Searcharxiv⌕ Search

arXiv subjects

Haozhe Wang

Publications and source records attributed to Haozhe Wang.

At least 37 records · Page 2Linked to original sources
arXiv2026

Bad Seeing or Bad Thinking? Rewarding Perception for Multimodal Reasoning

Achieving robust perception-reasoning synergy is a central goal for advanced Vision-Language Models (VLMs). Recent advancements have pursued this goal via architectural designs or agentic workflows. However, these approaches are often limited by static textual reasoning or complicated by the significant compute and engineering burden of external agentic complexity. Worse, this heavy investment does not yield proportional gains, often witnessing a "seesaw effect" on perception and reasoning. This motivates a fundamental rethinking of the true bottleneck. In this paper, we argue that the root cause of this trade-off is an ambiguity in modality credit assignment: when a VLM fails, is it due to flawed perception ("bad seeing") or flawed logic ("bad thinking")? To resolve this, we introduce a reinforcement learning framework that improves perception-reasoning synergy by reliably rewarding the perception fidelity. We explicitly decompose the generation process into interleaved perception and reasoning steps. This decoupling enables targeted supervision on perception. Crucially, we introduce Perception Verification (PV), leveraging a "blindfolded reasoning" proxy to reward perceptual fidelity independently of reasoning outcomes. Furthermore, to scale training across free-form VL tasks, we propose Structured Verbal Verification, which replaces high-variance LLM judging with structured algorithmic execution. These techniques are integrated into a Modality-Aware Credit Assignment (MoCA) mechanism, which routes rewards to the specific source of error -- either bad seeing or bad thinking -- enabling a single VLM to achieve simultaneous performance gains across a wide task spectrum.

cs.AI↗
arXiv2026

Antiferromagnetic Ordering Enhanced Magnetic Damping in Mn2Au/CoFeB Bilayers

Antiferromagnets (AFMs) hold significant potential for spintronic devices owing to their insensitivity to external magnetic fields and the absence of stray fields. Beyond these inherent advantages, an AFM can manipulate the magnetic dynamics of a ferromagnet (FM) layer in AFM/FM bilayers, whereas the mechanism of such manipulation remains controversial. Here, we investigate the magnetic dynamics of AFM/FM Mn2Au/CoFeB bilayers via Ferromagnetic Resonance (FMR). It is found that the Néel temperature of 2-nm-thick Mn2Au is as low as ~40 K, in sharp contrast to that of bulk Mn2Au, which exceeds 1000 K. In the Mn2Au(2 nm)/CoFeB(4 nm) bilayer, the magnetic damping $α$ of the CoFeB layer increases from 0.013 to 0.047 as temperature decreases from 160 K to 10 K, accompanied by a synchronous increase in the exchange coupling field H_rot. Such an increase in $α$ is attributed to the enhanced spin angular momentum transfer from CoFeB to Mn2Au, mediated through AFM-FM exchange coupling between Mn2Au and CoFeB, which is enhanced by the Mn2Au antiferromagnetic ordering as the temperature decreases. Our study provides deeper insights into AFM/FM dynamics and spintronic storage technology.

cond-mat.mtrl-sci↗
arXiv2026

Inversion diameter and treewidth

In an oriented graph $\overrightarrow{G}$, the inversion of a subset $X$ of vertices is the operation that reverses the orientation of all arcs with both end-vertices in $X$. The inversion graph of a graph $G$, denoted by $\mathcal{I}(G)$, is the graph whose vertices are orientations of $G$ in which two orientations $\overrightarrow{G_1}$ and $\overrightarrow{G_2}$ are adjacent if and only if there is an inversion transforming $\overrightarrow{G_1}$ into $\overrightarrow{G_2}$.The inversion diameter of a graph $G$ is the diameter of its inversion graph $\mathcal{I}(G)$, denoted by $\mathrm{diam}(\mathcal{I}(G))$.Havet, Hörsch, and Rambaud~(2024) first proved that for $G$ of treewidth $k$, $\mathrm{diam}(\mathcal{I}(G)) \le 2k$, and that there are graphs of treewidth $k$ with inversion diameter $k+2$.In this paper, we construct graphs of treewidth $k$ with inversion diameter $2k$, which implies that the previous upper bound $\mathrm{diam}(\mathcal{I}(G)) \le 2k$ is tight.Moreover, for graphs with maximum degree $Δ$, Havet, Hörsch, and Rambaud~(2024) proved $\mathrm{diam}(\mathcal{I}(G)) \le 2Δ-1$ and conjectured that $\mathrm{diam}(\mathcal{I}(G)) \le Δ$. We prove the conjecture when $Δ=3$ with the help of computer calculations.

