Selected work

Representative Papers

Visual Anomaly Synthesis for Model Selection in Data Scarcity

Sep 29, 2026

This study addresses the challenge of model validation in industrial defect detection caused by the scarcity of real anomalous samples. To this end, we propose a reference-free defect synthesis framework. Leveraging literature-derived defect taxonomy prompts and a severity-graded anomaly synthesis strategy, the method integrates the FLUX.2 generative model with ROI cropping, color-matched blending, and score-based filtering mechanisms to batch-generate defect images across graded severity levels for model selection and validation. Experimental results demonstrate that the proposed framework reduces model selection regret by approximately 50% on the MVTecAD dataset. Furthermore, it achieves an AUROC of 0.97 in a Pelton turbine case study, effectively evaluating detection sensitivity.

0 citationsRead paper

Optimal Restart Strategies for Parameter-dependent Optimization Algorithms

Jan 17, 2025

This work addresses the challenge of adaptively selecting the unknown optimal regularization parameter λ in parameter-dependent optimization algorithms, where excessively large λ incurs prohibitive computational cost while overly small λ yields low success probability. We propose the first classification framework for restart strategies based on bounded relative loss. Theoretically, we prove that multiplicative growth schemes admit an asymptotically optimal scaling factor independent of the true λ. Through rigorous parameter sensitivity analysis, worst-case modeling, and derivation of tight upper and lower bounds on relative loss, we establish boundedness of the relative loss under this strategy and derive an explicit closed-form optimal scaling factor that minimizes the worst-case relative loss. Crucially, this factor’s asymptotic optimality is agnostic to the unknown λ, thereby significantly enhancing both restart efficiency and robustness.

0 citationsRead paper

Recent publications

Latest Papers

Visual Anomaly Synthesis for Model Selection in Data Scarcity

Sep 29, 2026

This study addresses the challenge of model validation in industrial defect detection caused by the scarcity of real anomalous samples. To this end, we propose a reference-free defect synthesis framework. Leveraging literature-derived defect taxonomy prompts and a severity-graded anomaly synthesis strategy, the method integrates the FLUX.2 generative model with ROI cropping, color-matched blending, and score-based filtering mechanisms to batch-generate defect images across graded severity levels for model selection and validation. Experimental results demonstrate that the proposed framework reduces model selection regret by approximately 50% on the MVTecAD dataset. Furthermore, it achieves an AUROC of 0.97 in a Pelton turbine case study, effectively evaluating detection sensitivity.

0 citationsRead paper

Optimal Restart Strategies for Parameter-dependent Optimization Algorithms

Jan 17, 2025

This work addresses the challenge of adaptively selecting the unknown optimal regularization parameter λ in parameter-dependent optimization algorithms, where excessively large λ incurs prohibitive computational cost while overly small λ yields low success probability. We propose the first classification framework for restart strategies based on bounded relative loss. Theoretically, we prove that multiplicative growth schemes admit an asymptotically optimal scaling factor independent of the true λ. Through rigorous parameter sensitivity analysis, worst-case modeling, and derivation of tight upper and lower bounds on relative loss, we establish boundedness of the relative loss under this strategy and derive an explicit closed-form optimal scaling factor that minimizes the worst-case relative loss. Crucially, this factor’s asymptotic optimality is agnostic to the unknown λ, thereby significantly enhancing both restart efficiency and robustness.

0 citationsRead paper