arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2004.13146math.OCcs.LG
收藏

随机梯度下降中小批量对产量方差的影响

The Impact of the Mini-batch Size on the Variance of Gradients in Stochastic Gradient Descent

Northwestern University(西北大学)

Xin Qian, Diego Klabjan

摘要:

小批量随机梯度下降(SGD)算法广泛用于训练机器学习模型,特别是深度学习模型。我们研究了线性回归和两层线性网络下的SGD动态,通过关注梯度的方差,可以轻松扩展到更深的线性网络,这是第一次研究这种性质。在线性回归的情况下,我们表明,在每次迭代的梯度范数是一个减函数的小批量大小$b$,因此方差的随机梯度估计是一个减函数的$b$。对于具有$L_2$损失的深度神经网络,我们证明梯度的方差是$1/b$中的多项式。结果支持了一个重要的直觉,即较小的批量产生较低的损失函数值,这是研究人员普遍认为的。证明技术展示了随机梯度估计和初始权重之间的关系,这是有用的进一步研究SGD的动力学。我们根据经验为我们在各种数据集和常用的深度网络结构上的结果提供了进一步的见解。

英文摘要:

The mini-batch stochastic gradient descent (SGD) algorithm is widely used in training machine learning models, in particular deep learning models. We study SGD dynamics under linear regression and two-layer linear networks, with an easy extension to deeper linear networks, by focusing on the variance of the gradients, which is the first study of this nature. In the linear regression case, we show that in each iteration the norm of the gradient is a decreasing function of the mini-batch size $b$ and thus the variance of the stochastic gradient estimator is a decreasing function of $b$. For deep neural networks with $L_2$ loss we show that the variance of the gradient is a polynomial in $1/b$. The results back the important intuition that smaller batch sizes yield lower loss function values which is a common believe among the researchers. The proof techniques exhibit a relationship between stochastic gradient estimators and initial weights, which is useful for further research on the dynamics of SGD. We empirically provide further insights to our results on various datasets and commonly used deep network structures.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