math.CO↗
arXiv2026

Detector-Empowered Video Large Language Model for Efficient Spatio-Temporal Grounding

Multimodal large language models (MLLMs) are rapidly expanding from general video understanding to finer-grained understanding such as spatio-temporal video grounding (STVG) and reasoning. In these tasks, an MLLM must localize the user-queried target in time and space and take the results as evidence for reasoning. Existing MLLM methods mainly follow two paradigms: (1) Direct Localization, which outputs STVG results with extra alignment modules or specialized decoders; and (2) Candidate-based Selection, which first constructs tube-level candidates and then selects the relevant one by an MLLM. However, both suffer from a serious efficiency bottleneck: the former incurs linearly growing decoding cost as the queried temporal span increases, while the latter relies on costly candidate construction. To break this bottleneck, we propose DEViL, a detector-empowered Video-LLM with a simple key idea: offloading dense spatial grounding from the MLLM to a fully parallelizable, well-trained detector. Specifically, DEViL distills the query into a detector-compatible reference-semantic token, which replaces the detector's text embedding to enable spatial grounding in a single pass. Then, we design temporal consistency regularization to match objects across frames and enforce their coherence over time. In this way, DEViL avoids long coordinate decoding and heavy candidate pipelines. Extensive experiments show that DEViL achieves strong performance (43.1% m_vIoU on HC-STVG) with superior efficiency (14.33 FPS), while preserving the general reasoning capacity of the MLLM backbone.

cs.CV↗
arXiv2026

RationalRewards: Reasoning Rewards Scale Visual Generation Both Training and Test Time

Most reward models for visual generation reduce rich human judgments to a single unexplained score, discarding the reasoning that underlies preference. We show that teaching reward models to produce explicit, multi-dimensional critiques before scoring transforms them from passive evaluators into active optimization tools, improving generators in two complementary ways: at training time, structured rationales provide interpretable, fine-grained rewards for reinforcement learning; at test time, a Generate-Critique-Refine loop turns critiques into targeted prompt revisions that improve outputs without any parameter updates. To train such a reward model without costly rationale annotations, we introduce Preference-Anchored Rationalization (PARROT), a principled framework that recovers high-quality rationales from readily available preference data through anchored generation, consistency filtering, and distillation. The resulting model, RationalRewards (8B), achieves state-of-the-art preference prediction among open-source reward models, competitive with Gemini-2.5-Pro, while using 10-20x less training data than comparable baselines. As an RL reward, it consistently improves text-to-image and image-editing generators beyond scalar alternatives. Most strikingly, its test-time critique-and-refine loop matches or exceeds RL-based fine-tuning on several benchmarks, suggesting that structured reasoning can unlock latent capabilities in existing generators that suboptimal prompts fail to elicit.

cs.AI↗
arXiv2026

StructEval: Benchmarking LLMs' Capabilities to Generate Structural Outputs

As Large Language Models (LLMs) become integral to software development workflows, their ability to generate structured outputs has become critically important. We introduce StructEval, a comprehensive benchmark for evaluating LLMs' capabilities in producing both non-renderable (JSON, YAML, CSV) and renderable (HTML, React, SVG) structured formats. Unlike prior benchmarks, StructEval systematically evaluates structural fidelity across diverse formats through two paradigms: 1) generation tasks, producing structured output from natural language prompts, and \textbf{2)} conversion tasks, translating between structured formats. Our benchmark encompasses 18 formats and 44 types of task, with novel metrics for format adherence and structural correctness. Results reveal significant performance gaps-even state-of-the-art models like o1-mini achieve only 75.58 average score, with open-source alternatives lagging approximately 10 points behind. We find generation tasks more challenging than conversion tasks, and producing correct visual content more difficult than generating text-only structures.

