Label-Free Foundational Model Selection for Medical Image Classification under Distribution Shift via Pseudo Label Discrepancy
研究解决了在无标签目标域中选择基础模型的问题，通过提出一种基于SUDO框架的无标签选择标准AURCC来评估模型性能。
0 citationsRead paper
Selected work
Representative Papers
Value-Aware Product Recommendation by Customer Segmentation using a suitable High-Dimensional Similarity Measure
This study addresses the challenge posed by the high-dimensional sparsity of user-item interaction data and its adverse impact on recommendation profitability. To this end, the authors propose a value-aware recommendation approach that explicitly encodes item profitability within the user-item matrix and introduces a profitability-aware similarity metric tailored for high-dimensional sparse settings. This enables user segmentation based on the profitability of their purchase baskets. Building upon this segmentation, three profit-oriented recommendation strategies—profit share, item popularity, and expected profit—are developed. Experimental evaluations on both synthetic data and the UCI Online Retail real-world dataset demonstrate that the proposed method significantly enhances the overall profitability of recommender systems.
0 citationsRead paper
LRC codes over characteristic $2$
This work proposes a general construction method for linear locally repairable codes (LRCs) over finite fields of characteristic two, fully resolving—for the first time—the high-parameter design problem for LRCs in even characteristic. Leveraging the algebraic structure of finite fields, the method yields LRCs whose length, dimension, and minimum distance are all on the order of $q^4$, with locality $r = q - 1$. The efficacy of the proposed construction is explicitly verified for the cases $q = 4$ and $q = 8$, achieving the best-known parameter trade-offs for LRCs over even-characteristic finite fields to date.
0 citationsRead paper
Recent publications
Latest Papers
Label-Free Foundational Model Selection for Medical Image Classification under Distribution Shift via Pseudo Label Discrepancy
研究解决了在无标签目标域中选择基础模型的问题，通过提出一种基于SUDO框架的无标签选择标准AURCC来评估模型性能。
0 citationsRead paper
Value-Aware Product Recommendation by Customer Segmentation using a suitable High-Dimensional Similarity Measure
This study addresses the challenge posed by the high-dimensional sparsity of user-item interaction data and its adverse impact on recommendation profitability. To this end, the authors propose a value-aware recommendation approach that explicitly encodes item profitability within the user-item matrix and introduces a profitability-aware similarity metric tailored for high-dimensional sparse settings. This enables user segmentation based on the profitability of their purchase baskets. Building upon this segmentation, three profit-oriented recommendation strategies—profit share, item popularity, and expected profit—are developed. Experimental evaluations on both synthetic data and the UCI Online Retail real-world dataset demonstrate that the proposed method significantly enhances the overall profitability of recommender systems.
0 citationsRead paper
LRC codes over characteristic $2$
This work proposes a general construction method for linear locally repairable codes (LRCs) over finite fields of characteristic two, fully resolving—for the first time—the high-parameter design problem for LRCs in even characteristic. Leveraging the algebraic structure of finite fields, the method yields LRCs whose length, dimension, and minimum distance are all on the order of $q^4$, with locality $r = q - 1$. The efficacy of the proposed construction is explicitly verified for the cases $q = 4$ and $q = 8$, achieving the best-known parameter trade-offs for LRCs over even-characteristic finite fields to date.
0 citationsRead paper