Institution profile

University of Science and Technology

Academic institutionasia · kr
Official website
Research library12linked papers
Opportunities0open roles
Selected work

Representative Papers

Through Van Gogh's Eyes: Global Style Transfer with Diffusion Mod

Aug 11, 2026

This work addresses the limitations of existing artistic style transfer methods, which are often constrained by reliance on a single reference image or biased text prompts, hindering accurate modeling of an artist’s holistic style distribution. To overcome this, the paper introduces Global Style Transfer (GST), a novel Many-to-One paradigm that aggregates multiple artworks in the intermediate feature space of a diffusion model to learn a shared, artist-level style representation. Leveraging a training-free Global Style Guidance (GSG) mechanism alongside Content Alignment Guidance (CAG), the approach enables text-free style transfer that preserves semantic structure while allowing controllable deformations. Experiments on the WikiArt dataset demonstrate that the proposed method significantly outperforms current state-of-the-art techniques, achieving superior performance across three key metrics: style fidelity, content preservation, and output diversity.

0 citationsRead paper

Generalization bounds and sample complexity for remaining useful life prediction from complete degradation trajectories

Jul 26, 2026

This study addresses the challenge of remaining useful life (RUL) prediction, where complete degradation data are scarce and costly to obtain, and theoretical guidance on sample requirements is lacking. The authors establish a sample complexity framework for RUL prediction, providing the first distribution-free upper bound on mean squared error generalization and a matching minimax lower bound. They quantify how physical priors reduce data requirements and uncover performance degradation mechanisms caused by model misspecification and right-censored observations. Leveraging statistical learning theory, minimax analysis, and Bernstein-type inequalities—combined with exponential, power-law, and stretched-exponential degradation models—the theoretical results are validated on benchmark datasets for turbofan engines, batteries, and bearings, achieving average errors within a factor of 2–3. These findings yield actionable guidelines for data acquisition, model complexity selection, and physics-informed model evaluation.

0 citationsRead paper
Recent publications

Latest Papers

Through Van Gogh's Eyes: Global Style Transfer with Diffusion Mod

Aug 11, 2026

This work addresses the limitations of existing artistic style transfer methods, which are often constrained by reliance on a single reference image or biased text prompts, hindering accurate modeling of an artist’s holistic style distribution. To overcome this, the paper introduces Global Style Transfer (GST), a novel Many-to-One paradigm that aggregates multiple artworks in the intermediate feature space of a diffusion model to learn a shared, artist-level style representation. Leveraging a training-free Global Style Guidance (GSG) mechanism alongside Content Alignment Guidance (CAG), the approach enables text-free style transfer that preserves semantic structure while allowing controllable deformations. Experiments on the WikiArt dataset demonstrate that the proposed method significantly outperforms current state-of-the-art techniques, achieving superior performance across three key metrics: style fidelity, content preservation, and output diversity.

0 citationsRead paper

Generalization bounds and sample complexity for remaining useful life prediction from complete degradation trajectories

Jul 26, 2026

This study addresses the challenge of remaining useful life (RUL) prediction, where complete degradation data are scarce and costly to obtain, and theoretical guidance on sample requirements is lacking. The authors establish a sample complexity framework for RUL prediction, providing the first distribution-free upper bound on mean squared error generalization and a matching minimax lower bound. They quantify how physical priors reduce data requirements and uncover performance degradation mechanisms caused by model misspecification and right-censored observations. Leveraging statistical learning theory, minimax analysis, and Bernstein-type inequalities—combined with exponential, power-law, and stretched-exponential degradation models—the theoretical results are validated on benchmark datasets for turbofan engines, batteries, and bearings, achieving average errors within a factor of 2–3. These findings yield actionable guidelines for data acquisition, model complexity selection, and physics-informed model evaluation.

0 citationsRead paper