cs.SE↗
arXiv2026

LongCat-Next: Lexicalizing Modalities as Discrete Tokens

The prevailing Next-Token Prediction (NTP) paradigm has driven the success of large language models through discrete autoregressive modeling. However, contemporary multimodal systems remain language-centric, often treating non-linguistic modalities as external attachments, leading to fragmented architectures and suboptimal integration. To transcend this limitation, we introduce Discrete Native Autoregressive (DiNA), a unified framework that represents multimodal information within a shared discrete space, enabling a consistent and principled autoregressive modeling across modalities. A key innovation is the Discrete Native Any-resolution Visual Transformer (dNaViT), which performs tokenization and de-tokenization at arbitrary resolutions, transforming continuous visual signals into hierarchical discrete tokens. Building on this foundation, we develop LongCat-Next, a native multimodal model that processes text, vision, and audio under a single autoregressive objective with minimal modality-specific design. As an industrial-strength foundation model, it excels at seeing, painting, and talking within a single framework, achieving strong performance across a wide range of multimodal benchmarks. In particular, LongCat-Next addresses the long-standing performance ceiling of discrete vision modeling on understanding tasks and provides a unified approach to effectively reconcile the conflict between understanding and generation. As an attempt toward native multimodality, we open-source the LongCat-Next and its tokenizers, hoping to foster further research and development in the community. GitHub: https://github.com/meituan-longcat/LongCat-Next

cs.CV↗
arXiv2026

Understanding by Reconstruction: Reversing the Software Development Process for LLM Pretraining

While Large Language Models (LLMs) have achieved remarkable success in code generation, they often struggle with the deep, long-horizon reasoning required for complex software engineering. We attribute this limitation to the nature of standard pre-training data: static software repositories represent only the terminal state of an intricate intellectual process, abstracting away the intermediate planning, debugging, and iterative refinement. To bridge this gap, we propose a novel paradigm: understanding via reconstruction. We hypothesize that reverse-engineering the latent agentic trajectories -- the planning, reasoning, and debugging steps -- behind static repositories provides a far richer supervision signal than raw code alone. To operationalize this, we introduce a framework that synthesizes these trajectories using a multi-agent simulation. This process is grounded in the structural realities of the source repositories (e.g., dependency graphs and file hierarchies) to ensure fidelity. Furthermore, to guarantee the logical rigor of the synthetic data, we employ a search-based optimization technique that iteratively refines the Chain-of-Thought (CoT) reasoning to maximize the likelihood of the ground-truth code. Empirical results demonstrate that continuous pre-training on these reconstructed trajectories significantly enhances Llama-3-8B's performance across diverse benchmarks, including long-context understanding, coding proficiency, and agentic capabilities.

cs.SE↗
arXiv2026

SWE-QA-Pro: A Representative Benchmark and Scalable Training Recipe for Repository-Level Code Understanding

