过度泛化误差的稳健估计

Robust estimation of overgeneralization error

在最优收敛速度的驱动下,研究风险最小化的误差估计。通过运用相关不等式及性质,证得了在统计学习理论框架下的比较定理,不仅阐明了过度泛化误差与预测误差之间的关系,而且将矩条件弱化到1+α阶。

Motivated by the optimal convergence rate,this paper studies the error estimation of risk minimization.By using related inequalities and properties,the comparison theorem under the framework of statistical learning theory is proved,which not only explains the relationship between the excess generalization error and the prediction error,but also weakens the moment condition to the order of 1+α.