中文摘要：

对称IB（Symmetric Information Bottleneck）通过行、列压缩变量之间的相互协作来挖掘数据中的双向压缩模式.由于行、列压缩变量不能完全承载行、列基层变量中所蕴含的特征信息,从而导致对称IB所得的数据双向压缩模式与基层变量所蕴含的内在模式之间存在一定的偏离.针对该问题,通过最大化地保存压缩变量与基层变量交叉之间的互信息,将基层变量引入到数据的双向压缩中,使它们协助压缩变量共同来学习联合分布中的双向压缩模式,提出交叉对称IB：ICSIB（Inter-Correlated Symmetric Information Bottleneck）.ICSIB算法采用交错的顺序＂抽取-合并＂迭代过程来优化压缩变量与基层变量交叉之间的互信息,可保证得到目标函数的一个局部优解.实验结果表明,在基层特征变量的协助下,ICSIB算法得到的数据双向压缩模式更接近于数据中真实的内在模式,并可有效地应用于数据的联合聚类中.

英文摘要：

The symmetric IB aims to extract the double compressing patterns of data via the cooperation between compressed row and column variables.However,the compressed variables cannot completely carry the information resided in original variables,which results that there will be some deviation between the compressing patterns extracted by symmetric IB and the original patterns resided in original features.To solve this problem,this paper proposes an Inter-Correlated Symmetric Information Bottleneck（ICSIB）,which aims to maximize the inter-correlated mutual information between compressed variables and original variables,so that original feature variables can be involved in data double compressing process and can be used to help the compressed variables to learn double compressing patterns.The ICSIB algorithm can monotonically increase the objective function by an intertwining ＂draw-and-merge＂sequential iteration procedure,and guarantee to converge to a local maximum of the information.Our experimental results on benchmark data sets have demonstrated the effectiveness of the proposed method in the application of data double compressing and co-clustering.