利用少量标签训练集的半监督OLTV算法的设计

目前,大多数半监督学习算法都要使用大量的带标签训练样例集,当只有一个带标签训练样例时许多方法便不再适用。因此,基于两个充分视图工作的OLTV算法被提出来了。本文实现了此算法,并对充分视图的构建进行了讨论。对于特定的数据集,首先对它进行属性划分,得到两个充分视图;然后再执行OLTV算法,这样由一个初始标签训练样例便可以得到两个新的标签样例。最后在公共数据集上的实验验证了所提出算法的有效性。

Design of OLTV Algorithm on Semi-supervised Learning with Very Few Labeled Training Examples

Most of the current semi-supervised learning algorithms require a number of labeled training examples to be available.In particular,such methods cannot work well when there is only one labeled training example.Therefore,the OLTV Algorithm working under a two-view setting is proposed.In this paper,the algorithm on semi-supervised learning with very few labeled training examples is implemented based on OLTV Algorithm,and the construction of sufficient views is discussed.For the given dataset,we can split its attributes to two sufficient views,and carry out the OLTV Algorithm.Then additional labeled examples can be required from the original labeled example.The experiment’s results in some datasets prove the validity of this algorithm.