produce a highly accurate estimation of the next force values for different teleoperation activities. By ensuring that the estimation error is within a predefined threshold, the teleoperation system can safely relax its strict delay requirements. This enables the batching and transmission of multiple haptic packets within a single resource block, improving resource efficiency and facilitating scheduling in resource allocation. Through extensive simulations, we evaluated network performance in terms of reliability and capacity. Results show that, for both dynamic and rigid object interactions, the proposed method increases the number of reliably served users by up to 66%.

eess.SP↗
arXiv2026

Oxygen-vacancy quantum spin defects in silicon carbide

Optically addressable spin defects in wide-bandgap semiconductors are promising building blocks for quantum sensing and quantum networks. Establishing their atomic structure is essential for understanding functionality and enabling controlled engineering. In 4H-SiC, the PL5 and PL6 centers have long been recognized for their exceptional charge stability and room-temperature optically detected magnetic resonance (ODMR) performance, but their structural origin has remained elusive for over a decade. Here, we provide direct evidence for their oxygen-vacancy (${\rm O_C V_{Si}}$) origins through a combined chemical and isotopic control strategy. Under oxygen ion implantation, we observe over tenfold enhancement in the yield of PL5 and PL6 compared to nitrogen ion implantation. Furthermore, implantation with $^{17}{\rm O}$ ions produces PL5 and PL6 defects that exhibit a characteristic six-fold $^{17}{\rm O}$ hyperfine splitting in their ODMR spectra. These results affirm PL6 as the ${\rm O_C V_{Si}}$ defect in the $hh$ configuration. For PL5, the oxygen-related evidence, together with \textit{ab initio} calculations and additional measurements of the zero-field splitting and hyperfine structure, establishes it as the ${\rm O_C V_{Si}}$ defect in the $kh$ configuration. This unambiguous structural identification, achieved through materials-level chemical control, provides the microscopic foundation for deterministic engineering of these defects, paving the way for scalable photonic devices and high-sensitivity ensemble quantum sensors based on oxygen-vacancy centers.

cond-mat.mtrl-sci↗
arXiv2026

OnlinePG: Online Open-Vocabulary Panoptic Mapping with 3D Gaussian Splatting

Open-vocabulary scene understanding with online panoptic mapping is essential for embodied applications to perceive and interact with environments. However, existing methods are predominantly offline or lack instance-level understanding, limiting their applicability to real-world robotic tasks. In this paper, we propose OnlinePG, a novel and effective system that integrates geometric reconstruction and open-vocabulary perception using 3D Gaussian Splatting in an online setting. Technically, to achieve online panoptic mapping, we employ an efficient local-to-global paradigm with a sliding window. To build local consistency map, we construct a 3D segment clustering graph that jointly leverages geometric and semantic cues, fusing inconsistent segments within sliding window into complete instances. Subsequently, to update the global map, we construct explicit grids with spatial attributes for the local 3D Gaussian map and fuse them into the global map via robust bidirectional bipartite 3D Gaussian instance matching. Finally, we utilize the fused VLM features inside the 3D spatial attribute grids to achieve open-vocabulary scene understanding. Extensive experiments on widely used datasets demonstrate that our method achieves better performance among online approaches, while maintaining real-time efficiency.

cs.CV↗
arXiv2026

SAMA: Factorized Semantic Anchoring and Motion Alignment for Instruction-Guided Video Editing

Current instruction-guided video editing models struggle to simultaneously balance precise semantic modifications with faithful motion preservation. While existing approaches rely on injecting explicit external priors (e.g., VLM features or structural conditions) to mitigate these issues, this reliance severely bottlenecks model robustness and generalization. To overcome this limitation, we present SAMA (factorized Semantic Anchoring and Motion Alignment), a framework that factorizes video editing into semantic anchoring and motion modeling. First, we introduce Semantic Anchoring, which establishes a reliable visual anchor by jointly predicting semantic tokens and video latents at sparse anchor frames, enabling purely instruction-aware structural planning. Second, Motion Alignment pre-trains the same backbone on motion-centric video restoration pretext tasks (cube inpainting, speed perturbation, and tube shuffle), enabling the model to internalize temporal dynamics directly from raw videos. SAMA is optimized with a two-stage pipeline: a factorized pre-training stage that learns inherent semantic-motion representations without paired video-instruction editing data, followed by supervised fine-tuning on paired editing data. Remarkably, the factorized pre-training alone already yields strong zero-shot video editing ability, validating the proposed factorization. SAMA achieves state-of-the-art performance among open-source models and is competitive with leading commercial systems (e.g., Kling-Omni). Code, models, and datasets will be released.

