迁移学习支持向量回归机  

Transfer learning support vector regression

作　　者：史荧中[1,2] 王士同[1] 蒋亦樟[1] 刘培林[1,2] 

机构地区：[1]江南大学数字媒体学院,江苏无锡214122 [2]无锡职业技术学院物联网学院,江苏无锡214121

出　　处：《计算机应用》2013年第11期3084-3089,共6页journal of Computer Applications

基　　金：国家自然科学基金资助项目(61170122;61272210;61202311);江苏省自然科学基金资助项目(BK2011003;BK2012552)

摘　　要：传统的回归系统构建方法假设用于建模的数据是充分的,但若当前场景中重要数据信息缺失,则基于此数据集训练所得系统泛化能力较差。针对此缺陷,以支持向量回归机(SVR)为基础,提出了具有迁移学习能力的回归机系统,即迁移学习支持向量回归机(T-SVR)。T-SVR不仅能充分利用当前场景的数据信息,而且能有效地利用历史知识来学习,具有通过迁移历史场景知识来弥补当前场景信息缺失的能力。具体地,通过控制目标函数中当前模型与历史模型的相似性,使当前模型能在信息缺失和不足时从历史场景中得到有益信息,得到增强的当前场景模型。在模拟数据和酒类光谱数据集上的实验研究亦验证了在信息缺失场景下T-SVR较之于传统回归系统建模方法的更好适应性。The classical regression systems modeling methods suppose that the training data are sufficient, but partial information missing may weaken the generalization abilities of the regression systems constructed based on this dataset. In order to solve this problem, a regression system with the transfer learning abilities, i.e. Transfer learning Support Vector Regression （T-SVR for brevity） was proposed based on support vector regression. T-SVR could use the current data information sufficiently, and learn from the existing useful historical knowledge effectively, so that remedy the information lack in the current scene. Reinforced current model was obtained through controlling the similarity between current model and history model in the object function and current model can benefit from history scene when information is missing or insufficient. The experiments on simulation data and real data show that T-SVR has better adaptability than the traditional regression modeling method in the scene with information missing.

关 键 词：迁移学习 数据缺失 支持向量回归机 知识相关性 信息修补 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]