In this paper, we investigate the problem of optimization multivariateperformance measures, and propose a novel algorithm for it. Different fromtraditional machine learning methods which optimize simple loss functions tolearn prediction function, the problem studied in this paper is how to learneffective hyper-predictor for a tuple of data points, so that a complex lossfunction corresponding to a multivariate performance measure can be minimized.We propose to present the tuple of data points to a tuple of sparse codes via adictionary, and then apply a linear function to compare a sparse code against agive candidate class label. To learn the dictionary, sparse codes, andparameter of the linear function, we propose a joint optimization problem. Inthis problem, the both the reconstruction error and sparsity of sparse code,and the upper bound of the complex loss function are minimized. Moreover, theupper bound of the loss function is approximated by the sparse codes and thelinear function parameter. To optimize this problem, we develop an iterativealgorithm based on descent gradient methods to learn the sparse codes andhyper-predictor parameter alternately. Experiment results on some benchmarkdata sets show the advantage of the proposed methods over otherstate-of-the-art algorithms.
展开▼
