arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2003.06926cs.LGmath.OCphysics.data-anstat.ML
收藏

具有随机学习率的随机梯度下降算法

Stochastic gradient descent with random learning rate

Universidade de Santiago de Compostela(圣地亚哥-德孔波斯特拉大学);Inovalabs Digital S.L. (TECHEYE)(Inovalabs Digital有限公司(TECHEYE))

Daniele Musso

摘要:

我们建议优化具有均匀分布的随机学习率的神经网络。相关的随机梯度下降算法可以近似为连续的随机方程,并在福克-普朗克形式主义分析。在小学习率区域中,训练过程的特征在于取决于平均学习率、最小批量大小和优化算法的动量的有效温度。通过比较随机学习率协议与循环和常数协议,我们建议随机选择一般是最好的策略,在小学习率制度,产生更好的正则化,而无需额外的计算成本。我们通过对MNIST和CIFAR 10数据集上的图像分类的浅、全连接和深、卷积神经网络的实验提供了支持证据。

英文摘要:

We propose to optimize neural networks with a uniformly-distributed random learning rate. The associated stochastic gradient descent algorithm can be approximated by continuous stochastic equations and analyzed within the Fokker-Planck formalism. In the small learning rate regime, the training process is characterized by an effective temperature which depends on the average learning rate, the mini-batch size and the momentum of the optimization algorithm. By comparing the random learning rate protocol with cyclic and constant protocols, we suggest that the random choice is generically the best strategy in the small learning rate regime, yielding better regularization without extra computational cost. We provide supporting evidence through experiments on both shallow, fully-connected and deep, convolutional neural networks for image classification on the MNIST and CIFAR10 datasets.

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。

↑