Selected work

Representative Papers

Beyond Visual Evidence: Revealing and Mitigating Relational Privacy Leakage in Document MLLMs

Aug 13, 2026

This work addresses a critical privacy vulnerability in multimodal large language models for document understanding: when visual evidence is insufficient, these models often rely on memorized field correlations from training data to infer missing content, leading to relational leakage of sensitive personal information. To mitigate this risk, the paper introduces the first systematic analysis of such privacy leakage mechanisms and proposes a Dynamic Relation Unlearning Framework (DRUF), which integrates a relation decoupling unlearning module with a dynamic set updating mechanism to suppress high-risk field associations while preserving essential information extraction performance. Additionally, the authors construct DocPrivacyBench, the first privacy evaluation benchmark tailored to scenarios with missing visual evidence. Experiments demonstrate that DRUF significantly reduces privacy leakage risk—outperforming the strongest baseline by 4.8 percentage points—without compromising robustness in information extraction.

0 citationsRead paper

Beyond the Capability Boundary: Zeroth-Order Optimization for Self-Evolving LLM Agents

Aug 10, 2026

This work proposes the first zeroth-order optimization framework for self-evolution of large language model (LLM) agents, addressing the limitation that existing methods struggle to generate correct reasoning trajectories on challenging instances due to inherent capability boundaries. The approach perturbs LoRA parameters in parallel, estimates gradient signals via answer perplexity-based loss, and incorporates an adaptive search mechanism to update the model, thereby producing high-quality trajectories for supervised fine-tuning within a closed-loop evolutionary process. Requiring no human annotation, the method substantially increases the number of successful trajectories across multiple deep reasoning benchmarks and consistently outperforms strong baselines—particularly on difficult samples—effectively expanding the capability frontier of LLM agents.

0 citationsRead paper

SpecF2M: A Spectral-Aware Multi-task Network Estimating Axial Length and Refractive Error from Pediatric Fundus Photographs

Aug 07, 2026

This study addresses the challenges in pediatric myopia screening, which relies on invasive examinations that hinder widespread adoption. Fundus images exhibit low-contrast, spatially diffuse, and multiscale features associated with axial length (AL), spherical equivalent (SPH), and cylindrical power (CYL), whose anatomical correlates partially overlap yet remain inconsistent. To tackle this, the authors propose SpecF2M—the first spectrum-aware multitask network incorporating hybrid spatial–frequency domain modeling for pediatric fundus analysis—featuring an anatomy-guided enhancement module, a MixCNN-HSL hybrid backbone, and an expert-routing head to jointly estimate AL, SPH, and CYL. Evaluated on 6,966 pediatric fundus images, the method achieves mean absolute errors of 0.5347 mm for AL and 0.7062 D for SPH, significantly outperforming CNN- and ViT-based baselines, while revealing an asymmetric task coupling wherein CYL exhibits weaker association with myopic fundus patterns.

0 citationsRead paper

Interactive Reward Agent: GUI Task Evaluation via Environment-State Verification

Jul 28, 2026

Automated evaluation of GUI task completion remains challenging due to its reliance on environmental states beyond interface screenshots, such as system configurations and file contents. This work proposes the Interactive Reward Agent (IRA), which introduces environment state verification into GUI evaluation for the first time through a "propose-and-verify" framework. IRA dynamically gathers multi-source evidence by invoking system commands, application APIs, and GUI tools, enabling holistic assessment by fusing visual interface data with underlying environmental information. The authors release GUI-RewardBench, a new benchmark comprising 321 annotated trajectories, on which IRA achieves an accuracy of 86.9%, substantially outperforming existing methods. When employed to train GUI agents via reinforcement learning, IRA attains a 34.0% success rate on OSWorld.

0 citationsRead paper

OLEDLM: A Unified Language Model for OLED Molecular Design

Jul 22, 2026

This work addresses the challenges in OLED material design—namely, the vast chemical space, stringent quantum-chemical constraints, and scarcity of labeled data—by proposing the first inverse molecular generation framework tailored specifically for OLEDs. The approach employs a LLaMA-style Transformer-based causal language model, integrated with a BERT-pretrained property predictor and a multi-stage training strategy, further enhanced by reinforcement learning to optimize SMILES generation. This enables direct generation of molecules that satisfy rigorous structural constraints while targeting desired optoelectronic properties. Density functional theory (DFT) validation demonstrates that the framework efficiently explores the OLED-relevant chemical space, yielding structurally valid and high-performing novel candidates, thereby significantly narrowing the gap between general-purpose molecular generation and the specialized demands of optoelectronic material design.