Agentic repository-level code understanding is essential for automating complex software engineering tasks, yet the field lacks reliable benchmarks. Existing evaluations often overlook the long tail topics and rely on popular repositories where Large Language Models (LLMs) can cheat via memorized knowledge. To address this, we introduce SWE-QA-Pro, a benchmark constructed from diverse, long-tail repositories with executable environments. We enforce topical balance via issue-driven clustering to cover under-represented task types and apply a rigorous difficulty calibration process: questions solvable by direct-answer baselines are filtered out. This results in a dataset where agentic workflows significantly outperform direct answering (e.g., a ~13-point gap for Claude Sonnet 4.5), confirming the necessity of agentic codebase exploration. Furthermore, to tackle the scarcity of training data for such complex behaviors, we propose a scalable synthetic data pipeline that powers a two-stage training recipe: Supervised Fine-Tuning (SFT) followed by Reinforcement Learning from AI Feedback (RLAIF). This approach allows small open models to learn efficient tool usage and reasoning. Empirically, a Qwen3-8B model trained with our recipe surpasses GPT-4o by 2.3 points on SWE-QA-Pro and substantially narrows the gap to state-of-the-art proprietary models, demonstrating both the validity of our evaluation and the effectiveness of our agentic training workflow.

cs.SE↗
arXiv2026

Asynchronous Verified Semantic Caching for Tiered LLM Architectures

Large language models (LLMs) now sit in the critical path of search, assistance, and agentic workflows, making semantic caching essential for reducing inference cost and latency. Production deployments typically use a tiered static-dynamic design: a static cache of curated, offline vetted responses mined from logs, backed by a dynamic cache populated online. In practice, both tiers are commonly governed by a single embedding similarity threshold, which induces a hard tradeoff: conservative thresholds miss safe reuse opportunities, while aggressive thresholds risk serving semantically incorrect responses. We introduce Krites, an asynchronous, LLM-judged caching policy that expands static coverage without changing serving decisions. On the critical path, Krites behaves exactly like a standard static threshold policy. When the nearest static neighbor of the prompt falls just below the static threshold, Krites asynchronously invokes an LLM judge to verify whether the static response is acceptable for the new prompt. Approved matches are promoted into the dynamic cache, allowing future repeats and paraphrases to reuse curated static answers and expanding static reach over time. In trace-driven simulations on conversational and search workloads, Krites increases the fraction of requests served with curated static answers (direct static hits plus verified promotions) by up to3.9 times for conversational traffic and search-style queries relative to tuned baselines, with unchanged critical path latency.

cs.IR↗
arXiv2026

Adaptive RAN Slicing Control via Reward-Free Self-Finetuning Agents

The integration of Generative AI models into AI-native network systems offers a transformative path toward achieving autonomous and adaptive control. However, the application of such models to continuous control tasks is impeded by intrinsic architectural limitations, including finite context windows, the lack of explicit reward signals, and the degradation of the long context. This paper posits that the key to unlocking robust continuous control is enabling agents to internalize experience by distilling it into their parameters, rather than relying on prompt-based memory. To this end, we propose a novel self-finetuning framework that enables agentic systems to learn continuously through direct interaction with the environment, bypassing the need for handcrafted rewards. Our framework implements a bi-perspective reflection mechanism that generates autonomous linguistic feedback to construct preference datasets from interaction history. A subsequent preference-based fine-tuning process distills long-horizon experiences into the model's parameters. We evaluate our approach on a dynamic Radio Access Network (RAN) slicing task, a challenging multi-objective control problem that requires the resolution of acute trade-offs between spectrum efficiency, service quality, and reconfiguration stability under volatile network conditions. Experimental results show that our framework outperforms standard Reinforcement Learning (RL) baselines and existing Large Language Model (LLM)-based agents in sample efficiency, stability, and multi-metric optimization. These findings demonstrate the potential of self-improving generative agents for continuous control tasks, paving the way for future AI-native network infrastructure.

cs.AI↗
arXiv2026

Broadband multilayer metasurface absorbers with MXene resonators and topology optimized substrates

We present the synthesis of broadband multilayer metamaterial absorbers (MMA) based on MXenes, which are novel two-dimensional conductive materials with higher ohmic losses than copper. MXene resonator of different conductivity can be implemented at each layer or across the same layer, offering increased design flexibility. We examine the possibility of utilizing topology-optimized stereolithography (SLA) 3D-printed substrates as a complementary means for enhancing absorption. Combining MXenes with topology optimized 3D-printable structures paves the way for realizing MMAs of enhanced absorption.

