Selected work

Representative Papers

ZoRRO: A Zero-Weight Personalized Recommender System for Scalable News Recommendation

Jul 12, 2026

This work addresses the high training costs, substantial inference latency, and poor deployment scalability prevalent in large-scale news recommendation systems by proposing a training-free, zero-parameter personalized recommendation framework. The approach achieves personalization through efficient matching between user behavioral representations and news semantics, entirely circumventing neural network training or fine-tuning. Experimental results demonstrate that the method outperforms strong neural baselines in offline evaluations, while online A/B tests show click-through rates nearly on par with state-of-the-art models. Notably, it achieves over a 600× speedup in inference latency, offering compelling evidence of the often-overlooked discrepancy between offline metrics and online performance. To the best of our knowledge, this is the first practical news recommendation system that simultaneously delivers high performance and eliminates the need for model training.

0 citationsRead paper

Reference-Free Image Quality Assessment for Virtual Try-On via Human Feedback

Mar 13, 2026

Existing virtual try-on systems lack a no-reference, single-image quality assessment method that aligns with human perception. To address this gap, this work proposes VTON-IQA, a novel framework that first introduces VTON-QBench—the largest human-annotated benchmark to date for image quality evaluation across 14 state-of-the-art virtual try-on models—and then designs a Transformer-based interleaved cross-attention mechanism to explicitly model the interaction between garment fidelity and preservation of human details. Experiments demonstrate that VTON-IQA achieves highly consistent image-level quality predictions with human judgments under no-reference conditions, establishing the first generalizable and perceptually aligned evaluation standard for virtual try-on models.

0 citationsRead paper

Connectivity-Guided Sparsification of 2-FWL GNNs: Preserving Full Expressivity with Improved Efficiency

Nov 16, 2025

Higher-order graph neural networks (HOGNNs) achieve 2-FWL expressive power but incur O(n³) computational complexity due to three-node interactions; existing acceleration techniques often compromise expressivity. Method: We propose Co-Sparsify, the first framework to identify that three-node interactions are expressively necessary only within biconnected components (BCCs). Leveraging this insight, we design a structure-aware sparsification strategy that retains higher-order interactions exclusively inside BCCs—avoiding approximation or sampling. The method integrates 2-FWL message passing, BCC decomposition, connectivity-aware partitioning, and global readout. Contribution/Results: Co-Sparsify provably preserves 2-FWL equivalence while drastically reducing memory and computation. Experiments on ZINC, QM9, and other benchmarks demonstrate superior predictive performance over state-of-the-art HOGNNs, alongside significant efficiency gains.

0 citationsRead paper

Tactile Data Recording System for Clothing with Motion-Controlled Robotic Sliding

Nov 08, 2025

Understanding the physical determinants of tactile comfort in garments remains challenging due to the lack of controlled, high-fidelity tactile data capturing dynamic finger–fabric interactions. Method: We propose a robotic-arm-based tactile acquisition system that emulates fingertip sliding with precise control over velocity, direction, and normal force, while synchronously recording multimodal signals—tactile force, acceleration, and audio. Crucially, it enables non-destructive, full-garment tactile sensing with fine-grained motion annotations (e.g., speed, trajectory, contact state). Contribution/Results: We introduce the first motion-parameter-annotated garment tactile database. Experiments demonstrate that incorporating motion information significantly improves material classification accuracy by +12.7%, validating its value for building more robust and scalable models of tactile perception. This work establishes a physically grounded, experimentally validated framework for linking fabric mechanics to human tactile comfort.

0 citationsRead paper

Pinching Visuo-haptic Display: Investigating Cross-Modal Effects of Visual Textures on Electrostatic Cloth Tactile Sensations

Oct 12, 2025International Conference on Multimodal Interaction

This study investigates how visual texture modulates cross-modal tactile perception in electrostatic fabric haptic displays. We employed a conductive-fabric-based electrostatic adhesion haptic device synchronized with a spatially registered virtual reality fabric rendering system to deliver congruent or incongruent visual textures alongside tactile roughness cues. Results demonstrate that visual texture significantly modulates subjective perceptions of stiffness (voile), warmth (toweling), and roughness (denim), confirming that visual augmentation effectively extends the material expressiveness of single-dimension haptic devices—those capable only of modulating surface friction. In contrast, perceived thickness remained unaffected, as pinch-based interactions provided dominant haptic cues. This work presents the first systematic validation of cross-modal visual–tactile integration for dimensional expansion in fabric interaction, establishing both theoretical foundations and practical implementation strategies for multimodal textile interfaces.

