基于RSDE的领域自适应概率密度估计方法  

Density estimation based on RSDE for domain adaption

作　　者：许敏[1,2] 王士同[1] 顾鑫[1,3] 俞林[2] 

机构地区：[1]江南大学数字媒体学院,江苏无锡214122 [2]无锡职业技术学院电子与信息技术学院,江苏无锡214121 [3]无锡北方湖光光电有限公司研发部,江苏无锡214035

出　　处：《控制与决策》2013年第1期125-130,共6页Control and Decision

基　　金：国家自然科学基金项目(60903100;60975027;61170122);江苏省研究生创新工程项目(CXZZ12-0759)

摘　　要：同一应用领域不同时间、地点或设备,采集的样本数据可能存在扰动、噪音或缺失,如何对样本数据集进行有效的预处理是其进一步应用的前提.针对上述问题,提出一种新的基于压缩集密度估计(RSDE)算法的领域自适应概率密度估计方法A-RSDE,通过学习源域(训练域)知识,使目标域(测试域)概率密度估计更接近真实概率密度分布,并用基于近似最小包含球的核心集快速算法求解A-RSDE,将其应用于大数据集密度估计.Benchmark和UCI数据集上的实验表明,该算法具有较好的性能.Sample datasets are often collected from different times, places or devices. Due to the existence of the disturbance, noise and missing data, the collected datasets can not always keep the same distribution, and can even sometimes be required to concentrate them to reduce the computational burden, which can do the domain adaptation as the preprocessing step for the sample dataset before being fed into the next step. In order to achieve the above goal, a novel adaptive reduced set density estimator（A-RSDE） is proposed for adaptive probability density estimation by making full use of the source domain＇s （training dataset） knowledge of the probability density distribution, which lets the target domaln＇s （testing dataset） probability density estimation be closer to the true probability density distribution. Meanwhile, the fast core-sets based minimum enclosing ball（MEB） approximation algorithm is introduced to develop the proposed algorithm. Finally, the experiment on the benchmark data sets and UCI data sets show that the proposed method has better performance.

关 键 词：领域自适应 压缩集密度估计 最小包含球 核心集 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]