cs.CV↗
arXiv2026

TinyIO: Lightweight Reparameterized Inertial Odometry

Inertial odometry (IO) is a widely used approach for localization on mobile devices; however, obtaining a lightweight IO model that also achieves high accuracy remains challenging. To address this issue, we propose TinyIO, a lightweight IO method. During training, we adopt a multi-branch architecture to extract diverse motion features more effectively. At inference time, the trained multi-branch model is converted into an equivalent single-path architecture to reduce computational complexity. We further propose a Dual-Path Adaptive Attention mechanism (DPAA), which enhances TinyIO's perception of contextual motion along both channel and temporal dimensions with negligible additional parameters. Extensive experiments on public datasets demonstrate that our method attains a favorable trade-off between accuracy and model size. On the RoNIN dataset, TinyIO reduces the ATE by 23.53% compared with R-ResNet and decreases the parameter count by 3.68%.

cs.RO↗
arXiv2026

Interlayer Charge-density-wave Vector Phase Induced Structural Chirality

Chiral charge density waves (CDWs) have attracted intense interest due to their exotic quantum properties, yet the microscopic origin of structural chirality emerging from correlated charge order remains elusive. Here, we reveal that the interlayer phases of CDW wave vectors, an overlooked degree of freedom, play a crucial role in driving chiral structural displacements in layered CDW materials. By explicitly incorporating the interlayer phases in first-principles calculations, we successfully obtained the chiral structure of the CDW phases of AV$_3$Sb$_5$ (A= K, Rb, and Cs) and 1T-TiSe$_2$. The electronic and optical properties of the predicted chiral structures are consistent with experimental measurements of these materials in their CDW phases. We further predict that 1T-NbSe$_2$ is a promising material candidate for realizing chiral CDW order. Beyond materials prediction, our theory reveals that the chiral CDW can be manipulated by electron filling. Our study opens new avenues for discovering, designing, and engineering chiral CDW materials.

cond-mat.mtrl-sci↗
arXiv2026

Preferred Synthesis of Armchair Transition Metal Dichalcogenide Nanotubes

In this work, we present the synthesis of transition-metal dichalcogenide (TMDC) nanotubes with a preferred chiral angle. SnS2, MoS2, and WS2 are formed with high yield and structural purity inside the channels of boron nitride nanotubes. Atomic-resolution imaging, nano-area electron diffraction, and Circular Dichroism spectroscopy reveal that these synthesized TMDC nanotubes prefer to have an armchair configuration, with a probability up to 84%. Density functional theory reveals a negligible difference in the formation energy between armchair and zigzag nanotubes, suggesting that the chirality preference does not originate from the differences in structural stability. However, a detailed TEM investigation revealed that these TMDC nanotubes formed via a transition state of nanoribbons, and these nanoribbons are energetically more stable in a zigzag configuration. Subsequent machine learning potential molecular dynamics simulations verify that zigzag nanoribbons do roll up to form an armchair SnS2 nanotubes. Finally, this "zigzag nanoribbon to armchair nanotube" transition process is directly observed in real time by in-situ transmission electron microscopy. This work demonstrates the first, but likely general, experimental strategy for synthesizing chirality-preferred TMDC nanotubes.

cond-mat.mtrl-sci↗
arXiv2026

Can RL Improve Generalization of LLM Agents? An Empirical Study

Reinforcement fine-tuning (RFT) has shown promise for training LLM agents to perform multi-turn decision-making based on environment feedback. However, most existing evaluations remain largely in-domain: training and testing are conducted in the same environment or even on the same tasks. In real-world deployment, agents may operate in unseen environments with different background knowledge, observation spaces, and action interfaces. To characterize the generalization profile of RFT under such shifts, we conduct a systematic study along three axes: (1) within-environment generalization across task difficulty, (2) cross-environment transfer to unseen environments, and (3) sequential multi-environment training to quantify transfer and forgetting. Our results show that RFT generalizes well across task difficulty within an environment, but exhibits weaker transfer to unseen environments, which correlates with shifts in both semantic priors and observation/action interfaces. In contrast, sequential training yields promising downstream gains with minimal upstream forgetting, and mixture training across environments improves the overall balance. We further provide detailed analyses and deeper insights, and hope our work helps the community develop and deploy generalizable LLM agents.

cs.AI↗
arXiv2026

Scalable Training of Mixture-of-Experts Models with Megatron Core

Scaling Mixture-of-Experts (MoE) training introduces systems challenges absent in dense models. Because each token activates only a subset of experts, this sparsity allows total parameters to grow much faster than per-token computation, creating coupled constraints across memory, communication, and computation. Optimizing one dimension often shifts pressure to another, demanding co-design across the full system stack. We address these challenges for MoE training through integrated optimizations spanning memory (fine-grained recomputation, offloading, etc.), communication (optimized dispatchers, overlapping, etc.), and computation (Grouped GEMM, fusions, CUDA Graphs, etc.). The framework also provides Parallel Folding for flexible multi-dimensional parallelism, low-precision training support for FP8 and NVFP4, and efficient long-context training. On NVIDIA GB300 and GB200, it achieves 1,233/1,048 TFLOPS/GPU for DeepSeek-V3-685B and 974/919 TFLOPS/GPU for Qwen3-235B. As a performant, scalable, and production-ready open-source solution, it has been used across academia and industry for training MoE models ranging from billions to trillions of parameters on clusters scaling up to thousands of GPUs. This report explains how these techniques work, their trade-offs, and their interactions at the systems level, providing practical guidance for scaling MoE models with Megatron Core.

