基于相近原则的半指导直推学习机及其增量算法

Transductive Learning Machines Based on the Affinity-rule for Semi-supervised Problem and Its Incremental Algorithm

半指导问题是近来机器学习研究中的备受关注一个重要内容．本文以满足“在输入空间中相近的对象其输出也相近”这一源于直观事实的原则(相近原则)去解决半指导学习问题,给出在这个原则下的一个一般的直接推理方法—基于相近原则的半指导问题直推学习机,得到了这个问题的解析解及迭代算法,用模式分类实例验证该方法的有效性,并给出适于在线处理的增量学习算法,这些增量算法尤其还适于新增了有指导的信息的场合．

Semi-supervised problem, as one of important contents in machine learning, has gained much attention in recent years. In this paper, the so-called affinity-rule, which comes from the intuitive fact is that if two objects are close in input space then their outputs should also be close, is used to obtain a transductive inference for semi-supervised problem. To get the solution of semi-supervised learning problem satisfying the rule, transductive learning machines based on the affinity-rule are introduced as general method. The analytical solution for the problem and its iterated algorithm are obtained. Some examples about pattern classification are given to explain the validity of this method. A incremental learning algorithm adapted to on-line data processing is introduced. In addition, this incremental learning algorithm can be used to new supervised samples.