| 基于预测风险最小化的模型选择理论与方法研究 |
| |
| 引用本文： | 盛守照,王道波.基于预测风险最小化的模型选择理论与方法研究[J].系统工程,2004,22(4):100-103. |
| |
| 作者姓名： | 盛守照 王道波 |
| |
| 作者单位： | 南京航空航天大学,自动化学院,江苏,南京,210016 |
| |
| 基金项目： | 航空基金资助项目(01C52015) |
| |
| 摘 要： | 阐述基于预测风险最小化的模型选择问题，提出基于经验风险最小化原则的模型选择一致收敛性定理，解决有限样本下利用经验风险来最小化预测风险的问题，并分析模型选择欠学习或过学习问题的根源，给出一种模型选择的次优迭代算法。最后通过具体实例验证上述理论和方法的可行性和优越性。
|
| 关 键 词： | 模型选择理论 预测风险 最小化 统计学习理论 机器学习 |
| 文章编号： | 1001-4098(2004)04-0100-04 |
On Theory and Method of Model Selection Based on Predict Risk Minimization |
| |
| SHENG Shou-zhao,WANG Dao-bo.On Theory and Method of Model Selection Based on Predict Risk Minimization[J].Systems Engineering,2004,22(4):100-103. |
| |
| Authors: | SHENG Shou-zhao WANG Dao-bo |
| |
| Abstract: | In this paper, the problem of model selection based on predict risk minimization is mainly discussed; the theorem of consistency and convergence of model selection based on empirical risk minimization is proposed, and the problem of (predict) risk minimization with finite samples by means of empirical risk is resolved; the cause of underfitting and overfitting of model selection is analyzed in detail; then a new suboptimal algorithm for model selection is given. Their reliability and (advantage) are illustrated through concrete test. |
| |
| Keywords: | Model Selection Statistical Learning Theory Machine Learning |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