physics.app-ph↗
arXiv2026

RelaCtrl: Relevance-Guided Efficient Control for Diffusion Transformers

The Diffusion Transformer plays a pivotal role in advancing text-to-image and text-to-video generation, owing primarily to its inherent scalability. However, existing controlled diffusion transformer methods incur significant parameter and computational overheads and suffer from inefficient resource allocation due to their failure to account for the varying relevance of control information across different transformer layers. To address this, we propose the Relevance-Guided Efficient Controllable Generation framework, RelaCtrl, enabling efficient and resource-optimized integration of control signals into the Diffusion Transformer. First, we evaluate the relevance of each layer in the Diffusion Transformer to the control information by assessing the "ControlNet Relevance Score"-i.e., the impact of skipping each control layer on both the quality of generation and the control effectiveness during inference. Based on the strength of the relevance, we then tailor the positioning, parameter scale, and modeling capacity of the control layers to reduce unnecessary parameters and redundant computations. Additionally, to further improve efficiency, we replace the self-attention and FFN in the commonly used copy block with the carefully designed Two-Dimensional Shuffle Mixer (TDSM), enabling efficient implementation of both the token mixer and channel mixer. Both qualitative and quantitative experimental results demonstrate that our approach achieves superior performance with only 15% of the parameters and computational complexity compared to PixArt-delta.

cs.CV↗
arXiv2026

Physical Human-Robot Interaction for Grasping in Augmented Reality via Rigid-Soft Robot Synergy

Hybrid rigid-soft robots combine the precision of rigid manipulators with the compliance and adaptability of soft arms, offering a promising approach for versatile grasping in unstructured environments. However, coordinating hybrid robots remains challenging, due to difficulties in modeling, perception, and cross-domain kinematics. In this work, we present a novel augmented reality (AR)-based physical human-robot interaction framework that enables direct teleoperation of a hybrid rigid-soft robot for simple reaching and grasping tasks. Using an AR headset, users can interact with a simulated model of the robotic system integrated into a general-purpose physics engine, which is superimposed on the real system, allowing simulated execution prior to real-world deployment. To ensure consistent behavior between the virtual and physical robots, we introduce a real-to-simulation parameter identification pipeline that leverages the inherent geometric properties of the soft robot, enabling accurate modeling of its static and dynamic behavior as well as the control system's response.

cs.RO↗
arXiv2026

Dr. Bench: A Multidimensional Evaluation for Deep Research Agents, from Answers to Reports

As an embodiment of intelligence evolution toward interconnected architectures, Deep Research Agents (DRAs) systematically exhibit the capabilities in task decomposition, cross-source retrieval, multi-stage reasoning, information integration, and structured output, which markedly enhance performance on complex and open-ended tasks. However, existing benchmarks remain deficient in evaluation dimensions, response format, and scoring mechanisms, limiting their effectiveness in assessing such agents. This paper introduces Dr. Bench, a multidimensional evaluation framework tailored to DRAs and long-form report-style responses. The benchmark comprises 214 expert-curated challenging tasks across 10 broad domains, each accompanied by manually constructed reference bundles to support composite evaluation. This framework incorporates metrics for semantic quality, topical focus, and retrieval trustworthiness, enabling a comprehensive evaluation of long reports generated by DRAs. Extensive experimentation confirms the superior performance of mainstream DRAs over web-search-tool-augmented reasoning models, yet reveals considerable scope for further improvement. This study provides a robust foundation for capability assessment, architectural refinement, and paradigm advancement of DRAs.

cs.AI↗
arXiv2026

EvoCUA: Evolving Computer Use Agents via Learning from Scalable Synthetic Experience

