arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1909.03069cs.LGstat.ML
收藏

Differential Equation Units:从数据中学习激活函数的函数形式

Differential Equation Units: Learning Functional Forms of Activation Functions from Data

Amazon Web Services(亚马逊云科技);University of Massachusetts Amherst(马萨诸塞大学阿默斯特分校);Microsoft Inc.(微软公司)

MohamadAli Torkamani, Shiv Shankar, Amirmohammad Rooshenas, Phillip Wallis

摘要:

大多数深度神经网络使用简单的固定激活函数,如S形或整流线性单元,而不管域或网络结构如何。我们引入微分方程单元(DEU),现代神经网络的改进,使每个神经元学习一个特定的非线性激活函数从一个家庭的解决方案,一个常微分方程。具体地,每个神经元可以在训练期间基于网络的其他部分的行为来改变其功能形式。我们表明,使用神经元与DEU激活函数的结果在一个更紧凑的网络能够实现可比的,如果不是优越的,性能相比,更大的网络。

英文摘要:

Most deep neural networks use simple, fixed activation functions, such as sigmoids or rectified linear units, regardless of domain or network structure. We introduce differential equation units (DEUs), an improvement to modern neural networks, which enables each neuron to learn a particular nonlinear activation function from a family of solutions to an ordinary differential equation. Specifically, each neuron may change its functional form during training based on the behavior of the other parts of the network. We show that using neurons with DEU activation functions results in a more compact network capable of achieving comparable, if not superior, performance when is compared to much larger networks.

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。

↑