cs.DC↗
arXiv2026

Ab initio simulation of the first-order proton-ordering transition in water ice

Proton ordering in water ice is a paradigmatic order-disorder transition in a locally constrained system. The ice rules require exactly two hydrogens close to each oxygen, restricting the disorder to an exponentially large yet strongly correlated manifold of hydrogen-bond configurations. Within this constrained space, meV-scale energy differences drive the transition from disordered ice Ih to ordered ice XI, while distinct configurations are separated by eV-scale barriers. These barriers hinder equilibration in experiments, and efficient sampling of this space with the required energy accuracy has remained a long-standing challenge in simulation. We address this by combining a machine learning interatomic potential with loop updates that preserve the ice rules and continuous updates of atomic coordinates, enabling equilibrium sampling with ab initio accuracy and capturing configurational entropic effects. In systems of up to 360 water molecules, with over 10^6 samples retained per temperature point, the simulations reveal clear first-order transition signatures at 83 K: a negative Binder cumulant, a bimodal potential energy distribution, and a sharp step in the lattice aspect ratio. Nuclear quantum effects are estimated to lower the transition temperature by approximately 20 K, bringing the prediction closer to the experimental value of 72 K.

cond-mat.mtrl-sci↗
arXiv2026

Prolate-oblate shape competition and impact on charge radii in Bk isotopes

The nuclear charge radius provides a fundamental probe of nuclear structure, yet experimental data remain rare in the actinide region. Using the deformed relativistic Hartree-Bogoliubov theory in continuum (DRHBc) with the PC-PK1 functional, we carry out a systematic investigation of prolate-oblate shape competition in odd-$A$ Bk isotopes. Deformation is found to play an important role in the description of charge radii $r_c$ by extending the density distribution. Notably, $r_c$ exhibits a distinct shape dependence: for a given absolute quadrupole deformation $|β_2|$, oblate shapes yield larger charge radii than their prolate counterparts in well-deformed nuclei near the mid-shell region, where the empirical formula $r_c(β_2) = \left(1 + \frac{5}{4π}|β_2|^2\right) r_c(0)$ fails to capture the observed behavior. This enhancement is attributed to a central depression (or ``bubble" structure) in the proton density, which microscopically originates from the non-occupation of the spherical $3s_{1/2}$($Ω=1/2$) orbital in oblate minima. These findings establish a clear microscopic connection between nuclear shape, single-particle occupancy, and nuclear size.

nucl-th↗
arXiv2026

Physics and computation: An insight from non-Hermitian quantum computing

We elucidate the profound connection between physics and computation by proposing and examining the model of the non-Hermitian quantum computer (NQC). In addition to conventional quantum gates such as the Hadamard, phase, and CNOT gates, this model incorporates a non-unitary quantum gate $G$. We show that NQC is extraordinarily powerful, capable of solving not only all NP problems but also all problems within the complexity class $\text{P}^{\sharp\text{P}}$ in polynomial time. We investigate two physical schemes for implementing the non-unitary gate $G$ and find that the remarkable computational power of NQC originates from the exponentially large physical resources required.

quant-ph↗
arXiv2026

Why Reinforcement Fine-Tuning Enables MLLMs Preserve Prior Knowledge Better: A Data Perspective

Post-training algorithms such as Supervised Fine-Tuning (SFT) and Reinforcement Fine-Tuning (RFT) are widely used to adapt (multimodal) large language models to downstream tasks. While effective at task adaptation, their impact on retaining prior knowledge remains unclear. In this paper, we introduce jigsaw puzzles as a novel task absent from existing pretraining corpora and systematically study the behavior of SFT and RFT on the open-source Qwen2.5-VL series. Our experiments reveal a sharp trade-off: SFT enables rapid task acquisition but leads to catastrophic forgetting, whereas RFT learns more slowly but better maintains prior knowledge. We study this phenomenon through learning dynamics by examining both the magnitude and direction of how training data influence prior knowledge. Our analysis shows that RFT mainly reinforces correct samples naturally aligned with the base model's probability landscape, leading to weaker interference with prior knowledge. Moreover, training on RFT-simulated rollouts, which exert a smaller magnitude of influence and are better aligned in direction to prior knowledge, allows SFT to preserve prior knowledge better while rapidly learning new tasks. We further validate our framework on Qwen2.5 post-training in math and scientific QA, observing consistent forgetting and learning-dynamics trends. These findings suggest that the distribution of post-training data, rather than algorithmic differences alone, plays a central role in forgetting, and highlight RFT as a promising ingredient for stable continual post-training.

cs.CL↗