0 citationsRead paper

Recent publications

Latest Papers

Beyond Visual Evidence: Revealing and Mitigating Relational Privacy Leakage in Document MLLMs

Aug 13, 2026

This work addresses a critical privacy vulnerability in multimodal large language models for document understanding: when visual evidence is insufficient, these models often rely on memorized field correlations from training data to infer missing content, leading to relational leakage of sensitive personal information. To mitigate this risk, the paper introduces the first systematic analysis of such privacy leakage mechanisms and proposes a Dynamic Relation Unlearning Framework (DRUF), which integrates a relation decoupling unlearning module with a dynamic set updating mechanism to suppress high-risk field associations while preserving essential information extraction performance. Additionally, the authors construct DocPrivacyBench, the first privacy evaluation benchmark tailored to scenarios with missing visual evidence. Experiments demonstrate that DRUF significantly reduces privacy leakage risk—outperforming the strongest baseline by 4.8 percentage points—without compromising robustness in information extraction.

0 citationsRead paper

Beyond the Capability Boundary: Zeroth-Order Optimization for Self-Evolving LLM Agents

Aug 10, 2026

This work proposes the first zeroth-order optimization framework for self-evolution of large language model (LLM) agents, addressing the limitation that existing methods struggle to generate correct reasoning trajectories on challenging instances due to inherent capability boundaries. The approach perturbs LoRA parameters in parallel, estimates gradient signals via answer perplexity-based loss, and incorporates an adaptive search mechanism to update the model, thereby producing high-quality trajectories for supervised fine-tuning within a closed-loop evolutionary process. Requiring no human annotation, the method substantially increases the number of successful trajectories across multiple deep reasoning benchmarks and consistently outperforms strong baselines—particularly on difficult samples—effectively expanding the capability frontier of LLM agents.

0 citationsRead paper

SpecF2M: A Spectral-Aware Multi-task Network Estimating Axial Length and Refractive Error from Pediatric Fundus Photographs

Aug 07, 2026

This study addresses the challenges in pediatric myopia screening, which relies on invasive examinations that hinder widespread adoption. Fundus images exhibit low-contrast, spatially diffuse, and multiscale features associated with axial length (AL), spherical equivalent (SPH), and cylindrical power (CYL), whose anatomical correlates partially overlap yet remain inconsistent. To tackle this, the authors propose SpecF2M—the first spectrum-aware multitask network incorporating hybrid spatial–frequency domain modeling for pediatric fundus analysis—featuring an anatomy-guided enhancement module, a MixCNN-HSL hybrid backbone, and an expert-routing head to jointly estimate AL, SPH, and CYL. Evaluated on 6,966 pediatric fundus images, the method achieves mean absolute errors of 0.5347 mm for AL and 0.7062 D for SPH, significantly outperforming CNN- and ViT-based baselines, while revealing an asymmetric task coupling wherein CYL exhibits weaker association with myopic fundus patterns.

0 citationsRead paper

Interactive Reward Agent: GUI Task Evaluation via Environment-State Verification

Jul 28, 2026

Automated evaluation of GUI task completion remains challenging due to its reliance on environmental states beyond interface screenshots, such as system configurations and file contents. This work proposes the Interactive Reward Agent (IRA), which introduces environment state verification into GUI evaluation for the first time through a "propose-and-verify" framework. IRA dynamically gathers multi-source evidence by invoking system commands, application APIs, and GUI tools, enabling holistic assessment by fusing visual interface data with underlying environmental information. The authors release GUI-RewardBench, a new benchmark comprising 321 annotated trajectories, on which IRA achieves an accuracy of 86.9%, substantially outperforming existing methods. When employed to train GUI agents via reinforcement learning, IRA attains a 34.0% success rate on OSWorld.

0 citationsRead paper

OLEDLM: A Unified Language Model for OLED Molecular Design

Jul 22, 2026

This work addresses the challenges in OLED material design—namely, the vast chemical space, stringent quantum-chemical constraints, and scarcity of labeled data—by proposing the first inverse molecular generation framework tailored specifically for OLEDs. The approach employs a LLaMA-style Transformer-based causal language model, integrated with a BERT-pretrained property predictor and a multi-stage training strategy, further enhanced by reinforcement learning to optimize SMILES generation. This enables direct generation of molecules that satisfy rigorous structural constraints while targeting desired optoelectronic properties. Density functional theory (DFT) validation demonstrates that the framework efficiently explores the OLED-relevant chemical space, yielding structurally valid and high-performing novel candidates, thereby significantly narrowing the gap between general-purpose molecular generation and the specialized demands of optoelectronic material design.

0 citationsRead paper