稀疏正则化最小类散度半监督分类机

基于稀疏表示理论提出一种稀疏正则化最小类散度半监督分类机(SRMCV), 且对于模式分类问题, SRMCV通过引入稀疏Laplacian正则化项和类内散度信息以实现预测空间函数在全局稀疏表示结构下平滑变化, 同时通过类内数据散度结构进一步优化决策函数的判别方向, 此方法能解决现有SSL方法对模型参数敏感和在噪声学习环境下缺乏鲁棒性等问题, 其有效性已在实际数据集上通过实验验证.

Sparse Regularization for Minimum Class Variance Semi-supervised Classifier

This paper presents a sparse regularization minimum class scatter semi-supervised classifier （SRMCV） based on the theory of sparse representation. For the problems of pattern classification, SRMCV aims at achieving smooth changes of the predicted spatial function in the structure of the global sparse representation by introducing sparse Laplacian regularization term and the information of inner class scatter. In addition, through the scatter structure of inner class data, it is intended to further optimize the discriminant direction of the decision-making function. This method can solve some problems existing in the SSL method such as sensitivity to the model parameters, the lack of robustness in the noisy learning environment, etc. On the real data set, the experimental results manifest the effectiveness of the proposed method.