Selected work

Representative Papers

Remember Me, Not Save Me: A Collective Memory System for Evolving Virtual Identities in Augmented Reality

Dec 13, 2025International Conference on Virtual Reality Continuum and its Applications in Industry

This study investigates how to construct virtual identities in augmented reality (AR) that can evolve stable personalities through sustained public interaction. To this end, the authors propose a dynamic collective memory model to integrate and reconcile contradictory interaction memories, design a state-reflective virtual avatar to enable personality evolution, and introduce a geocultural contextual anchoring mechanism to strengthen local identity formation. By integrating AR, artificial intelligence, and narrative tension mechanisms, the system was deployed at the 2024 Jinan Biennale. Analysis of over 2,500 public interactions demonstrated the successful emergence of a stable ISTP personality profile, thereby validating the feasibility and novelty of transforming collective memory into a coherent virtual identity.

1 citationsRead paper

Admission Without Answers: Label-Free Certification and Experience Learning for LLM-Based Optimization Modeling

Aug 16, 2026

This study addresses the unreliability of knowledge admission in LLM experiential learning from unlabeled streams due to the absence of validation. We propose AdmitOR, a novel admission mechanism based on explicitly calibrated false discovery objectives that achieves high-reliability knowledge filtering through cross-model-family behavioral evidence, parameter-space resampling, and calibrated threshold decision-making. Experiments demonstrate that AdmitOR attains an admission precision of 0.927, reduces poisoning rates eightfold, and achieves a macro-accuracy of 58.4%, significantly outperforming majority voting and execution-success baselines. These results indicate substantial improvements in both skill library precision and generalization performance. Furthermore, this work reveals the impact of benchmark text distortion on transferability, highlighting critical considerations for deploying experiential learning systems in real-world unlabeled environments where validation signals are inherently unavailable.

0 citationsRead paper

DocPure: Prompt-Free Unified Document Restoration via Degradation-Aware Structure-Guided Wavelet Modulation

Aug 10, 2026

This work addresses the degradation of document image quality caused by blur, noise, compression artifacts, and shadows during acquisition and transmission by proposing the first unified restoration framework that operates without task-specific prompts. The method employs a degradation-aware structural autoencoder to predict a clean structural prior and introduces a structure-guided wavelet interaction mechanism that bridges frequency-domain representations and spatial semantics for high-fidelity recovery. Key innovations include degradation-aware routing regularization and a cross-band adaptive modulation strategy, which effectively integrate low-frequency structural cues to guide high-frequency detail reconstruction. Extensive experiments demonstrate that the proposed approach consistently outperforms state-of-the-art methods across diverse degradation scenarios, exhibiting superior generalization capability and restoration performance.

0 citationsRead paper

BiomedAP: A Vision-Informed Dual-Anchor Framework with Gated Cross-Modal Fusion for Robust Medical Vision-Language Adaptation

May 15, 2026

This work addresses the performance instability of medical vision–language models under noisy and heterogeneous clinical text descriptions caused by prompt variations. To this end, the authors propose a dual-anchor prompt learning framework that dynamically filters irrelevant textual information through a gated cross-modal fusion mechanism and jointly aligns image–text prompts. The high anchor enforces semantic consistency via expert-designed templates, while the low anchor enhances representational stability using few-shot visual prototypes. Integrated with parameter-efficient fine-tuning and gated cross-modal attention, the proposed model significantly outperforms existing methods across 11 medical benchmarks, demonstrating superior accuracy and robustness under both few-shot learning and prompt perturbation settings.

0 citationsRead paper

RevealLayer: Disentangling Hidden and Visible Layers via Occlusion-Aware Image Decomposition

May 12, 2026

This work addresses the challenges of layer decomposition in natural images—namely, difficulty in occlusion inpainting, non-robust disentanglement, and ambiguous boundaries—exacerbated by the absence of high-quality multi-layer datasets. The authors propose a diffusion-based layer decomposition framework that leverages region-aware attention, an occlusion-guided adapter, and a composite loss function to achieve precise RGBA layer separation and faithful reconstruction of occluded content. A key innovation is the introduction of an occlusion-aware mechanism that implicitly disentangles visible and hidden layers. To support this research, they also construct RevealLayer-100K, the first human-in-the-loop annotated multi-layer dataset, along with RevealLayerBench, a comprehensive benchmark for evaluation. Experiments demonstrate that the proposed method significantly outperforms existing approaches in layer accuracy, alpha matte sharpness, and reliability of occlusion recovery.

