论文部分内容阅读
在统计推断中,稳健性是指实际问题的数据来源与我们的模型假定有偏离时,所采用算法的结果受到的扰动很小,并且保持算法的预测性能.本文将统计稳健性的研究方法引入机器学习中,分析得到近邻估计这种局部学习能够在大样本的情形下收敛到 Bayes 最优估计,同时收敛条件可说明近邻估计是稳健估计.在模拟数据和真实数据库上进行实验,结果表明在某些离群点影响模型的情况下,仍保持监督学习预测的泛化性能.
In statistical inference, robustness means that when the source of the actual problem deviates from our model assumptions, the results of the proposed algorithm suffer little disturbance and maintain the predictive performance of the algorithm.In this paper, the method of statistical robustness is introduced In machine learning, we find that this local learning can converge to the Bayesian optimal estimator in the case of large samples, meanwhile, the convergence condition can indicate that the neighborhood estimation is a robust estimator.Experimental results on simulated and real databases show that in the Some outliers affect the model of the case, still maintain the supervisory performance of learning prediction.