1: \begin{abstract}

2: We firstly propose the new stochastic gradient estimate of unbiasedness and minimized variance in this paper. Secondly, we propose the two algorithms: Algorithm1 and Algorithm2 which apply the new stochastic gradient estimate to modern stochastic conjugate gradient algorithms SCGA \cite{kou2022mini} and CGVR \cite{jin2018stochastic}. Then we prove that the proposed algorithms can obtain linear convergence rate under assumptions of strong convexity and smoothness. Finally, numerical experiments show that the new stochastic gradient estimate can reduce variance of stochastic gradient effectively. And our algorithms compared with SCGA and CGVR can convergent faster in numerical experiments on ridge regression model.

3: \keywords{Stochastic conjugate gradient \and Variance reduction \and Linear convergence}    

4: % \PACS{PACS code1 \and PACS code2 \and more}

5: %\subclass{MSC code1 \and MSC code2 \and more}

6: \end{abstract}