组中选优机器学习问题建模和算法研究

项目名称： 组中选优机器学习问题建模和算法研究

项目编号： No.61271337

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 无线电电子学、电信技术

项目作者： 罗林开

作者单位： 厦门大学

项目金额： 72万元

中文摘要： 组中选优是机器学习尚待研究具有挑战性的新问题，具有常规机器学习所没有的新特点。本项目针对组间同类样本之间的比较带来负面影响的问题，研究降低其影响的数据预处理方法，以及在建模中对组间同类样本不进行比较的嵌入方法；在此基础上，设计体现分组特点的泛化性能定量指标，建立具有强泛化性能和组内非线性可分处理能力的组中选优机器学习新模型；并针对最优样本与非最优样本数量极度不平衡问题，研究不使用类权重的处理方法；分析新模型的性质，给出大规模问题时的高效算法；结合工艺参数寻优和投资优化等领域的组中选优问题，开展新模型、新算法的应用研究。组中选优作为一个新的基础性机器学习问题，本项目的研究可以丰富现有机器学习的模型和算法，拓广现有机器学习的应用范围，既有很强的创新性，亦有重要的应用价值。

中文关键词： 组中选优；支持向量机；最小序贯算法；；

英文摘要： Learning the rule of selecting the best one from group data (SBG) is a new machine learning problem. It brings some challenges for the existing machine learning models because of the new characteristics. To eliminate the negative impact on the comparison of the samples with same type between groups, this project first investigates the data preprocessing method,as well as the embedding methods in modeling without taking the comparison. Then, the quantitative measure of generalization performance for SBG is proposed. And some new models for the SBG learning problem are developed, in which a strong generalization performance and a good suitability for nonlinear separable problem within-group are guaranteed. Thirdly,to overcome the extremely unbalanced problem between the sizes of two classes, some methods without utilizing the weights of classes are investigated. Fourthly, the efficient algorithm for the new models with large scale data is presented after investigating the nature of the models. Finally, two applications on the optimizations of process parameters and investment are provided. SBG is a new foundation machine learning problem, this project will extend the models, algorithms and application ranges of the existing machine learning techinique,which is innovation in theory and has great application value.

英文关键词： Selecting the best in each group；SVM；SMO；；