大样本领域自适应支撑向量回归机

针对回归问题中存在采集数据不完整而导致预测性能降低的情况,根据支撑向量回归机(support vectorregression,简称SVR)等价于中心约束最小包含球(center-constrained minimum enclosing ball,简称CC-MEB)以及相似领域概率分布差异只与两域各自的最小包含球中心点位置有关的理论新结果,提出了针对大数据集的领域自适应核心集支撑向量回归机(adaptive-core vector regression,简称A-CVR).该算法利用源域CC-MEB 中心点对目标域CC-MEB 中心点进行校正,从而提高目标域的回归预测性能.实验结果表明,这种领域自适应算法可以弥补目标域缺失数据的不足,大大提高回归预测性能.

Support Vector Regression for Large Domain Adaptation

Incomplete data collection in regression analysis would lead to low prediction performance, which aises the issue of domain adaptation. It is well known that support vector regression (SVR) is equivalent to center-constrained minimum enclosing ball (CC-MEB). Also in solving the problem of how to effectively transfer the knowledge between the two fields, new theorems reveal that the difference between two probability distributions from two similar domains only depends on the centers of the two domains' minimum enclosing balls. Based on these developments, a fast adaptive-core vector regression (A-CVR) algorithm is proposed for large domain adaptation. The proposed algorithm uses the center of the source domain's CC-MEB to calibrate the center of the target domain's in order to improve the regression performance of the target domain. Experimental results show that the proposed domain adaptive algorithm can make up for the lack of data and greatly improve the performance of the target domain regression.