基于马氏采样的最小二乘正则化回归的最优速率

摘要

建立基于一致遍历的马氏采样的推广界，利用最小二乘正则化回归算法给出基于一致遍历马尔可夫链样本的误差分析，并得到最优的学习速率O（m－1）。%We establish the bound on the learning rates of regularization regression based on uniformly ergodic Markov chain samples. We give an error analysis and obtain the optimal learning rate O( m-1 ) for the least squares regularized regression algorithm based on uniformly ergodic Markov chain samples.