Multiview learning problem refers to the problem of learning a classifierfrom multiple view data. In this data set, each data points is presented bymultiple different views. In this paper, we propose a novel method for thisproblem. This method is based on two assumptions. The first assumption is thateach data point has an intact feature vector, and each view is obtained by alinear transformation from the intact vector. The second assumption is that theintact vectors are discriminative, and in the intact space, we have a linearclassifier to separate the positive class from the negative class. We define anintact vector for each data point, and a view-conditional transformation matrixfor each view, and propose to reconstruct the multiple view feature vectors bythe product of the corresponding intact vectors and transformation matrices.Moreover, we also propose a linear classifier in the intact space, and learn itjointly with the intact vectors. The learning problem is modeled by aminimization problem, and the objective function is composed of a Cauchy errorestimator-based view-conditional reconstruction term over all data points andviews, and a classification error term measured by hinge loss over all theintact vectors of all the data points. Some regularization terms are alsoimposed to different variables in the objective function. The minimizationproblem is solve by an iterative algorithm using alternate optimizationstrategy and gradient descent algorithm. The proposed algorithm shows itadvantage in the compression to other multiview learning algorithms onbenchmark data sets.
展开▼
