Sparsity Optimization Study In Multi-task Feature Learning And Disease Classification

The applications of sparse optimization method in many areas including computer vision, picture processing, biomedical informatics and so on arouse the great interest from the scholars from all over the world. According to the structure and the sparse form of data on the actual application background,choosing suitable sparse optimization models and designing the related algorithms can fast and efficiently solve the optimization problem. In this paper, we consider two aspects of the applications of the sparse optimization methods. Firstly, we consider multi-task feature learning, the related algorithm and the applications in letter classification and school effectiveness. Secondly we consider the sparse optimization model in the aspect of disease classification, designing the related algorithm,and then we combine the related algorithm with independence rule. At the same time, we apply the related algorithm and independence rule to disease classification.In the aspect of multi-task feature learning, we improve the multi-task feature learning via l_2,1-norml minimization and come up with the multi-task feature learning via l_2,p-norm minimization. Next, we consider the application of the related algorithm in Letter classification and School effectiveness. Matrix norm have joint sparsity, so we consider the matrix norm in designing the model. However because l_2,p-norml is non-convex and non-lipschitz continuous, solving this optimization problem is very challenging. This paper propose an iterative method to solve uniformly the difficult problem in p ?(0,1] and the algorithm convergence is also ensured. In particular,p?(0,1] meet the diversity of multi-task sparse optimization type, In the meantime, experimental results show our algorithm is effective.In the aspect of disease classification, based on classical joint sparse optimization and the independence rule of two sample, we propose the joint sparse independence classification rule. At the same time, we do some data experiment in three public dataset and the result indicate that the joint sparse independence rule combine effectively the “group” of the joint sparsity feature selection with the independence property of classification method and make classification accuracy higher and used time less.