S3VM和TSVM的不同

S3VM和TSVM的不同

转载：http://ycool.com/post/6vc4vf7

自己总结的一点学习心得，希望了解的人给予补充。

TSVM都是比较早的，后来的一般都叫S3VM。提出的人不一样。
Vapnik, 1998; Joachims, 1999) under the name Transductive Support Vector Machine (TSVM), and in (Bennett & Demiriz, 1998; Fung & Mangasarian,2001) under the name Semi-Supervised Support Vector Machine (S3VM).

A major line of research on extending SVMs to handle partially labeled datasets is based on the following idea: solve the standard SVM problem while treating the unknown labels as additional optimization variables. By maximizing the margin in the presence of unlabeled data, one learns a decision boundary that traverses through low data-density regions while respecting labels in the input space. 
In other words, this approach implements the cluster assumption for semi-supervised learning – that points in a data cluster have similar labels.
This idea was first introduced in [14] under the name Transductive SVM, but since it learns an inductive rule defined over the entire input space, we refer to this approach as Semi-supervised SVM (S3VM).

S3VMs were introduced as Transductive SVMs, originally designed for the task of directly estimating labels of unlabeled points. However S3VMs provide a decision boundary in the entire input space, and so they can provide labels of unseen test points as well. For this reason, we believe that S3VMs are inductive semi-supervised methods and not strictly transductive.

理论上可以证明，对于可分数据。TSVM、S3VM解必定是某种相近度量下相近直推学习机的解，并且它们都是通过控制VC维的上界以达到好的泛化性能的，只是VC维上界的控制函数的取法不同。