【源码】用随机平均梯度最小化有限和
我们分析了随机平均梯度(SAG)方法,以优化有限数量的光滑凸函数之和。
We analyze the stochastic average gradient(SAG) method for optimizing the sum of a finite number of smooth convexfunctions.
与随机梯度(SG)方法一样,SAG方法的迭代成本与总和中的项数多少无关。
Like stochastic gradient (SG) methods, theSAG method’s iteration cost is independent of the number of terms in the sum.
然而,通过引入以前存储的梯度值,SAG方法比黑盒SG方法实现了更快的收敛速度。
However, by incorporating a memory ofprevious gradient values the SAG method achieves a faster convergence rate thanblack-box SG methods.
进一步地,就梯度评价的数量而言,在很多情况下新方法的收敛速度也比黑盒的确定性梯度法更快。
Further, in many cases the convergence rateof the new method is also faster than black-box deterministic gradient methods,in terms of the number of gradient evaluations.
与本文相关的一个网站供参考:
https://www.cs.ubc.ca/~schmidtm/Software/SAG.html
下载英文原文及相关源码:
http://page2.dfpan.com/fs/dlc4j2d21529a16b7a1/
更多精彩文章请关注微信号: