Search arXiv⌕ Search

arXiv subjects

Hao Yan

Publications and source records attributed to Hao Yan.

At least 55 records · Page 3Linked to original sources
arXiv2026

DocSeeker: Structured Visual Reasoning with Evidence Grounding for Long Document Understanding

Existing Multimodal Large Language Models (MLLMs) suffer from significant performance degradation on the long document understanding task as document length increases. This stems from two fundamental challenges: 1) a low Signal-to-Noise Ratio (SNR), with crucial evidence buried in irrelevant pages; and 2) supervision scarcity, as datasets offering only final short answers provide a weak learning signal. In this paper, we address these challenges by proposing a paradigm that requires the model to execute a structured Analysis, Localization and Reasoning workflow. To instill this capability, we design a two-stage training framework: we first perform Supervised Fine-Tuning on high-quality data generated via an efficient knowledge distillation strategy. Subsequently, we employ an Evidence-aware Group Relative Policy Optimization which jointly optimizes for both evidence localization and answer accuracy. Additionally, we introduce a Evidence-Guided Resolution Allocation strategy to mitigate memory constraints of training on multi-pages documents. Extensive experiments demonstrate that DocSeeker achieves superior performance on both in-domain and out-of-domain tasks. We show it robustly generalizes from short-page training to ultra-long documents and is naturally synergistic with visual Retrieval-Augmented Generation systems, serving as a solid foundation for their implementation.

cs.AI↗
arXiv2026

Tstars-Tryon 1.0: Robust and Realistic Virtual Try-On for Diverse Fashion Items

Recent advances in image generation and editing have opened new opportunities for virtual try-on. However, existing methods still struggle to meet complex real-world demands. We present Tstars-Tryon 1.0, a commercial-scale virtual try-on system that is robust, realistic, versatile, and highly efficient. First, our system maintains a high success rate across challenging cases like extreme poses, severe illumination variations, motion blur, and other in-the-wild conditions. Second, it delivers highly photorealistic results with fine-grained details, faithfully preserving garment texture, material properties, and structural characteristics, while largely avoiding common AI-generated artifacts. Third, beyond apparel try-on, our model supports flexible multi-image composition (up to 6 reference images) across 8 fashion categories, with coordinated control over person identity and background. Fourth, to overcome the latency bottlenecks of commercial deployment, our system is heavily optimized for inference speed, delivering near real-time generation for a seamless user experience. These capabilities are enabled by an integrated system design spanning end-to-end model architecture, a scalable data engine, robust infrastructure, and a multi-stage training paradigm. Extensive evaluation and large-scale product deployment demonstrate that Tstars-Tryon1.0 achieves leading overall performance. To support future research, we also release a comprehensive benchmark. The model has been deployed at an industrial scale on the Taobao App, serving millions of users with tens of millions of requests.

cs.CV↗
arXiv2026

Continuous-Time Distribution Matching for Few-Step Diffusion Distillation

Step distillation has become a leading technique for accelerating diffusion models, among which Distribution Matching Distillation (DMD) and Consistency Distillation are two representative paradigms. While consistency methods enforce self-consistency along the full PF-ODE trajectory to steer it toward the clean data manifold, vanilla DMD relies on sparse supervision at a few predefined discrete timesteps. This restricted discrete-time formulation and mode-seeking nature of the reverse KL divergence tends to exhibit visual artifacts and over-smoothed outputs, often necessitating complex auxiliary modules -- such as GANs or reward models -- to restore visual fidelity. In this work, we introduce Continuous-Time Distribution Matching (CDM), migrating the DMD framework from discrete anchoring to continuous optimization for the first time. CDM achieves this through two continuous-time designs. First, we replace the fixed discrete schedule with a dynamic continuous schedule of random length, so that distribution matching is enforced at arbitrary points along sampling trajectories rather than only at a few fixed anchors. Second, we propose a continuous-time alignment objective that performs active off-trajectory matching on latents extrapolated via the student's velocity field, improving generalization and preserving fine visual details. Extensive experiments on different architectures, including SD3-Medium and Longcat-Image, demonstrate that CDM provides highly competitive visual fidelity for few-step image generation without relying on complex auxiliary objectives. Code is available at https://github.com/byliutao/cdm.

cs.CV↗
arXiv2026

Task Vector Geometry Underlies Dual Modes of Task Inference in Transformers

Transformers are effective at inferring the latent task from context via two inference modes: recognizing a task seen during training, and adapting to a novel one. Recent interpretability studies have identified from middle-layer representations task-specific directions, or task vectors, that steer model behavior. However, a lack of rigorous foundations hinders connecting internal representations to external model behavior: existing work fails to explain how task-vector geometry is shaped by the training distribution, and what geometry enables out-of-distribution (OOD) generalization. In this paper, we study these questions in a controlled synthetic setting by training small transformers from scratch on latent-task sequence distributions, which allows a principled mathematical characterization. We show that two inference modes can coexist within a single model. In-distribution behavior is governed by Bayesian task retrieval, implemented internally through convex combinations of learned task vectors. OOD behavior, by contrast, arises through extrapolative task learning, whose representations occupy a subspace nearly orthogonal to the task-vector subspace. Taken together, our results suggest that task-vector geometry, training distributions, and generalization behaviors are closely related.

cs.LG↗
arXiv2026

PDF-WuKong: A Large Multimodal Model for Efficient Long PDF Reading with End-to-End Sparse Sampling

Multimodal document understanding is a challenging task to process and comprehend large amounts of textual and visual information. Recent advances in Large Language Models (LLMs) have significantly improved the performance of this task. However, existing methods typically focus on either plain text or a limited number of document images, struggling to handle long PDF documents with interleaved text and images, especially for academic papers. In this paper, we introduce PDF-WuKong, a multimodal large language model (MLLM) that is designed to enhance multimodal question-answering (QA) for long PDF documents. PDF-WuKong incorporates a sparse sampler that operates on both text and image representations, significantly improving the efficiency and capability of the MLLM. The sparse sampler selects the paragraphs or diagrams most pertinent to user queries. To effectively train and evaluate our model, we construct PaperPDF, a dataset consisting of a broad collection of English and Chinese academic papers. Multiple strategies are proposed to build high-quality 1.1 million QA pairs along with their corresponding evidence sources. Experimental results demonstrate the superiority and high efficiency of our approach over other models on the task of long multimodal document understanding, surpassing proprietary products by an average of 8.6% on F1. Our code and dataset will be released at https://github.com/yh-hust/PDF-Wukong.

cs.CV↗
arXiv2026

Complementary-polarity double-layer LiTaO3 resonators for symmetry-selective SH2 excitation with ultrahigh electromechanical coupling (kt^2 = 25.7%)

We report a structurally simple double-layer lithium tantalate (LiTaO3) bulk acoustic resonator that enables symmetry-selective excitation of the second-order thickness-shear (SH2) mode with ultrahigh electromechanical coupling. Two 31 deg Y-oriented single-crystal LiTaO3 films are rotation-bonded with complementary polarization (+X/-X) and driven by a longitudinal electric field. Matching between the effective piezoelectric symmetry and the SH2 mode yields an effective electromechanical coupling coefficient of kt^2 = 25.7% at 5.24 MHz. To our knowledge, this is the highest kt^2 reported for a LiTaO3 resonator architecture to date. The measured response is dominated by the target SH2 mode, with only weak parasitic features in the operating band. The structure is also tunable: the resonance frequency and coupling can be adjusted through geometric parameters while maintaining stable modal behavior, indicating good process tolerance. Finite-element analysis further suggests straightforward frequency scaling beyond 5 GHz by reducing the film and electrode thickness while preserving approximately 25% kt^2. In addition, introducing a SiO2 compensation layer is predicted to improve the temperature coefficient of frequency to approximately -25 ppm/deg C. These results establish complementary-polarity double-layer LiTaO3 as a practical platform for high-coupling, spurious-suppressed acoustic resonators and provide a scalable route toward wideband ultrasonic resonators, filters, and related transducers.

physics.app-ph↗
arXiv2026

EquivFusion: Unifying Hardware Equivalence Checking from Algorithms to Netlists via MLIR

Ensuring functional consistency between high-level algorithmic models and low-level hardware implementations is a critical challenge, particularly as modern design flows increasingly span heterogeneous abstractions--from deep learning frameworks to hardware netlists. In this paper, we present EquivFusion, an end-to-end equivalence checking tool tailored for multi-modal circuit designs. Unlike traditional flows that rely on siloed tools or ad-hoc translation, EquivFusion leverages a verification-oriented MLIR lowering pipeline to unify diverse entry points, including PyTorch, C/C++, Chisel, Verilog, and gate-level netlists, into a common intermediate representation. This architecture enables automated, pairwise equivalence checking across diverse abstraction levels by rigorously translating designs into standard formal verification formats, i.e., SMT-LIB, BTOR2, AIGER. We demonstrate EquivFusion's feasibility to bridge the semantic gap between software specifications and hardware realizations, showcasing its effectiveness in facilitating "shift-left" formal verification for datapath-intensive hardware designs.

cs.AR↗
arXiv2026

Bi-Level Optimization for Single Domain Generalization

Generalizing from a single labeled source domain to unseen target domains, without access to any target data during training, remains a fundamental challenge in robust machine learning. We address this underexplored setting, known as Single Domain Generalization (SDG), by proposing BiSDG, a bi-level optimization framework that explicitly decouples task learning from domain modeling. BiSDG simulates distribution shifts through surrogate domains constructed via label-preserving transformations of the source data. To capture domain-specific context, we propose a domain prompt encoder that generates lightweight modulation signals to produce augmenting features via feature-wise linear modulation. The learning process is formulated as a bi-level optimization problem: the inner objective optimizes task performance under fixed prompts, while the outer objective maximizes generalization across the surrogate domains by updating the domain prompt encoder. We further develop a practical gradient approximation scheme that enables efficient bi-level training without second-order derivatives. Extensive experiments on various SGD benchmarks demonstrate that BiSDG consistently outperforms prior methods, setting new state-of-the-art performance in the SDG setting.

cs.LG↗
arXiv2026

Learning with Incomplete Context: Linear Contextual Bandits with Pretrained Imputation

The rise of large-scale pretrained models has made it feasible to generate predictive or synthetic features at low cost, raising the question of how to incorporate such surrogate predictions into downstream decision-making. We study this problem in the setting of online linear contextual bandits, where contexts may be complex, nonstationary, and only partially observed. In addition to bandit data, we assume access to an auxiliary dataset containing fully observed contexts--common in practice since such data are collected without adaptive interventions. We propose PULSE-UCB, an algorithm that leverages pretrained models trained on the auxiliary data to impute missing features during online decision-making. We establish regret guarantees that decompose into a standard bandit term plus an additional component reflecting pretrained model quality. In the i.i.d. context case with Hölder-smooth missing features, PULSE-UCB achieves near-optimal performance, supported by matching lower bounds. Our results quantify how uncertainty in predicted contexts affects decision quality and how much historical data is needed to improve downstream learning.

stat.ML↗
arXiv2026

Can LLMs Simulate Personas with Reversed Performance? A Systematic Investigation for Counterfactual Instruction Following in Math Reasoning Context

Large Language Models (LLMs) are now increasingly widely used to simulate personas in virtual environments, leveraging their instruction-following capability. However, we discovered that even state-of-the-art LLMs cannot simulate personas with reversed performance (e.g., student personas with low proficiency in educational settings), which impairs the simulation diversity and limits the practical applications of the simulated environments. In this work, using mathematical reasoning as a representative scenario, we propose the first benchmark dataset for evaluating LLMs on simulating personas with reversed performance, a capability that we dub "counterfactual instruction following". We evaluate both open-weight and closed-source LLMs on this task and find that LLMs, including the OpenAI o1 reasoning model, all struggle to follow counterfactual instructions for simulating reversedly performing personas. Intersectionally simulating both the performance level and the race population of a persona worsens the effect even further. These results highlight the challenges of counterfactual instruction following and the need for further research.

cs.CL↗
arXiv2026

Batch Prompting Suppresses Overthinking Reasoning Under Constraint: How Batch Prompting Suppresses Overthinking in Reasoning Models

