RL-赵-(六)：随机逼近与随机梯度下降02-2：Robbins-Monro（RM）算法【wₖ₊₁=wₖ-αₖĝ(wₖ,ηₖ)】【SA领域的开创性工作】

2、Robbins-Monro（RM）算法Robbins-Monro(RM)算法：wk+1 = wk − ak g~(wk,ηk ),k = 1,2,3,... \begin{aligned}w_{k+1}\:=\:w_k\:-\:a_k\:\tilde{g}(w_k,\eta_k\:),k\:=\:1,2,3,...\end{aligned}wk+1​=wk​−ak​g~​(wk​,ηk​),k=1,2,3,...​其中wkw_kwk​是root的第kkk次估计；g~(wk,ηk)=g(wk)+ηk\tilde{g}\left(w_k,\eta_k\right)=g(w_k)+\eta_kg~​(wk​,ηk​)=g(wk​)+ηk​是第kkk次带有噪声的观测；