Congratulations to Ph.D. Student Kehua Yuan’s Paper “Consistency-Enhanced Multi-Granularity Knowledge Modeling for Semi-Supervised Feature Selection” Accepted by TKDE
Recently, the paper "Consistency-Enhanced Multi-Granularity Knowledge Modeling for Semi-Supervised Feature Selection" by Ph.D. student Kehua Yuan, under the guidance of Prof. Duoqian Miao, has been accepted by IEEE Transactions on Knowledge and Data Engineering, a leading journal in artificial intelligence. The study addresses feature selection in high-dimensional weakly supervised data by proposing a consistency-enhanced semi-supervised multi-granularity knowledge modeling framework. Extensive experiments demonstrate its superior classification performance and robustness compared with existing methods.
近日，实验室苗夺谦教授指导的2022级博士研究生苑克花的论文"Consistency-Enhanced Multi-Granularity Knowledge Modeling for Semi-Supervised Feature Selection"被国际顶级期刊《IEEE Transactions on Knowledge and Data Engineering》（TKDE）正式录用。TKDE是人工智能与模式识别领域公认的顶级期刊，长期位列中科院一区Top期刊，同时为CCF-A类推荐期刊，在国际学术界具有重要影响力。
针对高维弱监督数据场景下特征选择所面临的标注稀缺、不确定性与鲁棒性等问题，该研究提出了一致性增强的半监督多粒度不确定性知识建模框架。该方法通过结合数据样本分布开展一致性样本学习，充分挖掘样本之间的潜在一致性信息，提升知识建模过程中的样本信息质量；进而，构建多粒度模糊近似模型，实现不同粒度层次下的知识表达，以此建立半监督多粒度不确定性度量机制，有效刻画数据的粒度层级中的不确定性变化；最后，依据信息增益原则设计高效半监督特征筛选机制，提升模型对噪声与不一致信息的稳健性。理论分析与大规模实验结果均验证了本文所提方法在稳健性与分类性能方面的优势。
图1. 半监督多粒度知识空间构建图