Large Reasoning Models (LRMs) achieve strong performance through explicit chain-of-thought reasoning but suffer from \textit{overthinking}: generating excessive reasoning tokens even for trivial queries. {Beyond inflating cost, overthinking can be self-defeating: models enter recursive self-doubt loops that exhaust token budgets without producing an answer, causing API timeouts that directly hurt accuracy.} We present an empirical study showing that \textbf{batch prompting}, originally introduced for throughput optimization, effectively suppresses overthinking at inference time. Across 13 diverse benchmarks with DeepSeek-R1 and OpenAI-o1, batch prompting {reduces reasoning tokens by 76\% (2{,}950$\mapsto$710), on average, while preserving or improving accuracy}. Through behavioral analysis, we find that batching induces three beneficial effects: (1) it reduces per-query reasoning effort when multiple queries share a context; (2) it enables pattern induction, where models generalize from earlier examples to solve later ones; and (3) it suppresses hedging behavior (e.g., ``\texttt{wait,}'' ``\texttt{let me double-check}'') that signals metacognitive loops. We also show that explicit prompt constraints (``\texttt{Use no more than 100 tokens in thinking.}'') fail to reduce overthinking; models either ignore them or sacrifice accuracy. These findings reframe batch prompting as more than a cost optimization: it is a practical inference-time technique that improves efficiency and reliability without model modification.

cs.CL↗
arXiv2026

PS$^2$: Parameterized Control for Fine-Grained Student Proficiency Simulation

Understanding how students with different proficiency levels respond to educational materials is a critical issue within the field of AI for Education. However, acquiring sufficient real student response data for a robust evaluation is often hindered by cost, ethics, and security constraints. Consequently, LLM-based student proficiency simulation, especially prompt-based methods, has emerged as a practical alternative under data-scarce conditions. Despite their promise, current methods still exhibit limited controllability with coarse-grained proficiency representations, high sensitivity to prompt design, and the lack of calibration with academic performance. Therefore, we propose Parameterized Student Proficiency Simulation (PS$^2$), an unsupervised and parameterized model-level framework that simulates students with different proficiencies by interpolating between a strong upper-bound LLM and a weaker, cognitive error-informed lower-bound student LLM via a hybrid ratio. Specifically, the lower-bound model is constructed by fine-tuning the weaker LM to exhibit cognitive errors when responding to educational materials. To ensure alignment with target proficiency levels, PS$^2$ further calibrates the interpolation ratio with academic performance. Experiments on two public datasets demonstrate that PS$^2$ achieves finer-grained and consistent proficiency simulation compared to existing baselines, leading to superior performance in student behavior similarity and item difficulty prediction.

cs.CY↗
arXiv2026

UniX: Unifying Autoregression and Diffusion for Chest X-Ray Understanding and Generation

Despite recent progress, medical foundation models still struggle to unify visual understanding and generation, as these tasks have inherently conflicting goals: semantic abstraction versus pixel-level reconstruction. Existing approaches, typically based on parameter-shared autoregressive architectures, frequently lead to compromised performance in one or both tasks. To address this, we present UniX, a next-generation unified medical foundation model for chest X-ray understanding and generation. UniX decouples the two tasks into an autoregressive branch for understanding and a diffusion branch for high-fidelity generation. Crucially, a cross-modal self-attention mechanism is introduced to dynamically guide the generation process with understanding features. Coupled with a rigorous data cleaning pipeline and a multi-stage training strategy, this architecture enables synergistic collaboration between tasks while leveraging the strengths of diffusion models for superior generation. On two representative benchmarks, UniX achieves a 46.1% improvement in understanding performance (Micro-F1) and a 24.2% gain in generation quality (FD-RadDino), using only a quarter of the parameters of LLM-CXR. By achieving performance on par with task-specific models, our work establishes a scalable paradigm for synergistic medical image understanding and generation. Codes and models are available at https://github.com/ZrH42/UniX.

cs.CV↗
arXiv2025

Partially Observable Markov Decision Process Framework for Operating Condition Optimization Using Real-Time Degradation Signals

In many engineering systems, proper predictive maintenance and operational control are essential to increase efficiency and reliability while reducing maintenance costs. However, one of the major challenges is that many sensors are used for system monitoring. Analyzing these sensors simultaneously for better predictive maintenance optimization is often very challenging. In this paper, we propose a systematic decision-making framework to improve the system performance in manufacturing practice, considering the real-time degradation signals generated by multiple sensors. Specifically, we propose a partially observed Markov decision process (POMDP) model to generate the optimal capacity and predictive maintenance policies, given the fact that the observation of the system state is imperfect. Such work provides a systematic approach that focuses on jointly controlling the operating conditions and preventive maintenance utilizing the real-time machine deterioration signals by incorporating the degradation constraint and non-observable states. We apply this technique to the bearing degradation data and NASA aircraft turbofan engine dataset, demonstrating the effectiveness of the proposed method.

