Institution profile

National Dong Hwa University

Academic institutionasia · tw
Official website
Research library6linked papers
Opportunities0open roles
Selected work

Representative Papers

A Width-Matched Comparison of Hybrid Quantum-Classical Self-Supervised Learning for Fingerprint Recognition

Sep 30, 2026

This study addresses the reliance of fingerprint recognition on extensive labeled data and the unclear attribution of benefits in quantum self-supervised learning. For the first time, a quantum feature extraction module (QuFeX) is embedded into three self-supervised frameworks—SimCLR, MoCo v2, and BYOL—using an equal-width controlled variable approach to construct hybrid models, which are systematically evaluated via KNN classification. Results indicate that these hybrid models significantly outperform classical baselines exclusively within contrastive learning settings, while exhibiting no reliable improvement in non-contrastive scenarios. This work reveals that quantum-enhanced gains are inherently dependent on the specific self-supervised objective employed, providing critical empirical evidence for understanding the true sources of advantage in quantum machine learning.

0 citationsRead paper

When Does Geometric View Synthesis Help Wine Label Retrieval? A Public One-Shot Benchmark Across Self-Supervised and Vision-Language Backbones

Sep 27, 2026

This study investigates the value of geometric view synthesis under pretrained encoders for one-shot wine label retrieval. Methodologically, we construct a thousand-class one-shot retrieval benchmark and integrate LoRA fine-tuning with SAM-based localization to systematically compare geometric augmentation effects across DINO and SigLIP architectures. Experimental results demonstrate that fine-tuned DINO achieves up to a threefold accuracy improvement, whereas frozen SigLIP features yield only marginal gains. By quantifying the differential benefits of geometric synthesis across distinct architectures, this work delineates the effective applicability boundaries of the proposed approach, providing empirical evidence for view augmentation strategies in one-shot retrieval scenarios.

0 citationsRead paper

Spatial Adapter: Structured Spatial Decomposition and Closed-Form Covariance for Frozen Predictors

May 11, 2026

This work addresses the limitations of frozen backbone networks in residual field modeling—specifically, their inability to adequately capture spatial structure and quantify uncertainty—by introducing Spatial Adapter, a parameter-efficient post-processing layer. The method constructs an identifiable low-rank representation through a structured spatial decomposition that enforces smoothness, sparsity, and orthogonality, enabling closed-form derivation of spatial covariance. An effective rank is determined via a data-adaptive spectral thresholding scheme, facilitating kriging interpolation and uncertainty quantification. By jointly optimizing spatial orthogonal bases and sample-specific scores using mini-batch ADMM, and integrating low-rank-plus-noise covariance estimation with a compact trend network, the approach successfully recovers diverse residual spatial structures across various tasks—from linear to deep spatiotemporal or vision backbones—while maintaining a parameter budget below K(N+T).

0 citationsRead paper

DeepKriging on the global Data

Apr 02, 2026

This study addresses the limitations of traditional spatial prediction methods on spherical domains, which often suffer from distance distortion due to reliance on Euclidean metrics or planar projections and struggle to scale to large datasets. To overcome these challenges, the authors propose Spherical DeepKriging—a novel framework that integrates intrinsic spherical thin-plate spline basis functions with deep learning to flexibly capture complex spatial structures on the sphere. This approach represents the first fusion of native spherical basis functions and neural networks for scalable geostatistical modeling. It effectively circumvents the constraints of classical kriging under global-scale and big-data scenarios, demonstrating superior predictive accuracy and computational scalability over existing methods in both synthetic and real-world global datasets.

0 citationsRead paper
Recent publications

Latest Papers

A Width-Matched Comparison of Hybrid Quantum-Classical Self-Supervised Learning for Fingerprint Recognition

Sep 30, 2026

This study addresses the reliance of fingerprint recognition on extensive labeled data and the unclear attribution of benefits in quantum self-supervised learning. For the first time, a quantum feature extraction module (QuFeX) is embedded into three self-supervised frameworks—SimCLR, MoCo v2, and BYOL—using an equal-width controlled variable approach to construct hybrid models, which are systematically evaluated via KNN classification. Results indicate that these hybrid models significantly outperform classical baselines exclusively within contrastive learning settings, while exhibiting no reliable improvement in non-contrastive scenarios. This work reveals that quantum-enhanced gains are inherently dependent on the specific self-supervised objective employed, providing critical empirical evidence for understanding the true sources of advantage in quantum machine learning.

0 citationsRead paper

When Does Geometric View Synthesis Help Wine Label Retrieval? A Public One-Shot Benchmark Across Self-Supervised and Vision-Language Backbones

Sep 27, 2026

This study investigates the value of geometric view synthesis under pretrained encoders for one-shot wine label retrieval. Methodologically, we construct a thousand-class one-shot retrieval benchmark and integrate LoRA fine-tuning with SAM-based localization to systematically compare geometric augmentation effects across DINO and SigLIP architectures. Experimental results demonstrate that fine-tuned DINO achieves up to a threefold accuracy improvement, whereas frozen SigLIP features yield only marginal gains. By quantifying the differential benefits of geometric synthesis across distinct architectures, this work delineates the effective applicability boundaries of the proposed approach, providing empirical evidence for view augmentation strategies in one-shot retrieval scenarios.

0 citationsRead paper

Spatial Adapter: Structured Spatial Decomposition and Closed-Form Covariance for Frozen Predictors

May 11, 2026

This work addresses the limitations of frozen backbone networks in residual field modeling—specifically, their inability to adequately capture spatial structure and quantify uncertainty—by introducing Spatial Adapter, a parameter-efficient post-processing layer. The method constructs an identifiable low-rank representation through a structured spatial decomposition that enforces smoothness, sparsity, and orthogonality, enabling closed-form derivation of spatial covariance. An effective rank is determined via a data-adaptive spectral thresholding scheme, facilitating kriging interpolation and uncertainty quantification. By jointly optimizing spatial orthogonal bases and sample-specific scores using mini-batch ADMM, and integrating low-rank-plus-noise covariance estimation with a compact trend network, the approach successfully recovers diverse residual spatial structures across various tasks—from linear to deep spatiotemporal or vision backbones—while maintaining a parameter budget below K(N+T).

0 citationsRead paper

DeepKriging on the global Data

Apr 02, 2026

This study addresses the limitations of traditional spatial prediction methods on spherical domains, which often suffer from distance distortion due to reliance on Euclidean metrics or planar projections and struggle to scale to large datasets. To overcome these challenges, the authors propose Spherical DeepKriging—a novel framework that integrates intrinsic spherical thin-plate spline basis functions with deep learning to flexibly capture complex spatial structures on the sphere. This approach represents the first fusion of native spherical basis functions and neural networks for scalable geostatistical modeling. It effectively circumvents the constraints of classical kriging under global-scale and big-data scenarios, demonstrating superior predictive accuracy and computational scalability over existing methods in both synthetic and real-world global datasets.

0 citationsRead paper