| 摘要 | 第1-5页 |
| Abstract | 第5-9页 |
| 1. Introduction and Background | 第9-19页 |
| ·Empirical Risk Minimization Principle | 第9-13页 |
| ·Regularization in Learning Theory | 第13-16页 |
| ·Least Square Regularization | 第16-17页 |
| ·Outline of the Thesis | 第17-19页 |
| 2. Empirical Risk Minimization for Regression with Unbounded Sampling | 第19-41页 |
| ·Settings for Regression with Unbounded Sampling | 第19-22页 |
| ·Bennet Inequality | 第22-24页 |
| ·Main Result on the ERM Algorithm | 第24-27页 |
| ·Error Decomposition for the ERM Algorithm | 第27-29页 |
| ·Error Term Involving f_z in the ERM Setting | 第29-33页 |
| ·Estimating the Sample Error and Total Error | 第33-36页 |
| ·ERM for Classification | 第36-41页 |
| 3. Regularized Least Square Regression with Unbounded Sampling | 第41-62页 |
| ·Setting and Learning Rates for Regularization | 第41-43页 |
| ·Error Decomposition in Regularization | 第43-44页 |
| ·Sample Error Estimation in Regularization | 第44-54页 |
| ·Sample error involving the regularizing function | 第44-47页 |
| ·Sample error term involving f_(z,λ) | 第47-52页 |
| ·Total sample error and weak estimation of the learning rate | 第52-54页 |
| ·Iteration and Strong Learning Rate | 第54-59页 |
| ·Comparison with Previous Results | 第59-62页 |
| 4. Extensions and Further Study | 第62-66页 |
| ·Error Analysis with Integral Operators | 第62-63页 |
| ·Online Learning | 第63-64页 |
| ·Learning with Non-i.i.d.Sampling | 第64-65页 |
| ·Semi-Supervised Learning | 第65-66页 |
| BIBLIOGRAPHY | 第66-73页 |
| Publieations | 第73-74页 |
| Acknwoledgement | 第74-75页 |