stat.AP↗
arXiv2025

Impedance-matched High-Overtone Thickness-Shear Bulk Acoustic Resonators with Scalable Mode Volume

High overtone bulk acoustic resonators are essential components in microwave signal processing and emerging quantum technologies; however, conventional designs suffer from limited impedance matching, spurious mode interference, and restricted scalability. Here we introduce a laterally excited high overtone thickness shear bulk acoustic resonator, abbreviated as X HTBAR, that overcomes these limitations through a fully planar excitation scheme. The X HTBAR employs a 3 micron thick 128 degree Y cut LiNbO3 piezoelectric film on a 500 micron high resistivity silicon substrate, enabling efficient excitation of thickness shear modes through lateral electrodes without the need for bottom electrodes and confining the acoustic field between the top electrodes. This configuration removes parasitic loss channels, increases energy transfer efficiency to greater than ninety nine percent, and provides a stable free spectral range of about 5.75 MHz with very small fluctuations. Experimental measurements show comb like phonon spectra spanning 0.1 to 1.8 GHz, high quality factors in the range of ten to the power of three to ten to the power of five, frequency quality products larger than ten to the power of thirteen at room temperature, and a low temperature coefficient of frequency. In addition, a gridded electrode design together with the intrinsic properties of 128 degree Y cut LiNbO3, including insensitivity to electrode spacing and a large electromechanical coupling coefficient, suppresses spurious modes and allows tunable mode volumes from 0.008 to 0.064 cubic millimeters. These combined features give X HTBAR devices excellent integration compatibility and strong immunity to electrode related perturbations, making them promising multimode phonon sources for large scale quantum interconnects and microwave photonic integrated circuits.

cond-mat.mes-hall↗
arXiv2025

VisuRiddles: Fine-grained Perception is a Primary Bottleneck for Multimodal Large Language Models in Abstract Visual Reasoning

Recent strides in multimodal large language models (MLLMs) have significantly advanced their performance in many reasoning tasks. However, Abstract Visual Reasoning (AVR) remains a critical challenge, primarily due to limitations in perceiving abstract graphics. To tackle this issue, we investigate the bottlenecks in current MLLMs and synthesize training data to improve their abstract visual perception. First, we propose VisuRiddles, a benchmark for AVR, featuring tasks meticulously constructed to assess models' reasoning capacities across five core dimensions and two high-level reasoning categories. Second, we introduce the Perceptual Riddle Synthesizer (PRS), an automated framework for generating riddles with fine-grained perceptual descriptions. PRS not only generates valuable training data for abstract graphics but also provides fine-grained perceptual description, crucially allowing for supervision over intermediate reasoning stages and thereby improving both training efficacy and model interpretability. Our extensive experimental results on VisuRiddles empirically validate that fine-grained visual perception is the principal bottleneck and our synthesis framework markedly enhances the performance of contemporary MLLMs on these challenging tasks. Our code and dataset will be released at https://github.com/yh-hust/VisuRiddles

cs.CV↗
arXiv2025

GRIDAI: Generating and Repairing Intrusion Detection Rules via Collaboration among Multiple LLM-based Agents

Rule-based network intrusion detection systems play a crucial role in the real-time detection of Web attacks. However, most existing works primarily focus on automatically generating detection rules for new attacks, often overlooking the relationships between new attacks and existing rules, which leads to significant redundancy within the ever-expanding ruleset. To address this issue, we propose GRIDAI, a novel end-to-end framework for the automated Generation and Repair of Intrusion Detection rules through collaboration among multiple LLM-based agents. Unlike traditional methods, GRIDAI first assesses the nature of incoming attack samples. If the sample represents a new attack type, it is used to generate a new rule. Otherwise, the sample is identified as a variant of an attack already covered by an existing rule and used to repair the rule by updating the corresponding signature, thereby enhancing its generalization capability. Additionally, to mitigate syntactic and semantic errors in rules caused by LLM hallucinations, we incorporate a tool-based real-time validation mechanism and a representative attack sample maintained for each rule, enabling fully automated rule generation and repair. Comprehensive experiments were conducted on a public dataset containing seven types of attacks and a private dataset with 43 attack types. The results demonstrate that GRIDAI accurately identifies the relationships between new attack samples and existing rules, efficiently generates and repairs rules to handle new attacks and variants, and effectively mitigates the impact of LLM hallucinations.

cs.CR↗