一种结合类别权重及多示例的多标记学习改进算法

发布时间：2018-07-10 05:01

本文选题：多标记学习 + MIML　；参考：《小型微型计算机系统》2017年04期

【摘要】：多标记学习主要用于解决单个样本同时属于多个类别的问题.传统的多标记算法在输入空间仅用单一示例表示多义性对象,过度简化了对象的复杂内涵,导致在表示阶段丢失重要信息.针对这一问题,提出一种结合类别权重及多示例的多标记学习改进算法CWMI-INSDIF.算法采用MIML(Multi-Instance Multi-Label learning)框架,在表示阶段,将学习样本分化为多示例包形式,在生成示例包过程中定义一组描述数据重要度的权重函数,并加入自适应惩罚策略,最终确定了学习样本中各部分信息的权重大小,从而在输入空间更好的描述了样本歧义性.算法给出了在公开数据集的实验结果.通过仿真分析,验证了本文提出的算法在学习性能和分类效果方面的提高.
[Abstract]:Multi-label learning is mainly used to solve the problem that a single sample belongs to multiple categories at the same time. The traditional multi-label algorithm only uses a single example to represent the polysemy object in the input space, which oversimplifies the complex connotation of the object and results in the loss of important information in the presentation phase. In order to solve this problem, an improved multi-label learning algorithm, CWMI-INSDIFF, is proposed, which combines class weight with multiple examples. The algorithm uses MIML (Multi-Instance Multi-Label learning) framework. In the presentation stage, the learning samples are divided into multi-sample packets, and a set of weight functions describing the importance of data are defined in the process of generating the sample packets, and adaptive penalty strategies are added. Finally, the weight of each part of the learning sample is determined, and the ambiguity of the sample is better described in the input space. The algorithm gives the experimental results in the open data set. Simulation results show that the proposed algorithm improves the learning performance and classification performance.
【作者单位】：南京工业大学计算机科学与技术学院;
【基金】：江苏省高校自然科学基金项目(12KJB510007)资助
【分类号】：TP181

【相似文献】