Data fusion and matching by maximizing statistical dependencies

摘要: Multi-view learning is a task of from multiple data sources where each source represents different view the same phenomenon. Typical examples include multimodal information retrieval and classification genes by combining heterogeneous genomic data. methods can be motivated two interrelated lines thoughts: if single not sufficient for task, other views complement information. Secondly, searching an agreement between may generalize better than view. In this thesis, novel unsupervised multi-view are proposed. methods, in general, work views. However, defining straightforward task. statistical dependency used to define Assuming that shared more interesting, find Based on principle, fast linear preprocessing method performs fusion during exploratory analysis introduced. Also, evaluation approach based compare vector representations bilingual corpora general assume co-occurred samples