演講公告

演講主題：Online Nonlinear Support Vector Machine for Large-Scale Classification
主講人：李育杰特聘教授(台灣科技大學資訊工程學系)
演講日期：2015年10月06日(星期二) 下午2:00 –3:00
演講地點：(光復校區) 科學一館223室
茶會時間：當天下午1:30 (科學一館205室)
摘要內容：
Online learning is an important technique for handling large-scale problems. In general, most of real-world classification problems are not linearly separable but most online learning algorithms give a linear model. Support vector machine (SVM) is one of the most popular nonlinear learning methods by taking the advantage of the kernel trick. Unfortunately, the computational overhead prohibits it for dealing with large scale problems. We propose an online nonlinear SVM algorithm with the reduced kernel trick. Similar to other online learning algorithms such as the passive and aggressive algorithm, we also have a closed form updating rule. Thus, it will be extremely fast for each updating. Moreover, we introduce a proximal model that “memorizes” the statistical information of the instances shown in the learning process. Combining the nonlinear SVM model and the proximal model, our proposed method is insensitive to the input order and is able to quickly achieve a reasonable good solution in a single pass.
相關檔案：演講1041006.doc

$返回$ go back