Published Paper Details:

ABSTRACT-Recently, Machine Learning has got a succeed scope on multi-view representation. Since the performance of data mining methods is heavily dependent on the expressive power of data representation, multi-view representation learning has become a very promising topic with wide applicability. It is an emerging direction in data mining which considers learning with multiple views to improve the generalization performance. Multi-view learning is also known as data fusion or data integration from multiple feature sets. In general, multi-view representation learning is capable of learning more informative and compact representation which leads to an improvement in predictors� performance. Thus, multi-view representation learning has been widely applied in numerous real-world applications including cross-media retrieval, natural language processing, video analysis, and recommender system. We propose two major categories for multi-view representation learning: (I) multi-view representation alignment, which aims to capture the relationships among multiple different views through feature alignment; (II) multi-view representation fusion, which seeks to fuse the separate features learned from multiple different views into a single compact representation. Both strategies seek to exploit the complementary knowledge contained in multiple views to comprehensively represent the data. In this project we implement the concept of canonical correlation analysis to retrieve the multi data. Inspired by the success of deep learning, deep multi-view representation learning has attracted much attention in cross media retrieval due to its ability of learning much more expressive cross-view representation.