0 citationsRead paper

Recent publications

Latest Papers

ZoRRO: A Zero-Weight Personalized Recommender System for Scalable News Recommendation

Jul 12, 2026

This work addresses the high training costs, substantial inference latency, and poor deployment scalability prevalent in large-scale news recommendation systems by proposing a training-free, zero-parameter personalized recommendation framework. The approach achieves personalization through efficient matching between user behavioral representations and news semantics, entirely circumventing neural network training or fine-tuning. Experimental results demonstrate that the method outperforms strong neural baselines in offline evaluations, while online A/B tests show click-through rates nearly on par with state-of-the-art models. Notably, it achieves over a 600× speedup in inference latency, offering compelling evidence of the often-overlooked discrepancy between offline metrics and online performance. To the best of our knowledge, this is the first practical news recommendation system that simultaneously delivers high performance and eliminates the need for model training.

0 citationsRead paper

Reference-Free Image Quality Assessment for Virtual Try-On via Human Feedback

Mar 13, 2026

Existing virtual try-on systems lack a no-reference, single-image quality assessment method that aligns with human perception. To address this gap, this work proposes VTON-IQA, a novel framework that first introduces VTON-QBench—the largest human-annotated benchmark to date for image quality evaluation across 14 state-of-the-art virtual try-on models—and then designs a Transformer-based interleaved cross-attention mechanism to explicitly model the interaction between garment fidelity and preservation of human details. Experiments demonstrate that VTON-IQA achieves highly consistent image-level quality predictions with human judgments under no-reference conditions, establishing the first generalizable and perceptually aligned evaluation standard for virtual try-on models.

0 citationsRead paper

Connectivity-Guided Sparsification of 2-FWL GNNs: Preserving Full Expressivity with Improved Efficiency

Nov 16, 2025

Higher-order graph neural networks (HOGNNs) achieve 2-FWL expressive power but incur O(n³) computational complexity due to three-node interactions; existing acceleration techniques often compromise expressivity. Method: We propose Co-Sparsify, the first framework to identify that three-node interactions are expressively necessary only within biconnected components (BCCs). Leveraging this insight, we design a structure-aware sparsification strategy that retains higher-order interactions exclusively inside BCCs—avoiding approximation or sampling. The method integrates 2-FWL message passing, BCC decomposition, connectivity-aware partitioning, and global readout. Contribution/Results: Co-Sparsify provably preserves 2-FWL equivalence while drastically reducing memory and computation. Experiments on ZINC, QM9, and other benchmarks demonstrate superior predictive performance over state-of-the-art HOGNNs, alongside significant efficiency gains.

0 citationsRead paper

Tactile Data Recording System for Clothing with Motion-Controlled Robotic Sliding

Nov 08, 2025

Understanding the physical determinants of tactile comfort in garments remains challenging due to the lack of controlled, high-fidelity tactile data capturing dynamic finger–fabric interactions. Method: We propose a robotic-arm-based tactile acquisition system that emulates fingertip sliding with precise control over velocity, direction, and normal force, while synchronously recording multimodal signals—tactile force, acceleration, and audio. Crucially, it enables non-destructive, full-garment tactile sensing with fine-grained motion annotations (e.g., speed, trajectory, contact state). Contribution/Results: We introduce the first motion-parameter-annotated garment tactile database. Experiments demonstrate that incorporating motion information significantly improves material classification accuracy by +12.7%, validating its value for building more robust and scalable models of tactile perception. This work establishes a physically grounded, experimentally validated framework for linking fabric mechanics to human tactile comfort.

0 citationsRead paper

Pinching Visuo-haptic Display: Investigating Cross-Modal Effects of Visual Textures on Electrostatic Cloth Tactile Sensations

Oct 12, 2025International Conference on Multimodal Interaction

This study investigates how visual texture modulates cross-modal tactile perception in electrostatic fabric haptic displays. We employed a conductive-fabric-based electrostatic adhesion haptic device synchronized with a spatially registered virtual reality fabric rendering system to deliver congruent or incongruent visual textures alongside tactile roughness cues. Results demonstrate that visual texture significantly modulates subjective perceptions of stiffness (voile), warmth (toweling), and roughness (denim), confirming that visual augmentation effectively extends the material expressiveness of single-dimension haptic devices—those capable only of modulating surface friction. In contrast, perceived thickness remained unaffected, as pinch-based interactions provided dominant haptic cues. This work presents the first systematic validation of cross-modal visual–tactile integration for dimensional expansion in fabric interaction, establishing both theoretical foundations and practical implementation strategies for multimodal textile interfaces.

0 citationsRead paper