演講公告 91年6月7日 (星期五)

摘要

Support vector machine (SVM) is a rapidly growing method in machine learning field. It gained its popularity due to its ease of computation and success in many real applications. So far most theoretical results in computational literatures provide generalization error bound in terms of margins, number of support vectors or empirical misclassification error. However, in statistics we emphasize more on Bayes optimal risk. In this talk, I'll present some results in computational literatures and the relation of SVM to Bayes optimal rule. Our recent result shows SVM can be improved but with higher computation complexity.