Abstract: In the present study a reinforcement support vector machine is developed for multiclass classification.Support vector machine and reinforcement learning are combined.Support vector machine classifier is trained on labeled data set.The unlabeled instance is queried according to querying strategy，and according to the critic of right or wrong about the classification result of the queried instances，the classifier is updated.The method is evaluated on a synthetic data set，as well as on real data sets，and it is shown to be valuable for the problem of multiclass classification in which the labeled instances are difficult to obtain.

Key words: multiclass classification, support vector machine, reinforcement learning

摘要： 提出了一种强化支持向量机方法，将支持向量机与强化学习结合，逐步对未知类别标记样本进行访问，根据对该样本分类结果正确与否的评价标记访问点的类别，并对当前的分类器进行更新，给出了更新分类器的规则。对模拟数据和真实数据分别进行了实验，表明该方法在保证分类精度的同时，大大降低了对已知类别标记的训练样本的数量要求，是处理已知类别标记样本获取困难的多类分类问题的一种有效的方法。

关键词: 多类分类, 支持向量机, 强化学习