Creating Virtual Examples based on Semi-empirical Formula for Adaptive Learning
Plenty of applications have to take into account dynamic factors. However traditional machine learning paradigm lacks adaptivity to deal with sudden change of the unknown underlying target function because its common assumption: training and test samples are drawn from the same feather space (input space) and the same distribution (target function). The difficulty is that it is short of new training data when change happens. A novel method, which creates virtual examples based on semi-empirical formula, is proposed to make up the loss of new samples when the target function alters suddenly. This method is tested and verified to perform fairly well on the robotic belt grinding platform, which suffers for dynamic factors when modeling the material removal rate. The experiments also prove that virtual examples as a kind of incorporation of prior knowledge are able to mind out the information contained in the semi-empirical formula for the learner.
Adaptive Learning Semi-empirical Formula Virtual Example Robotic Belt Grinding
Hongbo Lv Yixu Song Peifa Jia Lizhe Qi
Department of Computer Science and Technology Tsinghua University Beijing, China InterSmart Robotic Systems Co., Ltd. Langfang, HeBei, China
国际会议
2010 International Conference on Future Information Technology(2010年未来信息技术国际会议 ICFIT 2010)
长沙
英文
760-764
2010-12-14（万方平台首次上网日期，不代表论文的发表时间）