0 citationsRead paper

Recent publications

Latest Papers

Admission Without Answers: Label-Free Certification and Experience Learning for LLM-Based Optimization Modeling

Aug 16, 2026

This study addresses the unreliability of knowledge admission in LLM experiential learning from unlabeled streams due to the absence of validation. We propose AdmitOR, a novel admission mechanism based on explicitly calibrated false discovery objectives that achieves high-reliability knowledge filtering through cross-model-family behavioral evidence, parameter-space resampling, and calibrated threshold decision-making. Experiments demonstrate that AdmitOR attains an admission precision of 0.927, reduces poisoning rates eightfold, and achieves a macro-accuracy of 58.4%, significantly outperforming majority voting and execution-success baselines. These results indicate substantial improvements in both skill library precision and generalization performance. Furthermore, this work reveals the impact of benchmark text distortion on transferability, highlighting critical considerations for deploying experiential learning systems in real-world unlabeled environments where validation signals are inherently unavailable.

0 citationsRead paper

DocPure: Prompt-Free Unified Document Restoration via Degradation-Aware Structure-Guided Wavelet Modulation

Aug 10, 2026

This work addresses the degradation of document image quality caused by blur, noise, compression artifacts, and shadows during acquisition and transmission by proposing the first unified restoration framework that operates without task-specific prompts. The method employs a degradation-aware structural autoencoder to predict a clean structural prior and introduces a structure-guided wavelet interaction mechanism that bridges frequency-domain representations and spatial semantics for high-fidelity recovery. Key innovations include degradation-aware routing regularization and a cross-band adaptive modulation strategy, which effectively integrate low-frequency structural cues to guide high-frequency detail reconstruction. Extensive experiments demonstrate that the proposed approach consistently outperforms state-of-the-art methods across diverse degradation scenarios, exhibiting superior generalization capability and restoration performance.

0 citationsRead paper

BiomedAP: A Vision-Informed Dual-Anchor Framework with Gated Cross-Modal Fusion for Robust Medical Vision-Language Adaptation

May 15, 2026

This work addresses the performance instability of medical vision–language models under noisy and heterogeneous clinical text descriptions caused by prompt variations. To this end, the authors propose a dual-anchor prompt learning framework that dynamically filters irrelevant textual information through a gated cross-modal fusion mechanism and jointly aligns image–text prompts. The high anchor enforces semantic consistency via expert-designed templates, while the low anchor enhances representational stability using few-shot visual prototypes. Integrated with parameter-efficient fine-tuning and gated cross-modal attention, the proposed model significantly outperforms existing methods across 11 medical benchmarks, demonstrating superior accuracy and robustness under both few-shot learning and prompt perturbation settings.

0 citationsRead paper

RevealLayer: Disentangling Hidden and Visible Layers via Occlusion-Aware Image Decomposition

May 12, 2026

This work addresses the challenges of layer decomposition in natural images—namely, difficulty in occlusion inpainting, non-robust disentanglement, and ambiguous boundaries—exacerbated by the absence of high-quality multi-layer datasets. The authors propose a diffusion-based layer decomposition framework that leverages region-aware attention, an occlusion-guided adapter, and a composite loss function to achieve precise RGBA layer separation and faithful reconstruction of occluded content. A key innovation is the introduction of an occlusion-aware mechanism that implicitly disentangles visible and hidden layers. To support this research, they also construct RevealLayer-100K, the first human-in-the-loop annotated multi-layer dataset, along with RevealLayerBench, a comprehensive benchmark for evaluation. Experiments demonstrate that the proposed method significantly outperforms existing approaches in layer accuracy, alpha matte sharpness, and reliability of occlusion recovery.

0 citationsRead paper

Remember Me, Not Save Me: A Collective Memory System for Evolving Virtual Identities in Augmented Reality

Dec 13, 2025International Conference on Virtual Reality Continuum and its Applications in Industry

This study investigates how to construct virtual identities in augmented reality (AR) that can evolve stable personalities through sustained public interaction. To this end, the authors propose a dynamic collective memory model to integrate and reconcile contradictory interaction memories, design a state-reflective virtual avatar to enable personality evolution, and introduce a geocultural contextual anchoring mechanism to strengthen local identity formation. By integrating AR, artificial intelligence, and narrative tension mechanisms, the system was deployed at the 2024 Jinan Biennale. Analysis of over 2,500 public interactions demonstrated the successful emergence of a stable ISTP personality profile, thereby validating the feasibility and novelty of transforming collective memory into a coherent virtual identity.

1 citationsRead paper