一种半监督协同训练的正则化算法

半监督学习是机器学习领域的研究热点。协同训练研究数据有多个特征集时的半监督学习问题。从正则化角度研究协同训练,利用假设空间的度量结构定义学习函数的光滑性和一致性,在每个视图内的学习过程中以函数光滑性为约束条件,在多个视图的协同学习过程中以函数一致性为约束条件,创新性地提出一种两个层次的正则化算法,同时使用函数的光滑性和一致性进行正则化。实验表明,该算法较仅使用光滑性或仅使用一致性的正则化方法在预测性能上有显著提高。

New Regularization Method of Co-training

Semi-supervised learning is a hot research topic of machine learning. Co-training is a multi-view semi-superwised learning method. Co-training is studied from the regularization point of view. By exploiting the metric structure of the hypotheses space, the smoothness and consistency of a hypothesis were defined. A two levels regularization algorithm was presented which uses the smoothness to regularize the within-view learning process and uses the consistency to regularize the between-view learning process. The experimental results were presented on both synthetic and real world datascts.