Homology Structure Learning Algorithm And Its Application

Structure learning is a major method in the field of machine learning, and the current study based on the structure learning method is still unusual. In this paper, we mainly study the homology questions of structure feature learning through using the homology learning theory and propose the frame of homology structure learning which includes:1) Introducing the basic theories of homology, co-homology etc., which laid the theoretical foundation for this paper;2) Giving a homology structure learning algorithm(HSLA) which finds the solution to the data partitioning problem of variable margin from the perspective of homology.HSLA can effectively preserve the images structures in the process of classification;3) Giving a neighborhood homology learning algorithm(NHLA) which can characterize the images more exactly. NHLA not only can characterize the images accurately which margins are variable, but also can effectively preserve the images structures in the process of classification;4) Giving a co-homology learning algorithm(CHLA) which not only has simple calculation but also can characterize the images accurately. CHLA describes images based on both the topological and geometric features. Therefore, the features extracted by CHLA can characterize the shapes more exactly and can be used better for image recognition;5) Application: Experimenting on handwritten digit database and object image to verify the effectiveness of the algorithms.