摘要
提出了一种新的半监督两个视角的多示例聚类模型,整合文本视角和图像视角解决了伴有少量标签的多示例图像聚类问题。提出的模型首先嵌入概念分解和多示例核成为一个整体,学习每个视角的关联矩阵和两个视角所共享的聚类指示矩阵。而后,应用l_(2,1)范数学习最优的关联矩阵和聚类指示矩阵。进一步地,为了增加包之间的判别力,提出的模型强迫相同标签包的聚类指示向量间的相似性趋于1,不同标签包的指示向量间的相似性趋于0。最后,给出一种迭代更新算法优化提出的模型。实验结果表明,提出的模型优于现有的多示例聚类模型。
A novel semi-supervised two-view multi-instance clustering model is proposed, which bands text-view with image-view and solves the multi-instance image clustering problem with a small amount of label. Firstly, the proposed model embeds Concept Factorization and multi-instance kernel into a joint framework, which learns the association matrix of each view and the cluster indicator matrix shared by both views. Then, a l_(2,1)-norm is applied to learn the optimal association matrix and cluster indicator matrix. Furthermore, to enhance the discriminability between bags, the proposed model enforces the similarity of the cluster indicators for the bag with the same label to approximate 1 and the similarity with different labels to 0. Finally, an iterative updating algorithm is derived to solve the proposed model. The experimental results show that the proposed model is superior to other multiinstance clustering models.
作者
蔡昊
刘波
Cai Hao;Liu Bo(School of Automation,Guangdong University of Technology,Guangzhou 510006,China)
出处
《广东工业大学学报》
CAS
2021年第3期22-28,47,共8页
Journal of Guangdong University of Technology
基金
国家自然科学基金资助项目(61876044)。
关键词
多示例学习
多视角学习
概念分解
多示例核函数
multi-instance learning
multi-view learning
concept factorization
multi-instance kernel function