The development of native computer-use agents (CUA) represents a significant leap in multimodal AI. However, their potential is currently bottlenecked by the constraints of static data scaling. Existing paradigms relying primarily on passive imitation of static datasets struggle to capture the intricate causal dynamics inherent in long-horizon computer tasks. In this work, we introduce EvoCUA, a native computer use agentic model. Unlike static imitation, EvoCUA integrates data generation and policy optimization into a self-sustaining evolutionary cycle. To mitigate data scarcity, we develop a verifiable synthesis engine that autonomously generates diverse tasks coupled with executable validators. To enable large-scale experience acquisition, we design a scalable infrastructure orchestrating tens of thousands of asynchronous sandbox rollouts. Building on these massive trajectories, we propose an iterative evolving learning strategy to efficiently internalize this experience. This mechanism dynamically regulates policy updates by identifying capability boundaries -- reinforcing successful routines while transforming failure trajectories into rich supervision through error analysis and self-correction. Empirical evaluations on the OSWorld benchmark demonstrate that EvoCUA achieves a success rate of 56.7%, establishing a new open-source state-of-the-art. Notably, EvoCUA significantly outperforms the previous best open-source model, OpenCUA-72B (45.0%), and surpasses leading closed-weights models such as UI-TARS-2 (53.1%). Crucially, our results underscore the generalizability of this approach: the evolving paradigm driven by learning from experience yields consistent performance gains across foundation models of varying scales, establishing a robust and scalable path for advancing native agent capabilities.

cs.AI↗
arXiv2026

Forbidden second harmonics in centrosymmetric bilayer crystals

Optical spectroscopy based on second-order nonlinearity is a critical technique for characterizing two-dimensional (2D) crystals as well as bioimaging and quantum optics. It is generally believed that second-harmonic generation (SHG) in centrosymmetric crystals, such as graphene and other bilayer 2D crystals, is negligible without externally breaking the inversion symmetry. Here, we show that with a new homodyne detection technique, we can apparently circumvent this symmetry-imposed constraint and observe robust SHG in pristine centrosymmetric crystals, without any symmetry-breaking field. With its exceptional sensitivity, we resolve polarization-resolved SHG in bilayer hexagonal boron nitride (h-BN), bilayer 2H-WSe$_2$, and remarkably, Bernal-stacked bilayer graphene, allowing us to unambiguously identify the crystallographic orientation in these crystals via SHG for the first time. We also demonstrate that the new technique can be used to non-invasively detect uniaxial strain and optical geometric phase in these crystals. The observed SHG in our experiments is attributed to second-order nonlinearity in the quadrupole channel, which is controlled by the presence of the $C_2$ symmetry instead of the inversion symmetry. Our new technique expands the capability of nonlinear optical spectroscopy to encompass a large class of centrosymmetric materials that could never be measured before, and can be used for quantum sensing of moiré materials and twisted epitaxial films.

cond-mat.mes-hall↗
arXiv2026

Frustrated Magnetism in FeGe$_3$O$_4$ with a Chiral Trillium Network

The discovery of new magnetic ground states in geometrically frustrated lattices remains a central challenge in materials science. Here, we report the synthesis, structural characterization, and frustrated magnetic properties of FeGe$_3$O$_4$, a newly identified compound that crystallizes in the noncentrosymmetric cubic space group $P2_13$. In this structure, Fe atoms form an intricate double-trillium lattice with nearest-neighbor Fe--Fe distances of $\sim$4.2~Å, while Ge$^{2+}$ ions mediate magnetic interactions through Fe-Ge-Fe pathways. Field-dependent magnetization at 2~K shows a pronounced nonlinearity, reaching a maximum moment of 2.55(3)~$μ_\mathrm{B}$/Fe$^{2+}$ at 70~kOe without evidence of saturation. Magnetic susceptibility, heat capacity, and neutron scattering collectively reveal the onset of short-range magnetic interactions near 5~K, with no long-range ordering detected down to 0.06~K. Specific heat measurements demonstrate strong frustration: only $\sim$34\% of the expected magnetic entropy is recovered at 2.4~K. Taken together, these results establish FeGe$_3$O$_4$ as a rare example of a geometrically frustrated trillium-lattice magnet, offering a promising platform for exploring exotic quantum magnetic phenomena.

cond-mat.str-el↗